Upload 110 files

.dockerignore

@@ -0,0 +1,3 @@
+/.venv
+dist
+__pycache__/

.env-docker

@@ -0,0 +1,8 @@
+weight_root=/weights
+weight_uvr5_root=/assets/uvr5_weights
+index_root=/indices
+rmvpe_root=/assets/rmvpe
+hubert_path=/assets/hubert_base.pt
+save_uvr_path=/assets/uvr5_weights
+TEMP=/app/TEMP
+pretrained=/assets/pretrained

.gitignore

@@ -0,0 +1,8 @@
+/.venv
+/assets
+dist
+__pycache__/
+*.py[cod]
+*$py.class
+
+.DS_Store

Dockerfile

@@ -1,19 +1,47 @@
-FROM python:3.10-slim
+FROM alpine:3.19.1 as assets
 
-WORKDIR /app
+RUN apk add \
+      --update \
+      --no-cache \
+        bash \
+        git \
+        git-lfs\
+        dos2unix
+
+COPY --chmod=755 ./assets-download.sh /assets-download.sh
+
+#convert malformed line endings if cloned from Windows
+RUN dos2unix /assets-download.sh
+
+RUN /assets-download.sh 88e42f0cb3662ddc0dd263a4814206ce96d53214 assets
+
+FROM python:3.10.14-bullseye as app
 
-# System deps
-RUN apt-get update && apt-get install -y ffmpeg git && rm -rf /var/lib/apt/lists/*
+SHELL [ "/bin/bash", "-c" ]
+
+RUN apt update && \
+    apt install -y \
+      libsndfile1 \
+      libsndfile1-dev && \
+    apt clean && \
+    rm -rf /var/lib/apt/lists/*
+
+COPY --from=assets /assets /assets
+
+WORKDIR /app
 
-# Python deps
-COPY requirements.txt .
-RUN pip install --no-cache-dir -r requirements.txt
+COPY ./pyproject.toml .
 
-# Copy app + models
-COPY app/ ./app/
-COPY rvc/ ./rvc/
-COPY models/ ./models/
+RUN pip install \
+      --no-cache-dir \
+      "poetry==1.7.1" && \
+    poetry config virtualenvs.create false && \
+    poetry install \
+      --no-interaction \
+      --no-root && \
+    poetry cache clear --all .
 
-EXPOSE 7860
+COPY ./rvc ./rvc
+COPY ./.env-docker ./.env
 
-CMD ["uvicorn", "app.main:app", "--host", "0.0.0.0", "--port", "7860"]
+CMD [ "poetry", "run", "poe", "rvc-api" ]

api-request.sh

@@ -0,0 +1,59 @@
+#!/usr/bin/env bash
+#
+# Runs request to RVC API on localhost.
+
+set -e
+
+host="http://127.0.0.1:8000"
+
+url="${host}/inference"
+
+url+="?res_type=json"
+
+model_path=""
+index_path=""
+input_audio=""
+output_audio_suffix=""
+
+while [ $# -gt 0 ]; do
+  if [ "$1" == "--model_path" ]; then
+    model_path="$2"
+  elif [ "$1" == "--index_file" ]; then
+    index_path="$2"
+  elif [ "$1" == "--input_audio" ]; then
+    input_audio="$2"
+  else
+    arg_name="${1#--}"
+    arg_value="$2"
+
+    url+="&${arg_name}=${arg_value}"
+    output_audio_suffix+="-${arg_name}_${arg_value}"
+  fi
+
+  shift
+  shift
+done
+
+model_path_base="$(basename "${model_path}")"
+model_path_base_without_ext="${model_path_base%.*}"
+index_path_base="$(basename "${index_path}")"
+input_audio_base="$(basename "${input_audio}")"
+input_audio_dirname="$(dirname "${input_audio}")"
+output_audio_base_without_ext="${input_audio_base%.*}"
+output_audio="${input_audio_dirname}/${output_audio_base_without_ext}-${model_path_base_without_ext}${output_audio_suffix}.wav"
+
+cp "${model_path}" "./assets/weights/${model_path_base}"
+cp "${input_audio}" "./assets/audios/${input_audio_base}"
+
+if [ -f "${index_path}" ]; then
+  url+="&index_file=${index_path_base}"
+  cp "${index_path}" "./assets/indices/${index_path_base}"
+fi
+
+curl -X "POST" "${url}" \
+  -H "accept: application/json" \
+  -H "Content-Type: multipart/form-data" \
+  -F "modelpath=${model_path_base}" \
+  -F "input_audio=/audios/${input_audio_base}" \
+| jq -r '.audio' \
+| base64 -d > "${output_audio}"

assets-download.sh

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+#
+# Downloads required large files for RVC.
+
+function download() {
+  local path="$1"
+  echo "Downloading ${path}"
+  git lfs pull --include="${path}"
+}
+
+set -e
+
+REPO_FOLDER="VoiceConversionWebUI"
+
+assets_commit_hash="$1"
+assets_dir="$2"
+
+export GIT_CLONE_PROTECTION_ACTIVE=false
+export GIT_LFS_SKIP_SMUDGE=1
+
+git clone https://huggingface.co/lj1995/VoiceConversionWebUI "${REPO_FOLDER}"
+
+pushd "${REPO_FOLDER}"
+
+git config advice.detachedHead false
+
+git checkout "${assets_commit_hash}"
+
+unset GIT_LFS_SKIP_SMUDGE
+unset GIT_CLONE_PROTECTION_ACTIVE
+
+download "hubert_base.pt"
+download "pretrained"
+download "uvr5_weights"
+download "rmvpe.pt"
+download "rmvpe.onnx"
+
+rm -rf .git
+
+popd
+
+mkdir -p "${assets_dir}"
+
+mv "${REPO_FOLDER}/hubert_base.pt" "${assets_dir}/hubert_base.pt"
+
+mkdir -p "${assets_dir}/rmvpe"
+
+mv "${REPO_FOLDER}/rmvpe.pt" "${assets_dir}/rmvpe/rmvpe.pt"
+mv "${REPO_FOLDER}/rmvpe.onnx" "${assets_dir}/rmvpe/rmvpe.onnx"
+
+mv "${REPO_FOLDER}/pretrained" "${assets_dir}/pretrained"
+mv "${REPO_FOLDER}/uvr5_weights" "${assets_dir}/uvr5_weights"

docker-run.sh

@@ -0,0 +1,16 @@
+#!/usr/bin/env bash
+#
+# Runs RVC API in Docker.
+
+set -e
+
+tag="rvc"
+
+docker build -t "${tag}" .
+
+docker run -it \
+  -p 8000:8000 \
+  -v "${PWD}/assets/weights:/weights:ro" \
+  -v "${PWD}/assets/indices:/indices:ro" \
+  -v "${PWD}/assets/audios:/audios:ro" \
+  "${tag}"

docs/es/README.es.md

@@ -0,0 +1,171 @@
+<div align="center">
+
+<h1>Retrieval-based-Voice-Conversion</h1>
+Un framework de conversión de voz basado en VITS y fácil de usar.<br><br>
+
+[![madewithlove](https://img.shields.io/badge/hecho_con-%E2%9D%A4-red?style=for-the-badge&labelColor=orange
+)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion)
+
+<img src="https://counter.seku.su/cmoe?name=rvc&theme=r34" /><br>
+
+[![Licence](https://img.shields.io/github/license/RVC-Project/Retrieval-based-Voice-Conversion?style=for-the-badge)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion/blob/develop/LICENSE)
+
+[![Discord](https://img.shields.io/badge/Desarrolladores%20de%20RVC-Discord-7289DA?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/HcsmBBGyVk)
+
+</div>
+
+------
+
+
+> [!NOTE]
+> Actualmente en desarrollo... Proporcionado como biblioteca y API en rvc
+
+## Instalación y uso
+
+### Instalación estándar
+
+Primero, cree un directorio en su proyecto. La carpeta `assets` contendrá los modelos necesarios para la inferencia y el entrenamiento, y la carpeta `results` contendrá los resultados del entrenamiento.
+
+```sh
+rvc init
+```
+Esto creará la carpeta `assets` y `.env` en su directorio de trabajo.
+
+> [!WARNING]
+> El directorio debe de estar vacío o sin una carpeta de assets.
+
+### Instalación personalizada
+
+Si ya has descargado modelos o deseas cambiar estas configuraciones, edita el archivo `.env`.
+Si aún no tienes el archivo `.env`,
+
+```sh
+rvc env create
+```
+puedes crearlo.
+
+Además, para descargar un modelo, puedes utilizar
+
+```sh
+rvc dlmodel
+```
+o
+```
+rvc dlmodel {download_dir}
+```
+
+Finalmente, especifique la ubicación del modelo en el archivo env y estará listo.
+
+
+
+### Uso de la librería
+
+#### Inferir un audio
+```python
+from pathlib import Path
+
+from dotenv import load_dotenv
+from scipy.io import wavfile
+
+from rvc.modules.vc.modules import VC
+
+
+def main():
+      vc = VC()
+      vc.get_vc("{model.pth}")
+      tgt_sr, audio_opt, times, _ = vc.vc_inference(
+            1, Path("{InputAudio}")
+      )
+      wavfile.write("{OutputAudio}", tgt_sr, audio_opt)
+
+
+if __name__ == "__main__":
+      load_dotenv("{envPath}")
+      main()
+
+```
+
+### Uso en CLI
+
+#### Inferir un audio
+
+```sh
+rvc infer -m {model.pth} -i {input.wav} -o {output.wav}
+```
+
+| opción        | flag&nbsp; | tipo         | valor por defecto | descipción                                                                                                                                                                                                                                    |
+|---------------|------------|--------------|---------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| modelPath     | -m         | Path         | *requerido     | Ruta del modelo o nombre de archivo (se lee en el directorio establecido en env)                                                                                                                                                                                     |
+| inputPath     | -i         | Path         | *requerido     | Ruta o carpeta del audio de entrada                                                                                                                                                                                                                     |
+| outputPath    | -o         | Path         | *requerido     | Ruta o carpeta del audio de salida                                                                                                                                                                                                                    |
+| sid           | -s         | int          | 0             | ID del Orador/ Cantante                                                                                                                                                                                                                              |
+| f0_up_key     | -fu        | int          | 0             | Transponer (número entero, número de semitonos, subir una octava: 12, bajar una octava: -12)                                                                                                                                                      |
+| f0_method     | -fm        | str          | rmvpe         | Algoritmo de extracción de tono (pm, harvest, crepe, rmvpe)                                                                                                                                                                                          |
+| f0_file       | -ff        | Path \| None | None          | Archivo de curva F0 (opcional). Un tono por línea. Reemplaza el F0 predeterminado y la modulación de tono.                                                                                                                                                     |
+| index_file    | -if        | Path \| None | None          | Ruta al archivo index de características                                                                                                                                                                                                                 |
+| index_rate    | -if        | float        | 0.75          | Proporción de funciones de búsqueda (controla la fuerza del acento, demasiado alta tiene artifacting)                                                                                                                                                                      |
+| filter_radius | -fr        | int          | 3             | Si >=3: aplique el filtrado de mediana a los resultados del tono. El valor representa el radio del filtro y puede reducir la respiración                                                                                                               |
+| resample_sr   | -rsr       | int          | 0             | Vuelva a muestrear el audio de salida en el posprocesamiento hasta la frecuencia de muestreo final. Establecer en 0 para no remuestreo                                                                                                                                              |
+| rms_mix_rate  | -rmr       | float        | 0.25          | Ajuste la escala de la envolvente del volumen. Cuanto más cerca de 0, más imita el volumen de las voces originales. Puede ayudar a enmascarar el ruido y hacer que el volumen suene más natural cuando se establece en un nivel relativamente bajo. Más cerca de 1 habrá un volumen más alto y constante |
+| protect       | -p         | float        | 0.33          | Proteja las consonantes sordas y los sonidos respiratorios para evitar artefactos como el desgarro en la música electrónica. Establezca en 0.5 para desactivarlo. Disminuya el valor para aumentar la protección, pero puede reducir la precisión de la indexación                                 |
+
+### Uso de la API
+Primero, inicia el servidor.
+```sh
+rvc-api
+```
+o
+```sh
+poetry run poe rvc-api
+```
+
+#### Inferir audio
+
+##### Obtener como blob
+```sh
+curl -X 'POST' \
+      'http://127.0.0.1:8000/inference?res_type=blob' \
+      -H 'accept: application/json' \
+      -H 'Content-Type: multipart/form-data' \
+      -F 'modelpath={model.pth}' \
+      -F 'input={input audio path}'
+```
+
+##### Obtener como json (incluir tiempo)
+```sh
+curl -X 'POST' \
+      'http://127.0.0.1:8000/inference?res_type=json' \
+      -H 'accept: application/json' \
+      -H 'Content-Type: multipart/form-data' \
+      -F 'modelpath={model.pth}' \
+      -F 'input={input audio path}'
+```
+
+### Uso con Docker
+
+Compilar y ejecutar usando el script:
+
+```bash
+./docker-run.sh
+```
+
+**O** usar manuálmente:
+
+1. Compilar:
+
+   ```bash
+   docker build -t "rvc" .
+   ```
+
+2. Ejecutar:
+
+   ```bash
+   docker run -it \
+     -p 8000:8000 \
+     -v "${PWD}/assets/weights:/weights:ro" \
+     -v "${PWD}/assets/indices:/indices:ro" \
+     -v "${PWD}/assets/audios:/audios:ro" \
+     "rvc"
+   ```
+
+Recuerda que los pesos (weights), índices y audios de entrada se almacenan en `directorio-actual/assets`

docs/jp/README.ja.md

@@ -0,0 +1,119 @@
+<div align="center">
+
+<h1>Retrieval-based-Voice-Conversion</h1>
+An easy-to-use Voice Conversion framework based on VITS.<br><br>
+
+[![madewithlove](https://img.shields.io/badge/made_with-%E2%9D%A4-red?style=for-the-badge&labelColor=orange
+)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion)
+
+<img src="https://counter.seku.su/cmoe?name=rvc&theme=r34" /><br>
+
+[![Licence](https://img.shields.io/github/license/RVC-Project/Retrieval-based-Voice-Conversion?style=for-the-badge)](https://github.com/RVC-Project/Retrieval-based-Voice-Conversion/blob/develop/LICENSE)
+
+[![Discord](https://img.shields.io/badge/RVC%20Developers-Discord-7289DA?style=for-the-badge&logo=discord&logoColor=white)](https://discord.gg/HcsmBBGyVk)
+
+</div>
+
+------
+
+
+> [!NOTE]
+> 現在開発中です...rvcのライブラリとAPIを提供する予定です。
+
+## Installation and usage
+
+### Standard Setup
+
+最初にプロジェクトにディレクトリを作成します。`assets`フォルダには推論や学習に必要なモデル、`result`フォルダには学習の結果が保存されます。
+
+```sh
+rvc init
+```
+
+これにより、作業ディレクトリに`assets`フォルダと`.env`が作成されます。
+> [!WARNING]
+> この時、ディレクトリは空もしくは`assets`フォルダおよび`.env`ファイルがない状態にしてください
+
+### Custom Setup
+
+既にモデルをダウンロードしている場合や、これらの構成を変更したい場合、`.env`ファイルを編集してください。
+まだ`.env`ファイルがない場合、
+
+```sh
+rvc env create
+```
+
+にて作成できます。
+
+また、モデルをダウンロードするときは
+
+```sh
+rvc dlmodel
+```
+もしくは
+```
+rvc dlmodel {download_dir}
+```
+
+にてダウンロードできます。
+
+最後に、envファイルにてモデルの場所などを指定してあげれば、終了です！
+
+
+### CLI Usage
+
+#### Inference Audio
+
+```sh
+rvc infer -m {model.pth} -i {input.wav} -o {output.wav}
+```
+
+| option        | flag&nbsp; | type         | default value | description                                                                                                                                                                                                                                    |
+|---------------|------------|--------------|---------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
+| modelPath     | -m         | Path         | *required     | Model path or filename (reads in the directory set in env)                                                                                                                                                                                     |
+| inputPath     | -i         | Path         | *required     | Input audio path or folder                                                                                                                                                                                                                     |
+| outputPath    | -o         | Path         | *required     | Output audio path or folder                                                                                                                                                                                                                    |
+| sid           | -s         | int          | 0             | Speaker/Singer ID                                                                                                                                                                                                                              |
+| f0_up_key     | -fu        | int          | 0             | Transpose (integer, number of semitones, raise by an octave: 12, lower by an octave: -12)                                                                                                                                                      |
+| f0_method     | -fm        | str          | rmvpe         | pitch extraction algorithm (pm, harvest, crepe, rmvpe                                                                                                                                                                                          |
+| f0_file       | -ff        | Path \| None | None          | F0 curve file (optional). One pitch per line. Replaces the default F0 and pitch modulation                                                                                                                                                     |
+| index_file    | -if        | Path \| None | None          | Path to the feature index file                                                                                                                                                                                                                 |
+| index_rate    | -if        | float        | 0.75          | Search feature ratio (controls accent strength, too high has artifacting)                                                                                                                                                                      |
+| filter_radius | -fr        | int          | 3             | If >=3: apply median filtering to the harvested pitch results. The value represents the filter radius and can reduce breathiness                                                                                                               |
+| resample_sr   | -rsr       | int          | 0             | Resample the output audio in post-processing to the final sample rate. Set to 0 for no resampling                                                                                                                                              |
+| rms_mix_rate  | -rmr       | float        | 0.25          | Adjust the volume envelope scaling. Closer to 0, the more it mimicks the volume of the original vocals. Can help mask noise and make volume sound more natural when set relatively low. Closer to 1 will be more of a consistently loud volume |
+| protect       | -p         | float        | 0.33          | Protect voiceless consonants and breath sounds to prevent artifacts such as tearing in electronic music. Set to 0.5 to disable. Decrease the value to increase protection, but it may reduce indexing accuracy                                 |
+
+
+### API Usage
+最初に、サーバーを立ち上げます。
+```sh
+rvc-api
+```
+または
+```sh
+poetry run poe rvc-api
+```
+にて実行されます。
+
+#### Inference Audio
+
+##### blobでレスポンスを受け取る
+```sh
+curl -X 'POST' \
+      'http://127.0.0.1:8000/inference?res_type=blob' \
+      -H 'accept: application/json' \
+      -H 'Content-Type: multipart/form-data' \
+      -F 'modelpath={model.pth}' \
+      -F 'input={input audio path}'
+```
+
+##### jsonでレスポンス！(include time)
+```sh
+curl -X 'POST' \
+      'http://127.0.0.1:8000/inference?res_type=json' \
+      -H 'accept: application/json' \
+      -H 'Content-Type: multipart/form-data' \
+      -F 'modelpath={model.pth}' \
+      -F 'input={input audio path}'
+```

pyproject.toml

@@ -0,0 +1,44 @@
+[tool.poetry]
+name = "rvc"
+version = "0.3.5"
+description = "An easy-to-use Voice Conversion framework based on VITS."
+authors = ["Ftps <[email protected]>"]
+readme = "README.md"
+
+[tool.poetry.urls]
+github = "https://github.com/RVC-Project/Retrieval-based-Voice-Conversion"
+
+[tool.poetry.dependencies]
+python = "^3.10"
+torch = "^2.1.0"
+#fairseq = "^0.12.2"
+fairseq = {git = "https://github.com/Tps-F/fairseq.git", branch="main"}
+soundfile = "^0.12.1"
+librosa = "^0.10.1"
+praat-parselmouth = "^0.4.3"
+pyworld = "^0.3.4"
+torchcrepe = "^0.0.22"
+av = "^11.0.0"
+faiss-cpu = "^1.7.4"
+python-dotenv = "^1.0.0"
+pydub = "^0.25.1"
+click = "^8.1.7"
+tensorboardx = "^2.6.2.2"
+poethepoet = "^0.24.4"
+uvicorn = "^0.26.0"
+fastapi = "^0.109.0"
+python-multipart = "^0.0.6"
+numba = "0.59.0rc1"
+
+[tool.poetry.extras]
+api = ["uvicorn", "fastapi"]
+
+[tool.poetry.scripts]
+rvc = "rvc.wrapper.cli.cli:main"
+
+[tool.poe.tasks]
+rvc-api = "uvicorn rvc.wrapper.api.api:app --host 0.0.0.0 --port 8000 --reload"
+
+[build-system]
+requires = ["poetry-core"]
+build-backend = "poetry.core.masonry.api"

rvc/configs/__init__.py

@@ -0,0 +1 @@
+from rvc.configs.config import Config

rvc/configs/config.py

@@ -0,0 +1,197 @@
+import argparse
+import json
+import logging
+import os
+import sys
+from multiprocessing import cpu_count
+
+import torch
+
+try:
+    import intel_extension_for_pytorch as ipex
+
+    if torch.xpu.is_available():
+        from rvc.lib.ipex import ipex_init
+
+        ipex_init()
+except (ImportError, Exception):
+    pass
+
+logger: logging.Logger = logging.getLogger(__name__)
+
+
+version_config_list: list = [
+    os.path.join(root, file)
+    for root, dirs, files in os.walk(os.path.dirname(os.path.abspath(__file__)))
+    for file in files
+    if file.endswith(".json")
+]
+
+
+class Config:
+    def __new__(cls):
+        if not hasattr(cls, "_instance"):
+            cls._instance = super().__new__(cls)
+        return cls._instance
+
+    def __init__(self):
+        self.device: str = "cuda:0"
+        self.is_half: bool = True
+        self.use_jit: bool = False
+        self.n_cpu: int = cpu_count()
+        self.gpu_name: str | None = None
+        self.json_config = self.load_config_json()
+        self.gpu_mem: int | None = None
+        self.instead: str | None = None
+        (
+            self.python_cmd,
+            self.listen_port,
+            self.noparallel,
+            self.noautoopen,
+            self.dml,
+        ) = self.arg_parse()
+        self.x_pad, self.x_query, self.x_center, self.x_max = self.device_config()
+
+    @staticmethod
+    def load_config_json() -> dict:
+        return {
+            config_file: json.load(open(config_file, "r"))
+            for config_file in version_config_list
+        }
+
+    @staticmethod
+    def arg_parse() -> tuple:
+        parser: argparse.ArgumentParser = argparse.ArgumentParser()
+        parser.add_argument("--port", type=int, default=7865, help="Listen port")
+        parser.add_argument(
+            "--pycmd",
+            type=str,
+            default=sys.executable or "python",
+            help="Python command",
+        )
+        parser.add_argument(
+            "--noparallel", action="store_true", help="Disable parallel processing"
+        )
+        parser.add_argument(
+            "--noautoopen",
+            action="store_true",
+            help="Do not open in browser automatically",
+        )
+        parser.add_argument(
+            "--dml",
+            action="store_true",
+            help="torch_dml",
+        )
+        cmd_opts: argparse.Namespace
+        cmd_opts, _ = parser.parse_known_args()
+
+        cmd_opts.port = cmd_opts.port if 0 <= cmd_opts.port <= 65535 else 7865
+
+        return (
+            cmd_opts.pycmd,
+            cmd_opts.port,
+            cmd_opts.noparallel,
+            cmd_opts.noautoopen,
+            cmd_opts.dml,
+        )
+
+    @staticmethod
+    def has_mps() -> bool:
+        return torch.backends.mps.is_available() and not torch.zeros(1).to(
+            torch.device("mps")
+        )
+
+    @staticmethod
+    def has_xpu() -> bool:
+        return hasattr(torch, "xpu") and torch.xpu.is_available()
+
+    def params_config(self) -> tuple:
+        if self.gpu_mem is not None and self.gpu_mem <= 4:
+            x_pad = 1
+            x_query = 5
+            x_center = 30
+            x_max = 32
+        elif self.is_half:
+            # 6G PU_RAM conf
+            x_pad = 3
+            x_query = 10
+            x_center = 60
+            x_max = 65
+        else:
+            # 5G GPU_RAM conf
+            x_pad = 1
+            x_query = 6
+            x_center = 38
+            x_max = 41
+        return x_pad, x_query, x_center, x_max
+
+    def use_cuda(self) -> None:
+        if self.has_xpu():
+            self.device = self.instead = "xpu:0"
+            self.is_half = True
+        i_device = int(self.device.split(":")[-1])
+        self.gpu_name = torch.cuda.get_device_name(i_device)
+        if (
+            ("16" in self.gpu_name and "V100" not in self.gpu_name.upper())
+            or "P40" in self.gpu_name.upper()
+            or "P10" in self.gpu_name.upper()
+            or "1060" in self.gpu_name
+            or "1070" in self.gpu_name
+            or "1080" in self.gpu_name
+        ):
+            logger.info(f"Found GPU {self.gpu_name}, force to fp32")
+            self.is_half = False
+            self.use_fp32_config()
+        else:
+            logger.info(f"Found GPU {self.gpu_name}")
+        self.gpu_mem = int(
+            torch.cuda.get_device_properties(i_device).total_memory / 1024 / 1024 / 1024
+            + 0.4
+        )
+
+    def use_mps(self) -> None:
+        self.device = self.instead = "mps"
+        self.is_half = False
+        self.use_fp32_config()
+        self.params_config()
+
+    def use_dml(self) -> None:
+        import torch_directml
+
+        self.device = torch_directml.device(torch_directml.default_device())
+        self.is_half = False
+        self.params_config()
+
+    def use_cpu(self) -> None:
+        self.device = self.instead = "cpu"
+        self.is_half = False
+        self.use_fp32_config()
+        self.params_config()
+
+    def use_fp32_config(self) -> None:
+        for config_file, data in self.json_config.items():
+            try:
+                data["train"]["fp16_run"] = False
+                with open(config_file, "w") as json_file:
+                    json.dump(data, json_file, indent=4)
+            except Exception as e:
+                logger.info(f"Error updating {config_file}: {str(e)}")
+        logger.info("overwrite configs.json")
+
+    def device_config(self) -> tuple:
+        if torch.cuda.is_available():
+            self.use_cuda()
+        elif self.has_mps():
+            logger.info("No supported Nvidia GPU found")
+            self.use_mps()
+        elif self.dml:
+            self.use_dml()
+        else:
+            logger.info("No supported Nvidia GPU found")
+            self.device = self.instead = "cpu"
+            self.is_half = False
+            self.use_fp32_config()
+
+        logger.info(f"Use {self.dml or self.instead} instead")
+        logger.info(f"is_half:{self.is_half}, device:{self.device}")
+        return self.params_config()

rvc/configs/v1/32k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 32000,
+    "filter_length": 1024,
+    "hop_length": 320,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,4,2,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

rvc/configs/v1/40k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 40000,
+    "filter_length": 2048,
+    "hop_length": 400,
+    "win_length": 2048,
+    "n_mel_channels": 125,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,10,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

rvc/configs/v1/48k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 11520,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 2048,
+    "hop_length": 480,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,6,2,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [16,16,4,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

rvc/configs/v2/32k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 12800,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 32000,
+    "filter_length": 1024,
+    "hop_length": 320,
+    "win_length": 1024,
+    "n_mel_channels": 80,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [10,8,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [20,16,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

rvc/configs/v2/48k.json

@@ -0,0 +1,46 @@
+{
+  "train": {
+    "log_interval": 200,
+    "seed": 1234,
+    "epochs": 20000,
+    "learning_rate": 1e-4,
+    "betas": [0.8, 0.99],
+    "eps": 1e-9,
+    "batch_size": 4,
+    "fp16_run": true,
+    "lr_decay": 0.999875,
+    "segment_size": 17280,
+    "init_lr_ratio": 1,
+    "warmup_epochs": 0,
+    "c_mel": 45,
+    "c_kl": 1.0
+  },
+  "data": {
+    "max_wav_value": 32768.0,
+    "sampling_rate": 48000,
+    "filter_length": 2048,
+    "hop_length": 480,
+    "win_length": 2048,
+    "n_mel_channels": 128,
+    "mel_fmin": 0.0,
+    "mel_fmax": null
+  },
+  "model": {
+    "inter_channels": 192,
+    "hidden_channels": 192,
+    "filter_channels": 768,
+    "n_heads": 2,
+    "n_layers": 6,
+    "kernel_size": 3,
+    "p_dropout": 0,
+    "resblock": "1",
+    "resblock_kernel_sizes": [3,7,11],
+    "resblock_dilation_sizes": [[1,3,5], [1,3,5], [1,3,5]],
+    "upsample_rates": [12,10,2,2],
+    "upsample_initial_channel": 512,
+    "upsample_kernel_sizes": [24,20,4,4],
+    "use_spectral_norm": false,
+    "gin_channels": 256,
+    "spk_embed_dim": 109
+  }
+}

rvc/lib/audio.py

@@ -0,0 +1,70 @@
+import os
+import traceback
+from io import BytesIO
+
+import av
+import librosa
+import numpy as np
+
+
+def wav2(i, o, format):
+    inp = av.open(i, "rb")
+    if format == "m4a":
+        format = "mp4"
+    out = av.open(o, "wb", format=format)
+    if format == "ogg":
+        format = "libvorbis"
+    if format == "mp4":
+        format = "aac"
+
+    ostream = out.add_stream(format)
+
+    for frame in inp.decode(audio=0):
+        for p in ostream.encode(frame):
+            out.mux(p)
+
+    for p in ostream.encode(None):
+        out.mux(p)
+
+    out.close()
+    inp.close()
+
+
+def audio2(i, o, format, sr):
+    inp = av.open(i, "rb")
+    out = av.open(o, "wb", format=format)
+    if format == "ogg":
+        format = "libvorbis"
+    if format == "f32le":
+        format = "pcm_f32le"
+
+    ostream = out.add_stream(format, channels=1)
+    ostream.sample_rate = sr
+
+    for frame in inp.decode(audio=0):
+        for p in ostream.encode(frame):
+            out.mux(p)
+
+    out.close()
+    inp.close()
+
+
+def load_audio(file, sr):
+    if not os.path.exists(file):
+        raise RuntimeError(
+            "You input a wrong audio path that does not exists, please fix it!"
+        )
+    try:
+        with open(file, "rb") as f:
+            with BytesIO() as out:
+                audio2(f, out, "f32le", sr)
+                return np.frombuffer(out.getvalue(), np.float32).flatten()
+
+    except AttributeError:
+        audio = file[1] / 32768.0
+        if len(audio.shape) == 2:
+            audio = np.mean(audio, -1)
+        return librosa.resample(audio, orig_sr=file[0], target_sr=16000)
+
+    except Exception:
+        raise RuntimeError(traceback.format_exc())

rvc/lib/infer_pack/attentions.py

@@ -0,0 +1,459 @@
+import copy
+import math
+from typing import Optional
+
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+from rvc.lib.infer_pack import commons, modules
+from rvc.lib.infer_pack.modules import LayerNorm
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        window_size=10,
+        **kwargs
+    ):
+        super(Encoder, self).__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = int(n_layers)
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+
+        self.drop = nn.Dropout(p_dropout)
+        self.attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    window_size=window_size,
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask):
+        attn_mask = x_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        zippep = zip(
+            self.attn_layers, self.norm_layers_1, self.ffn_layers, self.norm_layers_2
+        )
+        for attn_layers, norm_layers_1, ffn_layers, norm_layers_2 in zippep:
+            y = attn_layers(x, x, attn_mask)
+            y = self.drop(y)
+            x = norm_layers_1(x + y)
+
+            y = ffn_layers(x, x_mask)
+            y = self.drop(y)
+            x = norm_layers_2(x + y)
+        x = x * x_mask
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size=1,
+        p_dropout=0.0,
+        proximal_bias=False,
+        proximal_init=True,
+        **kwargs
+    ):
+        super(Decoder, self).__init__()
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+
+        self.drop = nn.Dropout(p_dropout)
+        self.self_attn_layers = nn.ModuleList()
+        self.norm_layers_0 = nn.ModuleList()
+        self.encdec_attn_layers = nn.ModuleList()
+        self.norm_layers_1 = nn.ModuleList()
+        self.ffn_layers = nn.ModuleList()
+        self.norm_layers_2 = nn.ModuleList()
+        for i in range(self.n_layers):
+            self.self_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels,
+                    hidden_channels,
+                    n_heads,
+                    p_dropout=p_dropout,
+                    proximal_bias=proximal_bias,
+                    proximal_init=proximal_init,
+                )
+            )
+            self.norm_layers_0.append(LayerNorm(hidden_channels))
+            self.encdec_attn_layers.append(
+                MultiHeadAttention(
+                    hidden_channels, hidden_channels, n_heads, p_dropout=p_dropout
+                )
+            )
+            self.norm_layers_1.append(LayerNorm(hidden_channels))
+            self.ffn_layers.append(
+                FFN(
+                    hidden_channels,
+                    hidden_channels,
+                    filter_channels,
+                    kernel_size,
+                    p_dropout=p_dropout,
+                    causal=True,
+                )
+            )
+            self.norm_layers_2.append(LayerNorm(hidden_channels))
+
+    def forward(self, x, x_mask, h, h_mask):
+        """
+        x: decoder input
+        h: encoder output
+        """
+        self_attn_mask = commons.subsequent_mask(x_mask.size(2)).to(
+            device=x.device, dtype=x.dtype
+        )
+        encdec_attn_mask = h_mask.unsqueeze(2) * x_mask.unsqueeze(-1)
+        x = x * x_mask
+        for i in range(self.n_layers):
+            y = self.self_attn_layers[i](x, x, self_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_0[i](x + y)
+
+            y = self.encdec_attn_layers[i](x, h, encdec_attn_mask)
+            y = self.drop(y)
+            x = self.norm_layers_1[i](x + y)
+
+            y = self.ffn_layers[i](x, x_mask)
+            y = self.drop(y)
+            x = self.norm_layers_2[i](x + y)
+        x = x * x_mask
+        return x
+
+
+class MultiHeadAttention(nn.Module):
+    def __init__(
+        self,
+        channels,
+        out_channels,
+        n_heads,
+        p_dropout=0.0,
+        window_size=None,
+        heads_share=True,
+        block_length=None,
+        proximal_bias=False,
+        proximal_init=False,
+    ):
+        super(MultiHeadAttention, self).__init__()
+        assert channels % n_heads == 0
+
+        self.channels = channels
+        self.out_channels = out_channels
+        self.n_heads = n_heads
+        self.p_dropout = p_dropout
+        self.window_size = window_size
+        self.heads_share = heads_share
+        self.block_length = block_length
+        self.proximal_bias = proximal_bias
+        self.proximal_init = proximal_init
+        self.attn = None
+
+        self.k_channels = channels // n_heads
+        self.conv_q = nn.Conv1d(channels, channels, 1)
+        self.conv_k = nn.Conv1d(channels, channels, 1)
+        self.conv_v = nn.Conv1d(channels, channels, 1)
+        self.conv_o = nn.Conv1d(channels, out_channels, 1)
+        self.drop = nn.Dropout(p_dropout)
+
+        if window_size is not None:
+            n_heads_rel = 1 if heads_share else n_heads
+            rel_stddev = self.k_channels**-0.5
+            self.emb_rel_k = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+            self.emb_rel_v = nn.Parameter(
+                torch.randn(n_heads_rel, window_size * 2 + 1, self.k_channels)
+                * rel_stddev
+            )
+
+        nn.init.xavier_uniform_(self.conv_q.weight)
+        nn.init.xavier_uniform_(self.conv_k.weight)
+        nn.init.xavier_uniform_(self.conv_v.weight)
+        if proximal_init:
+            with torch.no_grad():
+                self.conv_k.weight.copy_(self.conv_q.weight)
+                self.conv_k.bias.copy_(self.conv_q.bias)
+
+    def forward(
+        self, x: torch.Tensor, c: torch.Tensor, attn_mask: Optional[torch.Tensor] = None
+    ):
+        q = self.conv_q(x)
+        k = self.conv_k(c)
+        v = self.conv_v(c)
+
+        x, _ = self.attention(q, k, v, mask=attn_mask)
+
+        x = self.conv_o(x)
+        return x
+
+    def attention(
+        self,
+        query: torch.Tensor,
+        key: torch.Tensor,
+        value: torch.Tensor,
+        mask: Optional[torch.Tensor] = None,
+    ):
+        # reshape [b, d, t] -> [b, n_h, t, d_k]
+        b, d, t_s = key.size()
+        t_t = query.size(2)
+        query = query.view(b, self.n_heads, self.k_channels, t_t).transpose(2, 3)
+        key = key.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+        value = value.view(b, self.n_heads, self.k_channels, t_s).transpose(2, 3)
+
+        scores = torch.matmul(query / math.sqrt(self.k_channels), key.transpose(-2, -1))
+        if self.window_size is not None:
+            assert (
+                t_s == t_t
+            ), "Relative attention is only available for self-attention."
+            key_relative_embeddings = self._get_relative_embeddings(self.emb_rel_k, t_s)
+            rel_logits = self._matmul_with_relative_keys(
+                query / math.sqrt(self.k_channels), key_relative_embeddings
+            )
+            scores_local = self._relative_position_to_absolute_position(rel_logits)
+            scores = scores + scores_local
+        if self.proximal_bias:
+            assert t_s == t_t, "Proximal bias is only available for self-attention."
+            scores = scores + self._attention_bias_proximal(t_s).to(
+                device=scores.device, dtype=scores.dtype
+            )
+        if mask is not None:
+            scores = scores.masked_fill(mask == 0, -1e4)
+            if self.block_length is not None:
+                assert (
+                    t_s == t_t
+                ), "Local attention is only available for self-attention."
+                block_mask = (
+                    torch.ones_like(scores)
+                    .triu(-self.block_length)
+                    .tril(self.block_length)
+                )
+                scores = scores.masked_fill(block_mask == 0, -1e4)
+        p_attn = F.softmax(scores, dim=-1)  # [b, n_h, t_t, t_s]
+        p_attn = self.drop(p_attn)
+        output = torch.matmul(p_attn, value)
+        if self.window_size is not None:
+            relative_weights = self._absolute_position_to_relative_position(p_attn)
+            value_relative_embeddings = self._get_relative_embeddings(
+                self.emb_rel_v, t_s
+            )
+            output = output + self._matmul_with_relative_values(
+                relative_weights, value_relative_embeddings
+            )
+        output = (
+            output.transpose(2, 3).contiguous().view(b, d, t_t)
+        )  # [b, n_h, t_t, d_k] -> [b, d, t_t]
+        return output, p_attn
+
+    def _matmul_with_relative_values(self, x, y):
+        """
+        x: [b, h, l, m]
+        y: [h or 1, m, d]
+        ret: [b, h, l, d]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0))
+        return ret
+
+    def _matmul_with_relative_keys(self, x, y):
+        """
+        x: [b, h, l, d]
+        y: [h or 1, m, d]
+        ret: [b, h, l, m]
+        """
+        ret = torch.matmul(x, y.unsqueeze(0).transpose(-2, -1))
+        return ret
+
+    def _get_relative_embeddings(self, relative_embeddings, length: int):
+        max_relative_position = 2 * self.window_size + 1
+        # Pad first before slice to avoid using cond ops.
+        pad_length: int = max(length - (self.window_size + 1), 0)
+        slice_start_position = max((self.window_size + 1) - length, 0)
+        slice_end_position = slice_start_position + 2 * length - 1
+        if pad_length > 0:
+            padded_relative_embeddings = F.pad(
+                relative_embeddings,
+                # commons.convert_pad_shape([[0, 0], [pad_length, pad_length], [0, 0]]),
+                [0, 0, pad_length, pad_length, 0, 0],
+            )
+        else:
+            padded_relative_embeddings = relative_embeddings
+        used_relative_embeddings = padded_relative_embeddings[
+            :, slice_start_position:slice_end_position
+        ]
+        return used_relative_embeddings
+
+    def _relative_position_to_absolute_position(self, x):
+        """
+        x: [b, h, l, 2*l-1]
+        ret: [b, h, l, l]
+        """
+        batch, heads, length, _ = x.size()
+        # Concat columns of pad to shift from relative to absolute indexing.
+        x = F.pad(
+            x,
+            #   commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, 1]])
+            [0, 1, 0, 0, 0, 0, 0, 0],
+        )
+
+        # Concat extra elements so to add up to shape (len+1, 2*len-1).
+        x_flat = x.view([batch, heads, length * 2 * length])
+        x_flat = F.pad(
+            x_flat,
+            # commons.convert_pad_shape([[0, 0], [0, 0], [0, int(length) - 1]])
+            [0, int(length) - 1, 0, 0, 0, 0],
+        )
+
+        # Reshape and slice out the padded elements.
+        x_final = x_flat.view([batch, heads, length + 1, 2 * length - 1])[
+            :, :, :length, length - 1 :
+        ]
+        return x_final
+
+    def _absolute_position_to_relative_position(self, x):
+        """
+        x: [b, h, l, l]
+        ret: [b, h, l, 2*l-1]
+        """
+        batch, heads, length, _ = x.size()
+        # padd along column
+        x = F.pad(
+            x,
+            # commons.convert_pad_shape([[0, 0], [0, 0], [0, 0], [0, int(length) - 1]])
+            [0, int(length) - 1, 0, 0, 0, 0, 0, 0],
+        )
+        x_flat = x.view([batch, heads, int(length**2) + int(length * (length - 1))])
+        # add 0's in the beginning that will skew the elements after reshape
+        x_flat = F.pad(
+            x_flat,
+            #    commons.convert_pad_shape([[0, 0], [0, 0], [int(length), 0]])
+            [length, 0, 0, 0, 0, 0],
+        )
+        x_final = x_flat.view([batch, heads, length, 2 * length])[:, :, :, 1:]
+        return x_final
+
+    def _attention_bias_proximal(self, length: int):
+        """Bias for self-attention to encourage attention to close positions.
+        Args:
+          length: an integer scalar.
+        Returns:
+          a Tensor with shape [1, 1, length, length]
+        """
+        r = torch.arange(length, dtype=torch.float32)
+        diff = torch.unsqueeze(r, 0) - torch.unsqueeze(r, 1)
+        return torch.unsqueeze(torch.unsqueeze(-torch.log1p(torch.abs(diff)), 0), 0)
+
+
+class FFN(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        filter_channels,
+        kernel_size,
+        p_dropout=0.0,
+        activation: str = None,
+        causal=False,
+    ):
+        super(FFN, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.activation = activation
+        self.causal = causal
+        self.is_activation = True if activation == "gelu" else False
+        # if causal:
+        #     self.padding = self._causal_padding
+        # else:
+        #     self.padding = self._same_padding
+
+        self.conv_1 = nn.Conv1d(in_channels, filter_channels, kernel_size)
+        self.conv_2 = nn.Conv1d(filter_channels, out_channels, kernel_size)
+        self.drop = nn.Dropout(p_dropout)
+
+    def padding(self, x: torch.Tensor, x_mask: torch.Tensor) -> torch.Tensor:
+        if self.causal:
+            padding = self._causal_padding(x * x_mask)
+        else:
+            padding = self._same_padding(x * x_mask)
+        return padding
+
+    def forward(self, x: torch.Tensor, x_mask: torch.Tensor):
+        x = self.conv_1(self.padding(x, x_mask))
+        if self.is_activation:
+            x = x * torch.sigmoid(1.702 * x)
+        else:
+            x = torch.relu(x)
+        x = self.drop(x)
+
+        x = self.conv_2(self.padding(x, x_mask))
+        return x * x_mask
+
+    def _causal_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l: int = self.kernel_size - 1
+        pad_r: int = 0
+        # padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(
+            x,
+            #   commons.convert_pad_shape(padding)
+            [pad_l, pad_r, 0, 0, 0, 0],
+        )
+        return x
+
+    def _same_padding(self, x):
+        if self.kernel_size == 1:
+            return x
+        pad_l: int = (self.kernel_size - 1) // 2
+        pad_r: int = self.kernel_size // 2
+        # padding = [[0, 0], [0, 0], [pad_l, pad_r]]
+        x = F.pad(
+            x,
+            #   commons.convert_pad_shape(padding)
+            [pad_l, pad_r, 0, 0, 0, 0],
+        )
+        return x

rvc/lib/infer_pack/commons.py

@@ -0,0 +1,172 @@
+import math
+from typing import List, Optional
+
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import functional as F
+
+
+def init_weights(m, mean=0.0, std=0.01):
+    classname = m.__class__.__name__
+    if classname.find("Conv") != -1:
+        m.weight.data.normal_(mean, std)
+
+
+def get_padding(kernel_size, dilation=1):
+    return int((kernel_size * dilation - dilation) / 2)
+
+
+# def convert_pad_shape(pad_shape):
+#     l = pad_shape[::-1]
+#     pad_shape = [item for sublist in l for item in sublist]
+#     return pad_shape
+
+
+def kl_divergence(m_p, logs_p, m_q, logs_q):
+    """KL(P||Q)"""
+    kl = (logs_q - logs_p) - 0.5
+    kl += (
+        0.5 * (torch.exp(2.0 * logs_p) + ((m_p - m_q) ** 2)) * torch.exp(-2.0 * logs_q)
+    )
+    return kl
+
+
+def rand_gumbel(shape):
+    """Sample from the Gumbel distribution, protect from overflows."""
+    uniform_samples = torch.rand(shape) * 0.99998 + 0.00001
+    return -torch.log(-torch.log(uniform_samples))
+
+
+def rand_gumbel_like(x):
+    g = rand_gumbel(x.size()).to(dtype=x.dtype, device=x.device)
+    return g
+
+
+def slice_segments(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, :, idx_str:idx_end]
+    return ret
+
+
+def slice_segments2(x, ids_str, segment_size=4):
+    ret = torch.zeros_like(x[:, :segment_size])
+    for i in range(x.size(0)):
+        idx_str = ids_str[i]
+        idx_end = idx_str + segment_size
+        ret[i] = x[i, idx_str:idx_end]
+    return ret
+
+
+def rand_slice_segments(x, x_lengths=None, segment_size=4):
+    b, d, t = x.size()
+    if x_lengths is None:
+        x_lengths = t
+    ids_str_max = x_lengths - segment_size + 1
+    ids_str = (torch.rand([b]).to(device=x.device) * ids_str_max).to(dtype=torch.long)
+    ret = slice_segments(x, ids_str, segment_size)
+    return ret, ids_str
+
+
+def get_timing_signal_1d(length, channels, min_timescale=1.0, max_timescale=1.0e4):
+    position = torch.arange(length, dtype=torch.float)
+    num_timescales = channels // 2
+    log_timescale_increment = math.log(float(max_timescale) / float(min_timescale)) / (
+        num_timescales - 1
+    )
+    inv_timescales = min_timescale * torch.exp(
+        torch.arange(num_timescales, dtype=torch.float) * -log_timescale_increment
+    )
+    scaled_time = position.unsqueeze(0) * inv_timescales.unsqueeze(1)
+    signal = torch.cat([torch.sin(scaled_time), torch.cos(scaled_time)], 0)
+    signal = F.pad(signal, [0, 0, 0, channels % 2])
+    signal = signal.view(1, channels, length)
+    return signal
+
+
+def add_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return x + signal.to(dtype=x.dtype, device=x.device)
+
+
+def cat_timing_signal_1d(x, min_timescale=1.0, max_timescale=1.0e4, axis=1):
+    b, channels, length = x.size()
+    signal = get_timing_signal_1d(length, channels, min_timescale, max_timescale)
+    return torch.cat([x, signal.to(dtype=x.dtype, device=x.device)], axis)
+
+
+def subsequent_mask(length):
+    mask = torch.tril(torch.ones(length, length)).unsqueeze(0).unsqueeze(0)
+    return mask
+
+
+@torch.jit.script
+def fused_add_tanh_sigmoid_multiply(input_a, input_b, n_channels):
+    n_channels_int = n_channels[0]
+    in_act = input_a + input_b
+    t_act = torch.tanh(in_act[:, :n_channels_int, :])
+    s_act = torch.sigmoid(in_act[:, n_channels_int:, :])
+    acts = t_act * s_act
+    return acts
+
+
+# def convert_pad_shape(pad_shape):
+#     l = pad_shape[::-1]
+#     pad_shape = [item for sublist in l for item in sublist]
+#     return pad_shape
+
+
+def convert_pad_shape(pad_shape: List[List[int]]) -> List[int]:
+    return torch.tensor(pad_shape).flip(0).reshape(-1).int().tolist()
+
+
+def shift_1d(x):
+    x = F.pad(x, convert_pad_shape([[0, 0], [0, 0], [1, 0]]))[:, :, :-1]
+    return x
+
+
+def sequence_mask(length: torch.Tensor, max_length: Optional[int] = None):
+    if max_length is None:
+        max_length = length.max()
+    x = torch.arange(max_length, dtype=length.dtype, device=length.device)
+    return x.unsqueeze(0) < length.unsqueeze(1)
+
+
+def generate_path(duration, mask):
+    """
+    duration: [b, 1, t_x]
+    mask: [b, 1, t_y, t_x]
+    """
+    device = duration.device
+
+    b, _, t_y, t_x = mask.shape
+    cum_duration = torch.cumsum(duration, -1)
+
+    cum_duration_flat = cum_duration.view(b * t_x)
+    path = sequence_mask(cum_duration_flat, t_y).to(mask.dtype)
+    path = path.view(b, t_x, t_y)
+    path = path - F.pad(path, convert_pad_shape([[0, 0], [1, 0], [0, 0]]))[:, :-1]
+    path = path.unsqueeze(1).transpose(2, 3) * mask
+    return path
+
+
+def clip_grad_value_(parameters, clip_value, norm_type=2):
+    if isinstance(parameters, torch.Tensor):
+        parameters = [parameters]
+    parameters = list(filter(lambda p: p.grad is not None, parameters))
+    norm_type = float(norm_type)
+    if clip_value is not None:
+        clip_value = float(clip_value)
+
+    total_norm = 0
+    for p in parameters:
+        param_norm = p.grad.data.norm(norm_type)
+        total_norm += param_norm.item() ** norm_type
+        if clip_value is not None:
+            p.grad.data.clamp_(min=-clip_value, max=clip_value)
+    total_norm = total_norm ** (1.0 / norm_type)
+    return total_norm

rvc/lib/infer_pack/models.py

@@ -0,0 +1,1426 @@
+import logging
+import math
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import AvgPool1d, Conv1d, Conv2d, ConvTranspose1d
+from torch.nn import functional as F
+from torch.nn.utils import remove_weight_norm, spectral_norm, weight_norm
+
+from rvc.lib.infer_pack import attentions, commons, modules
+from rvc.lib.infer_pack.commons import get_padding, init_weights
+
+has_xpu = bool(hasattr(torch, "xpu") and torch.xpu.is_available())
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super(TextEncoder256, self).__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(
+        self, phone: torch.Tensor, pitch: Optional[torch.Tensor], lengths: torch.Tensor
+    ):
+        if pitch is None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super(TextEncoder768, self).__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone: torch.Tensor, pitch: torch.Tensor, lengths: torch.Tensor):
+        if pitch is None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super(ResidualCouplingBlock, self).__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        g: Optional[torch.Tensor] = None,
+        reverse: bool = False,
+    ):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in self.flows[::-1]:
+                x, _ = flow.forward(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for i in range(self.n_flows):
+            for hook in self.flows[i * 2]._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.flows[i * 2])
+
+        return self
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super(PosteriorEncoder, self).__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(
+        self, x: torch.Tensor, x_lengths: torch.Tensor, g: Optional[torch.Tensor] = None
+    ):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.enc._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.enc)
+        return self
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x: torch.Tensor, g: Optional[torch.Tensor] = None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def __prepare_scriptable__(self):
+        for l in self.ups:
+            for hook in l._forward_pre_hooks.values():
+                # The hook we want to remove is an instance of WeightNorm class, so
+                # normally we would do `if isinstance(...)` but this class is not accessible
+                # because of shadowing, so we check the module name directly.
+                # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+
+        for l in self.resblocks:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        return self
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(torch.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        if uv.device.type == "privateuseone":  # for DirectML
+            uv = uv.float()
+        return uv
+
+    def forward(self, f0: torch.Tensor, upp: int):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            device = next(self.parameters(), None)
+            if device is not None:
+                device = device.device
+            else:
+                device = f0.device
+            align_corners = device.type != "xpu"
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in range(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1意味着n_har的乘积无法后处理优化
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1意味着后面的cumsum无法再优化
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=float(upp),
+                mode="linear",
+                align_corners=align_corners,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=float(upp), mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * torch.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=float(upp), mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+        # self.ddtype:int = -1
+
+    def forward(self, x: torch.Tensor, upp: int = 1):
+        # if self.ddtype ==-1:
+        #     self.ddtype = self.l_linear.weight.dtype
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        # print(x.dtype,sine_wavs.dtype,self.l_linear.weight.dtype)
+        # if self.is_half:
+        #     sine_wavs = sine_wavs.half()
+        # sine_merge = self.l_tanh(self.l_linear(sine_wavs.to(x)))
+        # print(sine_wavs.dtype,self.ddtype)
+        # if sine_wavs.dtype != self.l_linear.weight.dtype:
+        sine_wavs = sine_wavs.to(dtype=self.l_linear.weight.dtype)
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=math.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = math.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = math.prod(upsample_rates)
+
+        self.lrelu_slope = modules.LRELU_SLOPE
+
+    def forward(self, x, f0, g: Optional[torch.Tensor] = None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+        # torch.jit.script() does not support direct indexing of torch modules
+        # That's why I wrote this
+        for i, (ups, noise_convs) in enumerate(zip(self.ups, self.noise_convs)):
+            if i < self.num_upsamples:
+                x = F.leaky_relu(x, self.lrelu_slope)
+                x = ups(x)
+                x_source = noise_convs(har_source)
+                x = x + x_source
+                xs: Optional[torch.Tensor] = None
+                l = [i * self.num_kernels + j for j in range(self.num_kernels)]
+                for j, resblock in enumerate(self.resblocks):
+                    if j in l:
+                        if xs is None:
+                            xs = resblock(x)
+                        else:
+                            xs += resblock(x)
+                # This assertion cannot be ignored! \
+                # If ignored, it will cause torch.jit.script() compilation errors
+                assert isinstance(xs, torch.Tensor)
+                x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for l in self.ups:
+            for hook in l._forward_pre_hooks.values():
+                # The hook we want to remove is an instance of WeightNorm class, so
+                # normally we would do `if isinstance(...)` but this class is not accessible
+                # because of shadowing, so we check the module name directly.
+                # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        for l in self.resblocks:
+            for hook in self.resblocks._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        return self
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMs256NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super(SynthesizerTrnMs256NSFsid, self).__init__()
+        if isinstance(sr, str):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        logger.debug(
+            "gin_channels: "
+            + str(gin_channels)
+            + ", self.spk_embed_dim: "
+            + str(self.spk_embed_dim)
+        )
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.dec._forward_pre_hooks.values():
+            # The hook we want to remove is an instance of WeightNorm class, so
+            # normally we would do `if isinstance(...)` but this class is not accessible
+            # because of shadowing, so we check the module name directly.
+            # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.dec)
+        for hook in self.flow._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.flow)
+        if hasattr(self, "enc_q"):
+            for hook in self.enc_q._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.enc_q)
+        return self
+
+    @torch.jit.ignore
+    def forward(
+        self,
+        phone: torch.Tensor,
+        phone_lengths: torch.Tensor,
+        pitch: torch.Tensor,
+        pitchf: torch.Tensor,
+        y: torch.Tensor,
+        y_lengths: torch.Tensor,
+        ds: Optional[torch.Tensor] = None,
+    ):  # 这里ds是id，[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    @torch.jit.export
+    def infer(
+        self,
+        phone: torch.Tensor,
+        phone_lengths: torch.Tensor,
+        pitch: torch.Tensor,
+        nsff0: torch.Tensor,
+        sid: torch.Tensor,
+        rate: Optional[torch.Tensor] = None,
+    ):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate is not None:
+            assert isinstance(rate, torch.Tensor)
+            head = int(z_p.shape[2] * (1 - rate.item()))
+            z_p = z_p[:, :, head:]
+            x_mask = x_mask[:, :, head:]
+            nsff0 = nsff0[:, head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        **kwargs
+    ):
+        super(SynthesizerTrnMs768NSFsid, self).__init__()
+        if isinstance(sr, str):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+        )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        logger.debug(
+            "gin_channels: "
+            + str(gin_channels)
+            + ", self.spk_embed_dim: "
+            + str(self.spk_embed_dim)
+        )
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.dec._forward_pre_hooks.values():
+            # The hook we want to remove is an instance of WeightNorm class, so
+            # normally we would do `if isinstance(...)` but this class is not accessible
+            # because of shadowing, so we check the module name directly.
+            # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.dec)
+        for hook in self.flow._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.flow)
+        if hasattr(self, "enc_q"):
+            for hook in self.enc_q._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.enc_q)
+        return self
+
+    @torch.jit.ignore
+    def forward(
+        self, phone, phone_lengths, pitch, pitchf, y, y_lengths, ds
+    ):  # 这里ds是id，[bs,1]
+        # print(1,pitch.shape)#[bs,t]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        # print(-1,pitchf.shape,ids_slice,self.segment_size,self.hop_length,self.segment_size//self.hop_length)
+        pitchf = commons.slice_segments2(pitchf, ids_slice, self.segment_size)
+        # print(-2,pitchf.shape,z_slice.shape)
+        o = self.dec(z_slice, pitchf, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    @torch.jit.export
+    def infer(
+        self,
+        phone: torch.Tensor,
+        phone_lengths: torch.Tensor,
+        pitch: torch.Tensor,
+        nsff0: torch.Tensor,
+        sid: torch.Tensor,
+        rate: Optional[torch.Tensor] = None,
+    ):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate is not None:
+            head = int(z_p.shape[2] * (1.0 - rate.item()))
+            z_p = z_p[:, :, head:]
+            x_mask = x_mask[:, :, head:]
+            nsff0 = nsff0[:, head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, nsff0, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs256NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super(SynthesizerTrnMs256NSFsid_nono, self).__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder256(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        logger.debug(
+            "gin_channels: "
+            + str(gin_channels)
+            + ", self.spk_embed_dim: "
+            + str(self.spk_embed_dim)
+        )
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.dec._forward_pre_hooks.values():
+            # The hook we want to remove is an instance of WeightNorm class, so
+            # normally we would do `if isinstance(...)` but this class is not accessible
+            # because of shadowing, so we check the module name directly.
+            # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.dec)
+        for hook in self.flow._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.flow)
+        if hasattr(self, "enc_q"):
+            for hook in self.enc_q._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.enc_q)
+        return self
+
+    @torch.jit.ignore
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # 这里ds是id，[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    @torch.jit.export
+    def infer(
+        self,
+        phone: torch.Tensor,
+        phone_lengths: torch.Tensor,
+        sid: torch.Tensor,
+        rate: Optional[torch.Tensor] = None,
+    ):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate is not None:
+            head = int(z_p.shape[2] * (1.0 - rate.item()))
+            z_p = z_p[:, :, head:]
+            x_mask = x_mask[:, :, head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class SynthesizerTrnMs768NSFsid_nono(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr=None,
+        **kwargs
+    ):
+        super(SynthesizerTrnMs768NSFsid_nono, self).__init__()
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = float(p_dropout)
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        self.enc_p = TextEncoder768(
+            inter_channels,
+            hidden_channels,
+            filter_channels,
+            n_heads,
+            n_layers,
+            kernel_size,
+            float(p_dropout),
+            f0=False,
+        )
+        self.dec = Generator(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        logger.debug(
+            "gin_channels: "
+            + str(gin_channels)
+            + ", self.spk_embed_dim: "
+            + str(self.spk_embed_dim)
+        )
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.dec._forward_pre_hooks.values():
+            # The hook we want to remove is an instance of WeightNorm class, so
+            # normally we would do `if isinstance(...)` but this class is not accessible
+            # because of shadowing, so we check the module name directly.
+            # https://github.com/pytorch/pytorch/blob/be0ca00c5ce260eb5bcec3237357f7a30cc08983/torch/nn/utils/__init__.py#L3
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.dec)
+        for hook in self.flow._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.flow)
+        if hasattr(self, "enc_q"):
+            for hook in self.enc_q._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.enc_q)
+        return self
+
+    @torch.jit.ignore
+    def forward(self, phone, phone_lengths, y, y_lengths, ds):  # 这里ds是id，[bs,1]
+        g = self.emb_g(ds).unsqueeze(-1)  # [b, 256, 1]##1是t，广播的
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z, m_q, logs_q, y_mask = self.enc_q(y, y_lengths, g=g)
+        z_p = self.flow(z, y_mask, g=g)
+        z_slice, ids_slice = commons.rand_slice_segments(
+            z, y_lengths, self.segment_size
+        )
+        o = self.dec(z_slice, g=g)
+        return o, ids_slice, x_mask, y_mask, (z, z_p, m_p, logs_p, m_q, logs_q)
+
+    @torch.jit.export
+    def infer(
+        self,
+        phone: torch.Tensor,
+        phone_lengths: torch.Tensor,
+        sid: torch.Tensor,
+        rate: Optional[torch.Tensor] = None,
+    ):
+        g = self.emb_g(sid).unsqueeze(-1)
+        m_p, logs_p, x_mask = self.enc_p(phone, None, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * torch.randn_like(m_p) * 0.66666) * x_mask
+        if rate is not None:
+            head = int(z_p.shape[2] * (1.0 - rate.item()))
+            z_p = z_p[:, :, head:]
+            x_mask = x_mask[:, :, head:]
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec(z * x_mask, g=g)
+        return o, x_mask, (z, z_p, m_p, logs_p)
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            if has_xpu and x.dtype == torch.bfloat16:
+                x = F.pad(x.to(dtype=torch.float16), (0, n_pad), "reflect").to(
+                    dtype=torch.bfloat16
+                )
+            else:
+                x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

rvc/lib/infer_pack/models_onnx.py

@@ -0,0 +1,821 @@
+import logging
+import math
+
+logger = logging.getLogger(__name__)
+
+import numpy as np
+import torch
+from torch import nn
+from torch.nn import AvgPool1d, Conv1d, Conv2d, ConvTranspose1d
+from torch.nn import functional as F
+from torch.nn.utils import remove_weight_norm, spectral_norm, weight_norm
+
+from rvc.lib.infer_pack import attentions, commons, modules
+from rvc.lib.infer_pack.commons import get_padding, init_weights
+
+
+class TextEncoder256(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(256, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class TextEncoder768(nn.Module):
+    def __init__(
+        self,
+        out_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        f0=True,
+    ):
+        super().__init__()
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.emb_phone = nn.Linear(768, hidden_channels)
+        self.lrelu = nn.LeakyReLU(0.1, inplace=True)
+        if f0 == True:
+            self.emb_pitch = nn.Embedding(256, hidden_channels)  # pitch 256
+        self.encoder = attentions.Encoder(
+            hidden_channels, filter_channels, n_heads, n_layers, kernel_size, p_dropout
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, phone, pitch, lengths):
+        if pitch == None:
+            x = self.emb_phone(phone)
+        else:
+            x = self.emb_phone(phone) + self.emb_pitch(pitch)
+        x = x * math.sqrt(self.hidden_channels)  # [b, t, h]
+        x = self.lrelu(x)
+        x = torch.transpose(x, 1, -1)  # [b, h, t]
+        x_mask = torch.unsqueeze(commons.sequence_mask(lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.encoder(x * x_mask, x_mask)
+        stats = self.proj(x) * x_mask
+
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        return m, logs, x_mask
+
+
+class ResidualCouplingBlock(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        n_flows=4,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.n_flows = n_flows
+        self.gin_channels = gin_channels
+
+        self.flows = nn.ModuleList()
+        for i in range(n_flows):
+            self.flows.append(
+                modules.ResidualCouplingLayer(
+                    channels,
+                    hidden_channels,
+                    kernel_size,
+                    dilation_rate,
+                    n_layers,
+                    gin_channels=gin_channels,
+                    mean_only=True,
+                )
+            )
+            self.flows.append(modules.Flip())
+
+    def forward(self, x, x_mask, g=None, reverse=False):
+        if not reverse:
+            for flow in self.flows:
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        else:
+            for flow in reversed(self.flows):
+                x, _ = flow(x, x_mask, g=g, reverse=reverse)
+        return x
+
+    def remove_weight_norm(self):
+        for i in range(self.n_flows):
+            self.flows[i * 2].remove_weight_norm()
+
+
+class PosteriorEncoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        out_channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+    ):
+        super().__init__()
+        self.in_channels = in_channels
+        self.out_channels = out_channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+
+        self.pre = nn.Conv1d(in_channels, hidden_channels, 1)
+        self.enc = modules.WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            gin_channels=gin_channels,
+        )
+        self.proj = nn.Conv1d(hidden_channels, out_channels * 2, 1)
+
+    def forward(self, x, x_lengths, g=None):
+        x_mask = torch.unsqueeze(commons.sequence_mask(x_lengths, x.size(2)), 1).to(
+            x.dtype
+        )
+        x = self.pre(x) * x_mask
+        x = self.enc(x, x_mask, g=g)
+        stats = self.proj(x) * x_mask
+        m, logs = torch.split(stats, self.out_channels, dim=1)
+        z = (m + torch.randn_like(m) * torch.exp(logs)) * x_mask
+        return z, m, logs, x_mask
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+
+class Generator(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels=0,
+    ):
+        super(Generator, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+    def forward(self, x, g=None):
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+class SineGen(torch.nn.Module):
+    """Definition of sine generator
+    SineGen(samp_rate, harmonic_num = 0,
+            sine_amp = 0.1, noise_std = 0.003,
+            voiced_threshold = 0,
+            flag_for_pulse=False)
+    samp_rate: sampling rate in Hz
+    harmonic_num: number of harmonic overtones (default 0)
+    sine_amp: amplitude of sine-wavefrom (default 0.1)
+    noise_std: std of Gaussian noise (default 0.003)
+    voiced_thoreshold: F0 threshold for U/V classification (default 0)
+    flag_for_pulse: this SinGen is used inside PulseGen (default False)
+    Note: when flag_for_pulse is True, the first time step of a voiced
+        segment is always sin(np.pi) or cos(0)
+    """
+
+    def __init__(
+        self,
+        samp_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        noise_std=0.003,
+        voiced_threshold=0,
+        flag_for_pulse=False,
+    ):
+        super(SineGen, self).__init__()
+        self.sine_amp = sine_amp
+        self.noise_std = noise_std
+        self.harmonic_num = harmonic_num
+        self.dim = self.harmonic_num + 1
+        self.sampling_rate = samp_rate
+        self.voiced_threshold = voiced_threshold
+
+    def _f02uv(self, f0):
+        # generate uv signal
+        uv = torch.ones_like(f0)
+        uv = uv * (f0 > self.voiced_threshold)
+        return uv
+
+    def forward(self, f0, upp):
+        """sine_tensor, uv = forward(f0)
+        input F0: tensor(batchsize=1, length, dim=1)
+                  f0 for unvoiced steps should be 0
+        output sine_tensor: tensor(batchsize=1, length, dim)
+        output uv: tensor(batchsize=1, length, 1)
+        """
+        with torch.no_grad():
+            f0 = f0[:, None].transpose(1, 2)
+            f0_buf = torch.zeros(f0.shape[0], f0.shape[1], self.dim, device=f0.device)
+            # fundamental component
+            f0_buf[:, :, 0] = f0[:, :, 0]
+            for idx in np.arange(self.harmonic_num):
+                f0_buf[:, :, idx + 1] = f0_buf[:, :, 0] * (
+                    idx + 2
+                )  # idx + 2: the (idx+1)-th overtone, (idx+2)-th harmonic
+            rad_values = (f0_buf / self.sampling_rate) % 1  ###%1意味着n_har的乘积无法后处理优化
+            rand_ini = torch.rand(
+                f0_buf.shape[0], f0_buf.shape[2], device=f0_buf.device
+            )
+            rand_ini[:, 0] = 0
+            rad_values[:, 0, :] = rad_values[:, 0, :] + rand_ini
+            tmp_over_one = torch.cumsum(rad_values, 1)  # % 1  #####%1意味着后面的cumsum无法再优化
+            tmp_over_one *= upp
+            tmp_over_one = F.interpolate(
+                tmp_over_one.transpose(2, 1),
+                scale_factor=upp,
+                mode="linear",
+                align_corners=True,
+            ).transpose(2, 1)
+            rad_values = F.interpolate(
+                rad_values.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(
+                2, 1
+            )  #######
+            tmp_over_one %= 1
+            tmp_over_one_idx = (tmp_over_one[:, 1:, :] - tmp_over_one[:, :-1, :]) < 0
+            cumsum_shift = torch.zeros_like(rad_values)
+            cumsum_shift[:, 1:, :] = tmp_over_one_idx * -1.0
+            sine_waves = torch.sin(
+                torch.cumsum(rad_values + cumsum_shift, dim=1) * 2 * np.pi
+            )
+            sine_waves = sine_waves * self.sine_amp
+            uv = self._f02uv(f0)
+            uv = F.interpolate(
+                uv.transpose(2, 1), scale_factor=upp, mode="nearest"
+            ).transpose(2, 1)
+            noise_amp = uv * self.noise_std + (1 - uv) * self.sine_amp / 3
+            noise = noise_amp * torch.randn_like(sine_waves)
+            sine_waves = sine_waves * uv + noise
+        return sine_waves, uv, noise
+
+
+class SourceModuleHnNSF(torch.nn.Module):
+    """SourceModule for hn-nsf
+    SourceModule(sampling_rate, harmonic_num=0, sine_amp=0.1,
+                 add_noise_std=0.003, voiced_threshod=0)
+    sampling_rate: sampling_rate in Hz
+    harmonic_num: number of harmonic above F0 (default: 0)
+    sine_amp: amplitude of sine source signal (default: 0.1)
+    add_noise_std: std of additive Gaussian noise (default: 0.003)
+        note that amplitude of noise in unvoiced is decided
+        by sine_amp
+    voiced_threshold: threhold to set U/V given F0 (default: 0)
+    Sine_source, noise_source = SourceModuleHnNSF(F0_sampled)
+    F0_sampled (batchsize, length, 1)
+    Sine_source (batchsize, length, 1)
+    noise_source (batchsize, length 1)
+    uv (batchsize, length, 1)
+    """
+
+    def __init__(
+        self,
+        sampling_rate,
+        harmonic_num=0,
+        sine_amp=0.1,
+        add_noise_std=0.003,
+        voiced_threshod=0,
+        is_half=True,
+    ):
+        super(SourceModuleHnNSF, self).__init__()
+
+        self.sine_amp = sine_amp
+        self.noise_std = add_noise_std
+        self.is_half = is_half
+        # to produce sine waveforms
+        self.l_sin_gen = SineGen(
+            sampling_rate, harmonic_num, sine_amp, add_noise_std, voiced_threshod
+        )
+
+        # to merge source harmonics into a single excitation
+        self.l_linear = torch.nn.Linear(harmonic_num + 1, 1)
+        self.l_tanh = torch.nn.Tanh()
+
+    def forward(self, x, upp=None):
+        sine_wavs, uv, _ = self.l_sin_gen(x, upp)
+        if self.is_half:
+            sine_wavs = sine_wavs.half()
+        sine_merge = self.l_tanh(self.l_linear(sine_wavs))
+        return sine_merge, None, None  # noise, uv
+
+
+class GeneratorNSF(torch.nn.Module):
+    def __init__(
+        self,
+        initial_channel,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        gin_channels,
+        sr,
+        is_half=False,
+    ):
+        super(GeneratorNSF, self).__init__()
+        self.num_kernels = len(resblock_kernel_sizes)
+        self.num_upsamples = len(upsample_rates)
+
+        self.f0_upsamp = torch.nn.Upsample(scale_factor=np.prod(upsample_rates))
+        self.m_source = SourceModuleHnNSF(
+            sampling_rate=sr, harmonic_num=0, is_half=is_half
+        )
+        self.noise_convs = nn.ModuleList()
+        self.conv_pre = Conv1d(
+            initial_channel, upsample_initial_channel, 7, 1, padding=3
+        )
+        resblock = modules.ResBlock1 if resblock == "1" else modules.ResBlock2
+
+        self.ups = nn.ModuleList()
+        for i, (u, k) in enumerate(zip(upsample_rates, upsample_kernel_sizes)):
+            c_cur = upsample_initial_channel // (2 ** (i + 1))
+            self.ups.append(
+                weight_norm(
+                    ConvTranspose1d(
+                        upsample_initial_channel // (2**i),
+                        upsample_initial_channel // (2 ** (i + 1)),
+                        k,
+                        u,
+                        padding=(k - u) // 2,
+                    )
+                )
+            )
+            if i + 1 < len(upsample_rates):
+                stride_f0 = np.prod(upsample_rates[i + 1 :])
+                self.noise_convs.append(
+                    Conv1d(
+                        1,
+                        c_cur,
+                        kernel_size=stride_f0 * 2,
+                        stride=stride_f0,
+                        padding=stride_f0 // 2,
+                    )
+                )
+            else:
+                self.noise_convs.append(Conv1d(1, c_cur, kernel_size=1))
+
+        self.resblocks = nn.ModuleList()
+        for i in range(len(self.ups)):
+            ch = upsample_initial_channel // (2 ** (i + 1))
+            for j, (k, d) in enumerate(
+                zip(resblock_kernel_sizes, resblock_dilation_sizes)
+            ):
+                self.resblocks.append(resblock(ch, k, d))
+
+        self.conv_post = Conv1d(ch, 1, 7, 1, padding=3, bias=False)
+        self.ups.apply(init_weights)
+
+        if gin_channels != 0:
+            self.cond = nn.Conv1d(gin_channels, upsample_initial_channel, 1)
+
+        self.upp = np.prod(upsample_rates)
+
+    def forward(self, x, f0, g=None):
+        har_source, noi_source, uv = self.m_source(f0, self.upp)
+        har_source = har_source.transpose(1, 2)
+        x = self.conv_pre(x)
+        if g is not None:
+            x = x + self.cond(g)
+
+        for i in range(self.num_upsamples):
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            x = self.ups[i](x)
+            x_source = self.noise_convs[i](har_source)
+            x = x + x_source
+            xs = None
+            for j in range(self.num_kernels):
+                if xs is None:
+                    xs = self.resblocks[i * self.num_kernels + j](x)
+                else:
+                    xs += self.resblocks[i * self.num_kernels + j](x)
+            x = xs / self.num_kernels
+        x = F.leaky_relu(x)
+        x = self.conv_post(x)
+        x = torch.tanh(x)
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.ups:
+            remove_weight_norm(l)
+        for l in self.resblocks:
+            l.remove_weight_norm()
+
+
+sr2sr = {
+    "32k": 32000,
+    "40k": 40000,
+    "48k": 48000,
+}
+
+
+class SynthesizerTrnMsNSFsidM(nn.Module):
+    def __init__(
+        self,
+        spec_channels,
+        segment_size,
+        inter_channels,
+        hidden_channels,
+        filter_channels,
+        n_heads,
+        n_layers,
+        kernel_size,
+        p_dropout,
+        resblock,
+        resblock_kernel_sizes,
+        resblock_dilation_sizes,
+        upsample_rates,
+        upsample_initial_channel,
+        upsample_kernel_sizes,
+        spk_embed_dim,
+        gin_channels,
+        sr,
+        version,
+        **kwargs,
+    ):
+        super().__init__()
+        if type(sr) == type("strr"):
+            sr = sr2sr[sr]
+        self.spec_channels = spec_channels
+        self.inter_channels = inter_channels
+        self.hidden_channels = hidden_channels
+        self.filter_channels = filter_channels
+        self.n_heads = n_heads
+        self.n_layers = n_layers
+        self.kernel_size = kernel_size
+        self.p_dropout = p_dropout
+        self.resblock = resblock
+        self.resblock_kernel_sizes = resblock_kernel_sizes
+        self.resblock_dilation_sizes = resblock_dilation_sizes
+        self.upsample_rates = upsample_rates
+        self.upsample_initial_channel = upsample_initial_channel
+        self.upsample_kernel_sizes = upsample_kernel_sizes
+        self.segment_size = segment_size
+        self.gin_channels = gin_channels
+        # self.hop_length = hop_length#
+        self.spk_embed_dim = spk_embed_dim
+        if version == "v1":
+            self.enc_p = TextEncoder256(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        else:
+            self.enc_p = TextEncoder768(
+                inter_channels,
+                hidden_channels,
+                filter_channels,
+                n_heads,
+                n_layers,
+                kernel_size,
+                p_dropout,
+            )
+        self.dec = GeneratorNSF(
+            inter_channels,
+            resblock,
+            resblock_kernel_sizes,
+            resblock_dilation_sizes,
+            upsample_rates,
+            upsample_initial_channel,
+            upsample_kernel_sizes,
+            gin_channels=gin_channels,
+            sr=sr,
+            is_half=kwargs["is_half"],
+        )
+        self.enc_q = PosteriorEncoder(
+            spec_channels,
+            inter_channels,
+            hidden_channels,
+            5,
+            1,
+            16,
+            gin_channels=gin_channels,
+        )
+        self.flow = ResidualCouplingBlock(
+            inter_channels, hidden_channels, 5, 1, 3, gin_channels=gin_channels
+        )
+        self.emb_g = nn.Embedding(self.spk_embed_dim, gin_channels)
+        self.speaker_map = None
+        logger.debug(
+            f"gin_channels: {gin_channels}, self.spk_embed_dim: {self.spk_embed_dim}"
+        )
+
+    def remove_weight_norm(self):
+        self.dec.remove_weight_norm()
+        self.flow.remove_weight_norm()
+        self.enc_q.remove_weight_norm()
+
+    def construct_spkmixmap(self, n_speaker):
+        self.speaker_map = torch.zeros((n_speaker, 1, 1, self.gin_channels))
+        for i in range(n_speaker):
+            self.speaker_map[i] = self.emb_g(torch.LongTensor([[i]]))
+        self.speaker_map = self.speaker_map.unsqueeze(0)
+
+    def forward(self, phone, phone_lengths, pitch, nsff0, g, rnd, max_len=None):
+        if self.speaker_map is not None:  # [N, S]  *  [S, B, 1, H]
+            g = g.reshape((g.shape[0], g.shape[1], 1, 1, 1))  # [N, S, B, 1, 1]
+            g = g * self.speaker_map  # [N, S, B, 1, H]
+            g = torch.sum(g, dim=1)  # [N, 1, B, 1, H]
+            g = g.transpose(0, -1).transpose(0, -2).squeeze(0)  # [B, H, N]
+        else:
+            g = g.unsqueeze(0)
+            g = self.emb_g(g).transpose(1, 2)
+
+        m_p, logs_p, x_mask = self.enc_p(phone, pitch, phone_lengths)
+        z_p = (m_p + torch.exp(logs_p) * rnd) * x_mask
+        z = self.flow(z_p, x_mask, g=g, reverse=True)
+        o = self.dec((z * x_mask)[:, :, :max_len], nsff0, g=g)
+        return o
+
+
+class MultiPeriodDiscriminator(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminator, self).__init__()
+        periods = [2, 3, 5, 7, 11, 17]
+        # periods = [3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class MultiPeriodDiscriminatorV2(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(MultiPeriodDiscriminatorV2, self).__init__()
+        # periods = [2, 3, 5, 7, 11, 17]
+        periods = [2, 3, 5, 7, 11, 17, 23, 37]
+
+        discs = [DiscriminatorS(use_spectral_norm=use_spectral_norm)]
+        discs = discs + [
+            DiscriminatorP(i, use_spectral_norm=use_spectral_norm) for i in periods
+        ]
+        self.discriminators = nn.ModuleList(discs)
+
+    def forward(self, y, y_hat):
+        y_d_rs = []  #
+        y_d_gs = []
+        fmap_rs = []
+        fmap_gs = []
+        for i, d in enumerate(self.discriminators):
+            y_d_r, fmap_r = d(y)
+            y_d_g, fmap_g = d(y_hat)
+            # for j in range(len(fmap_r)):
+            #     print(i,j,y.shape,y_hat.shape,fmap_r[j].shape,fmap_g[j].shape)
+            y_d_rs.append(y_d_r)
+            y_d_gs.append(y_d_g)
+            fmap_rs.append(fmap_r)
+            fmap_gs.append(fmap_g)
+
+        return y_d_rs, y_d_gs, fmap_rs, fmap_gs
+
+
+class DiscriminatorS(torch.nn.Module):
+    def __init__(self, use_spectral_norm=False):
+        super(DiscriminatorS, self).__init__()
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(Conv1d(1, 16, 15, 1, padding=7)),
+                norm_f(Conv1d(16, 64, 41, 4, groups=4, padding=20)),
+                norm_f(Conv1d(64, 256, 41, 4, groups=16, padding=20)),
+                norm_f(Conv1d(256, 1024, 41, 4, groups=64, padding=20)),
+                norm_f(Conv1d(1024, 1024, 41, 4, groups=256, padding=20)),
+                norm_f(Conv1d(1024, 1024, 5, 1, padding=2)),
+            ]
+        )
+        self.conv_post = norm_f(Conv1d(1024, 1, 3, 1, padding=1))
+
+    def forward(self, x):
+        fmap = []
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap
+
+
+class DiscriminatorP(torch.nn.Module):
+    def __init__(self, period, kernel_size=5, stride=3, use_spectral_norm=False):
+        super(DiscriminatorP, self).__init__()
+        self.period = period
+        self.use_spectral_norm = use_spectral_norm
+        norm_f = weight_norm if use_spectral_norm == False else spectral_norm
+        self.convs = nn.ModuleList(
+            [
+                norm_f(
+                    Conv2d(
+                        1,
+                        32,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        32,
+                        128,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        128,
+                        512,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        512,
+                        1024,
+                        (kernel_size, 1),
+                        (stride, 1),
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+                norm_f(
+                    Conv2d(
+                        1024,
+                        1024,
+                        (kernel_size, 1),
+                        1,
+                        padding=(get_padding(kernel_size, 1), 0),
+                    )
+                ),
+            ]
+        )
+        self.conv_post = norm_f(Conv2d(1024, 1, (3, 1), 1, padding=(1, 0)))
+
+    def forward(self, x):
+        fmap = []
+
+        # 1d to 2d
+        b, c, t = x.shape
+        if t % self.period != 0:  # pad first
+            n_pad = self.period - (t % self.period)
+            x = F.pad(x, (0, n_pad), "reflect")
+            t = t + n_pad
+        x = x.view(b, c, t // self.period, self.period)
+
+        for l in self.convs:
+            x = l(x)
+            x = F.leaky_relu(x, modules.LRELU_SLOPE)
+            fmap.append(x)
+        x = self.conv_post(x)
+        fmap.append(x)
+        x = torch.flatten(x, 1, -1)
+
+        return x, fmap

rvc/lib/infer_pack/modules.py

@@ -0,0 +1,615 @@
+import copy
+import math
+from typing import Optional, Tuple
+
+import numpy as np
+import scipy
+import torch
+from torch import nn
+from torch.nn import AvgPool1d, Conv1d, Conv2d, ConvTranspose1d
+from torch.nn import functional as F
+from torch.nn.utils import remove_weight_norm, weight_norm
+
+from rvc.lib.infer_pack import commons
+from rvc.lib.infer_pack.commons import get_padding, init_weights
+from rvc.lib.infer_pack.transforms import piecewise_rational_quadratic_transform
+
+LRELU_SLOPE = 0.1
+
+
+class LayerNorm(nn.Module):
+    def __init__(self, channels, eps=1e-5):
+        super(LayerNorm, self).__init__()
+        self.channels = channels
+        self.eps = eps
+
+        self.gamma = nn.Parameter(torch.ones(channels))
+        self.beta = nn.Parameter(torch.zeros(channels))
+
+    def forward(self, x):
+        x = x.transpose(1, -1)
+        x = F.layer_norm(x, (self.channels,), self.gamma, self.beta, self.eps)
+        return x.transpose(1, -1)
+
+
+class ConvReluNorm(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        hidden_channels,
+        out_channels,
+        kernel_size,
+        n_layers,
+        p_dropout,
+    ):
+        super(ConvReluNorm, self).__init__()
+        self.in_channels = in_channels
+        self.hidden_channels = hidden_channels
+        self.out_channels = out_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = float(p_dropout)
+        assert n_layers > 1, "Number of layers should be larger than 0."
+
+        self.conv_layers = nn.ModuleList()
+        self.norm_layers = nn.ModuleList()
+        self.conv_layers.append(
+            nn.Conv1d(
+                in_channels, hidden_channels, kernel_size, padding=kernel_size // 2
+            )
+        )
+        self.norm_layers.append(LayerNorm(hidden_channels))
+        self.relu_drop = nn.Sequential(nn.ReLU(), nn.Dropout(float(p_dropout)))
+        for _ in range(n_layers - 1):
+            self.conv_layers.append(
+                nn.Conv1d(
+                    hidden_channels,
+                    hidden_channels,
+                    kernel_size,
+                    padding=kernel_size // 2,
+                )
+            )
+            self.norm_layers.append(LayerNorm(hidden_channels))
+        self.proj = nn.Conv1d(hidden_channels, out_channels, 1)
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(self, x, x_mask):
+        x_org = x
+        for i in range(self.n_layers):
+            x = self.conv_layers[i](x * x_mask)
+            x = self.norm_layers[i](x)
+            x = self.relu_drop(x)
+        x = x_org + self.proj(x)
+        return x * x_mask
+
+
+class DDSConv(nn.Module):
+    """
+    Dialted and Depth-Separable Convolution
+    """
+
+    def __init__(self, channels, kernel_size, n_layers, p_dropout=0.0):
+        super(DDSConv, self).__init__()
+        self.channels = channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.p_dropout = float(p_dropout)
+
+        self.drop = nn.Dropout(float(p_dropout))
+        self.convs_sep = nn.ModuleList()
+        self.convs_1x1 = nn.ModuleList()
+        self.norms_1 = nn.ModuleList()
+        self.norms_2 = nn.ModuleList()
+        for i in range(n_layers):
+            dilation = kernel_size**i
+            padding = (kernel_size * dilation - dilation) // 2
+            self.convs_sep.append(
+                nn.Conv1d(
+                    channels,
+                    channels,
+                    kernel_size,
+                    groups=channels,
+                    dilation=dilation,
+                    padding=padding,
+                )
+            )
+            self.convs_1x1.append(nn.Conv1d(channels, channels, 1))
+            self.norms_1.append(LayerNorm(channels))
+            self.norms_2.append(LayerNorm(channels))
+
+    def forward(self, x, x_mask, g: Optional[torch.Tensor] = None):
+        if g is not None:
+            x = x + g
+        for i in range(self.n_layers):
+            y = self.convs_sep[i](x * x_mask)
+            y = self.norms_1[i](y)
+            y = F.gelu(y)
+            y = self.convs_1x1[i](y)
+            y = self.norms_2[i](y)
+            y = F.gelu(y)
+            y = self.drop(y)
+            x = x + y
+        return x * x_mask
+
+
+class WN(torch.nn.Module):
+    def __init__(
+        self,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        gin_channels=0,
+        p_dropout=0,
+    ):
+        super(WN, self).__init__()
+        assert kernel_size % 2 == 1
+        self.hidden_channels = hidden_channels
+        self.kernel_size = (kernel_size,)
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.gin_channels = gin_channels
+        self.p_dropout = float(p_dropout)
+
+        self.in_layers = torch.nn.ModuleList()
+        self.res_skip_layers = torch.nn.ModuleList()
+        self.drop = nn.Dropout(float(p_dropout))
+
+        if gin_channels != 0:
+            cond_layer = torch.nn.Conv1d(
+                gin_channels, 2 * hidden_channels * n_layers, 1
+            )
+            self.cond_layer = torch.nn.utils.weight_norm(cond_layer, name="weight")
+
+        for i in range(n_layers):
+            dilation = dilation_rate**i
+            padding = int((kernel_size * dilation - dilation) / 2)
+            in_layer = torch.nn.Conv1d(
+                hidden_channels,
+                2 * hidden_channels,
+                kernel_size,
+                dilation=dilation,
+                padding=padding,
+            )
+            in_layer = torch.nn.utils.weight_norm(in_layer, name="weight")
+            self.in_layers.append(in_layer)
+
+            # last one is not necessary
+            if i < n_layers - 1:
+                res_skip_channels = 2 * hidden_channels
+            else:
+                res_skip_channels = hidden_channels
+
+            res_skip_layer = torch.nn.Conv1d(hidden_channels, res_skip_channels, 1)
+            res_skip_layer = torch.nn.utils.weight_norm(res_skip_layer, name="weight")
+            self.res_skip_layers.append(res_skip_layer)
+
+    def forward(
+        self, x: torch.Tensor, x_mask: torch.Tensor, g: Optional[torch.Tensor] = None
+    ):
+        output = torch.zeros_like(x)
+        n_channels_tensor = torch.IntTensor([self.hidden_channels])
+
+        if g is not None:
+            g = self.cond_layer(g)
+
+        for i, (in_layer, res_skip_layer) in enumerate(
+            zip(self.in_layers, self.res_skip_layers)
+        ):
+            x_in = in_layer(x)
+            if g is not None:
+                cond_offset = i * 2 * self.hidden_channels
+                g_l = g[:, cond_offset : cond_offset + 2 * self.hidden_channels, :]
+            else:
+                g_l = torch.zeros_like(x_in)
+
+            acts = commons.fused_add_tanh_sigmoid_multiply(x_in, g_l, n_channels_tensor)
+            acts = self.drop(acts)
+
+            res_skip_acts = res_skip_layer(acts)
+            if i < self.n_layers - 1:
+                res_acts = res_skip_acts[:, : self.hidden_channels, :]
+                x = (x + res_acts) * x_mask
+                output = output + res_skip_acts[:, self.hidden_channels :, :]
+            else:
+                output = output + res_skip_acts
+        return output * x_mask
+
+    def remove_weight_norm(self):
+        if self.gin_channels != 0:
+            torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            torch.nn.utils.remove_weight_norm(l)
+
+    def __prepare_scriptable__(self):
+        if self.gin_channels != 0:
+            for hook in self.cond_layer._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(self.cond_layer)
+        for l in self.in_layers:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        for l in self.res_skip_layers:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        return self
+
+
+class ResBlock1(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3, 5)):
+        super(ResBlock1, self).__init__()
+        self.convs1 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[2],
+                        padding=get_padding(kernel_size, dilation[2]),
+                    )
+                ),
+            ]
+        )
+        self.convs1.apply(init_weights)
+
+        self.convs2 = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=1,
+                        padding=get_padding(kernel_size, 1),
+                    )
+                ),
+            ]
+        )
+        self.convs2.apply(init_weights)
+        self.lrelu_slope = LRELU_SLOPE
+
+    def forward(self, x: torch.Tensor, x_mask: Optional[torch.Tensor] = None):
+        for c1, c2 in zip(self.convs1, self.convs2):
+            xt = F.leaky_relu(x, self.lrelu_slope)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c1(xt)
+            xt = F.leaky_relu(xt, self.lrelu_slope)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c2(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs1:
+            remove_weight_norm(l)
+        for l in self.convs2:
+            remove_weight_norm(l)
+
+    def __prepare_scriptable__(self):
+        for l in self.convs1:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        for l in self.convs2:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        return self
+
+
+class ResBlock2(torch.nn.Module):
+    def __init__(self, channels, kernel_size=3, dilation=(1, 3)):
+        super(ResBlock2, self).__init__()
+        self.convs = nn.ModuleList(
+            [
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[0],
+                        padding=get_padding(kernel_size, dilation[0]),
+                    )
+                ),
+                weight_norm(
+                    Conv1d(
+                        channels,
+                        channels,
+                        kernel_size,
+                        1,
+                        dilation=dilation[1],
+                        padding=get_padding(kernel_size, dilation[1]),
+                    )
+                ),
+            ]
+        )
+        self.convs.apply(init_weights)
+        self.lrelu_slope = LRELU_SLOPE
+
+    def forward(self, x, x_mask: Optional[torch.Tensor] = None):
+        for c in self.convs:
+            xt = F.leaky_relu(x, self.lrelu_slope)
+            if x_mask is not None:
+                xt = xt * x_mask
+            xt = c(xt)
+            x = xt + x
+        if x_mask is not None:
+            x = x * x_mask
+        return x
+
+    def remove_weight_norm(self):
+        for l in self.convs:
+            remove_weight_norm(l)
+
+    def __prepare_scriptable__(self):
+        for l in self.convs:
+            for hook in l._forward_pre_hooks.values():
+                if (
+                    hook.__module__ == "torch.nn.utils.weight_norm"
+                    and hook.__class__.__name__ == "WeightNorm"
+                ):
+                    torch.nn.utils.remove_weight_norm(l)
+        return self
+
+
+class Log(nn.Module):
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        g: Optional[torch.Tensor] = None,
+        reverse: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
+        if not reverse:
+            y = torch.log(torch.clamp_min(x, 1e-5)) * x_mask
+            logdet = torch.sum(-y, [1, 2])
+            return y, logdet
+        else:
+            x = torch.exp(x) * x_mask
+            return x
+
+
+class Flip(nn.Module):
+    # torch.jit.script() Compiled functions \
+    # can't take variable number of arguments or \
+    # use keyword-only arguments with defaults
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        g: Optional[torch.Tensor] = None,
+        reverse: bool = False,
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
+        x = torch.flip(x, [1])
+        if not reverse:
+            logdet = torch.zeros(x.size(0)).to(dtype=x.dtype, device=x.device)
+            return x, logdet
+        else:
+            return x, torch.zeros([1], device=x.device)
+
+
+class ElementwiseAffine(nn.Module):
+    def __init__(self, channels):
+        super(ElementwiseAffine, self).__init__()
+        self.channels = channels
+        self.m = nn.Parameter(torch.zeros(channels, 1))
+        self.logs = nn.Parameter(torch.zeros(channels, 1))
+
+    def forward(self, x, x_mask, reverse=False, **kwargs):
+        if not reverse:
+            y = self.m + torch.exp(self.logs) * x
+            y = y * x_mask
+            logdet = torch.sum(self.logs * x_mask, [1, 2])
+            return y, logdet
+        else:
+            x = (x - self.m) * torch.exp(-self.logs) * x_mask
+            return x
+
+
+class ResidualCouplingLayer(nn.Module):
+    def __init__(
+        self,
+        channels,
+        hidden_channels,
+        kernel_size,
+        dilation_rate,
+        n_layers,
+        p_dropout=0,
+        gin_channels=0,
+        mean_only=False,
+    ):
+        assert channels % 2 == 0, "channels should be divisible by 2"
+        super(ResidualCouplingLayer, self).__init__()
+        self.channels = channels
+        self.hidden_channels = hidden_channels
+        self.kernel_size = kernel_size
+        self.dilation_rate = dilation_rate
+        self.n_layers = n_layers
+        self.half_channels = channels // 2
+        self.mean_only = mean_only
+
+        self.pre = nn.Conv1d(self.half_channels, hidden_channels, 1)
+        self.enc = WN(
+            hidden_channels,
+            kernel_size,
+            dilation_rate,
+            n_layers,
+            p_dropout=float(p_dropout),
+            gin_channels=gin_channels,
+        )
+        self.post = nn.Conv1d(hidden_channels, self.half_channels * (2 - mean_only), 1)
+        self.post.weight.data.zero_()
+        self.post.bias.data.zero_()
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        g: Optional[torch.Tensor] = None,
+        reverse: bool = False,
+    ):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0) * x_mask
+        h = self.enc(h, x_mask, g=g)
+        stats = self.post(h) * x_mask
+        if not self.mean_only:
+            m, logs = torch.split(stats, [self.half_channels] * 2, 1)
+        else:
+            m = stats
+            logs = torch.zeros_like(m)
+
+        if not reverse:
+            x1 = m + x1 * torch.exp(logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            logdet = torch.sum(logs, [1, 2])
+            return x, logdet
+        else:
+            x1 = (x1 - m) * torch.exp(-logs) * x_mask
+            x = torch.cat([x0, x1], 1)
+            return x, torch.zeros([1])
+
+    def remove_weight_norm(self):
+        self.enc.remove_weight_norm()
+
+    def __prepare_scriptable__(self):
+        for hook in self.enc._forward_pre_hooks.values():
+            if (
+                hook.__module__ == "torch.nn.utils.weight_norm"
+                and hook.__class__.__name__ == "WeightNorm"
+            ):
+                torch.nn.utils.remove_weight_norm(self.enc)
+        return self
+
+
+class ConvFlow(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        filter_channels,
+        kernel_size,
+        n_layers,
+        num_bins=10,
+        tail_bound=5.0,
+    ):
+        super(ConvFlow, self).__init__()
+        self.in_channels = in_channels
+        self.filter_channels = filter_channels
+        self.kernel_size = kernel_size
+        self.n_layers = n_layers
+        self.num_bins = num_bins
+        self.tail_bound = tail_bound
+        self.half_channels = in_channels // 2
+
+        self.pre = nn.Conv1d(self.half_channels, filter_channels, 1)
+        self.convs = DDSConv(filter_channels, kernel_size, n_layers, p_dropout=0.0)
+        self.proj = nn.Conv1d(
+            filter_channels, self.half_channels * (num_bins * 3 - 1), 1
+        )
+        self.proj.weight.data.zero_()
+        self.proj.bias.data.zero_()
+
+    def forward(
+        self,
+        x: torch.Tensor,
+        x_mask: torch.Tensor,
+        g: Optional[torch.Tensor] = None,
+        reverse=False,
+    ):
+        x0, x1 = torch.split(x, [self.half_channels] * 2, 1)
+        h = self.pre(x0)
+        h = self.convs(h, x_mask, g=g)
+        h = self.proj(h) * x_mask
+
+        b, c, t = x0.shape
+        h = h.reshape(b, c, -1, t).permute(0, 1, 3, 2)  # [b, cx?, t] -> [b, c, t, ?]
+
+        unnormalized_widths = h[..., : self.num_bins] / math.sqrt(self.filter_channels)
+        unnormalized_heights = h[..., self.num_bins : 2 * self.num_bins] / math.sqrt(
+            self.filter_channels
+        )
+        unnormalized_derivatives = h[..., 2 * self.num_bins :]
+
+        x1, logabsdet = piecewise_rational_quadratic_transform(
+            x1,
+            unnormalized_widths,
+            unnormalized_heights,
+            unnormalized_derivatives,
+            inverse=reverse,
+            tails="linear",
+            tail_bound=self.tail_bound,
+        )
+
+        x = torch.cat([x0, x1], 1) * x_mask
+        logdet = torch.sum(logabsdet * x_mask, [1, 2])
+        if not reverse:
+            return x, logdet
+        else:
+            return x

rvc/lib/infer_pack/modules/F0Predictor/DioF0Predictor.py

@@ -0,0 +1,91 @@
+import numpy as np
+import pyworld
+
+from rvc.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+
+
+class DioF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.dio(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        for index, pitch in enumerate(f0):
+            f0[index] = round(pitch, 1)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

rvc/lib/infer_pack/modules/F0Predictor/F0Predictor.py

@@ -0,0 +1,16 @@
+class F0Predictor(object):
+    def compute_f0(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length]
+        """
+        pass
+
+    def compute_f0_uv(self, wav, p_len):
+        """
+        input: wav:[signal_length]
+               p_len:int
+        output: f0:[signal_length//hop_length],uv:[signal_length//hop_length]
+        """
+        pass

rvc/lib/infer_pack/modules/F0Predictor/HarvestF0Predictor.py

@@ -0,0 +1,87 @@
+import numpy as np
+import pyworld
+
+from rvc.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+
+
+class HarvestF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def resize_f0(self, x, target_len):
+        source = np.array(x)
+        source[source < 0.001] = np.nan
+        target = np.interp(
+            np.arange(0, len(source) * target_len, len(source)) / target_len,
+            np.arange(0, len(source)),
+            source,
+        )
+        res = np.nan_to_num(target)
+        return res
+
+    def compute_f0(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_ceil=self.f0_max,
+            f0_floor=self.f0_min,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.fs)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))[0]
+
+    def compute_f0_uv(self, wav, p_len=None):
+        if p_len is None:
+            p_len = wav.shape[0] // self.hop_length
+        f0, t = pyworld.harvest(
+            wav.astype(np.double),
+            fs=self.sampling_rate,
+            f0_floor=self.f0_min,
+            f0_ceil=self.f0_max,
+            frame_period=1000 * self.hop_length / self.sampling_rate,
+        )
+        f0 = pyworld.stonemask(wav.astype(np.double), f0, t, self.sampling_rate)
+        return self.interpolate_f0(self.resize_f0(f0, p_len))

rvc/lib/infer_pack/modules/F0Predictor/PMF0Predictor.py

@@ -0,0 +1,98 @@
+import numpy as np
+import parselmouth
+
+from rvc.lib.infer_pack.modules.F0Predictor.F0Predictor import F0Predictor
+
+
+class PMF0Predictor(F0Predictor):
+    def __init__(self, hop_length=512, f0_min=50, f0_max=1100, sampling_rate=44100):
+        self.hop_length = hop_length
+        self.f0_min = f0_min
+        self.f0_max = f0_max
+        self.sampling_rate = sampling_rate
+
+    def interpolate_f0(self, f0):
+        """
+        对F0进行插值处理
+        """
+
+        data = np.reshape(f0, (f0.size, 1))
+
+        vuv_vector = np.zeros((data.size, 1), dtype=np.float32)
+        vuv_vector[data > 0.0] = 1.0
+        vuv_vector[data <= 0.0] = 0.0
+
+        ip_data = data
+
+        frame_number = data.size
+        last_value = 0.0
+        for i in range(frame_number):
+            if data[i] <= 0.0:
+                j = i + 1
+                for j in range(i + 1, frame_number):
+                    if data[j] > 0.0:
+                        break
+                if j < frame_number - 1:
+                    if last_value > 0.0:
+                        step = (data[j] - data[i - 1]) / float(j - i)
+                        for k in range(i, j):
+                            ip_data[k] = data[i - 1] + step * (k - i + 1)
+                    else:
+                        for k in range(i, j):
+                            ip_data[k] = data[j]
+                else:
+                    for k in range(i, frame_number):
+                        ip_data[k] = last_value
+            else:
+                ip_data[i] = data[i]  # 这里可能存在一个没有必要的拷贝
+                last_value = data[i]
+
+        return ip_data[:, 0], vuv_vector[:, 0]
+
+    def compute_f0(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0
+
+    def compute_f0_uv(self, wav, p_len=None):
+        x = wav
+        if p_len is None:
+            p_len = x.shape[0] // self.hop_length
+        else:
+            assert abs(p_len - x.shape[0] // self.hop_length) < 4, "pad length error"
+        time_step = self.hop_length / self.sampling_rate * 1000
+        f0 = (
+            parselmouth.Sound(x, self.sampling_rate)
+            .to_pitch_ac(
+                time_step=time_step / 1000,
+                voicing_threshold=0.6,
+                pitch_floor=self.f0_min,
+                pitch_ceiling=self.f0_max,
+            )
+            .selected_array["frequency"]
+        )
+
+        pad_size = (p_len - len(f0) + 1) // 2
+        if pad_size > 0 or p_len - len(f0) - pad_size > 0:
+            f0 = np.pad(f0, [[pad_size, p_len - len(f0) - pad_size]], mode="constant")
+        f0, uv = self.interpolate_f0(f0)
+        return f0, uv

rvc/lib/infer_pack/onnx_inference.py

@@ -0,0 +1,149 @@
+import logging
+
+import librosa
+import numpy as np
+import onnxruntime
+import soundfile
+
+logger = logging.getLogger(__name__)
+
+
+class ContentVec:
+    def __init__(self, vec_path="pretrained/vec-768-layer-12.onnx", device=None):
+        logger.info("Load model(s) from {}".format(vec_path))
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(vec_path, providers=providers)
+
+    def __call__(self, wav):
+        return self.forward(wav)
+
+    def forward(self, wav):
+        feats = wav
+        if feats.ndim == 2:  # double channels
+            feats = feats.mean(-1)
+        assert feats.ndim == 1, feats.ndim
+        feats = np.expand_dims(np.expand_dims(feats, 0), 0)
+        onnx_input = {self.model.get_inputs()[0].name: feats}
+        logits = self.model.run(None, onnx_input)[0]
+        return logits.transpose(0, 2, 1)
+
+
+def get_f0_predictor(f0_predictor, hop_length, sampling_rate, **kargs):
+    if f0_predictor == "pm":
+        from lib.infer_pack.modules.F0Predictor.PMF0Predictor import PMF0Predictor
+
+        f0_predictor_object = PMF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "harvest":
+        from lib.infer_pack.modules.F0Predictor.HarvestF0Predictor import (
+            HarvestF0Predictor,
+        )
+
+        f0_predictor_object = HarvestF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    elif f0_predictor == "dio":
+        from lib.infer_pack.modules.F0Predictor.DioF0Predictor import DioF0Predictor
+
+        f0_predictor_object = DioF0Predictor(
+            hop_length=hop_length, sampling_rate=sampling_rate
+        )
+    else:
+        raise Exception("Unknown f0 predictor")
+    return f0_predictor_object
+
+
+class OnnxRVC:
+    def __init__(
+        self,
+        model_path,
+        sr=40000,
+        hop_size=512,
+        vec_path="vec-768-layer-12",
+        device="cpu",
+    ):
+        vec_path = f"pretrained/{vec_path}.onnx"
+        self.vec_model = ContentVec(vec_path, device)
+        if device == "cpu" or device is None:
+            providers = ["CPUExecutionProvider"]
+        elif device == "cuda":
+            providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+        elif device == "dml":
+            providers = ["DmlExecutionProvider"]
+        else:
+            raise RuntimeError("Unsportted Device")
+        self.model = onnxruntime.InferenceSession(model_path, providers=providers)
+        self.sampling_rate = sr
+        self.hop_size = hop_size
+
+    def forward(self, hubert, hubert_length, pitch, pitchf, ds, rnd):
+        onnx_input = {
+            self.model.get_inputs()[0].name: hubert,
+            self.model.get_inputs()[1].name: hubert_length,
+            self.model.get_inputs()[2].name: pitch,
+            self.model.get_inputs()[3].name: pitchf,
+            self.model.get_inputs()[4].name: ds,
+            self.model.get_inputs()[5].name: rnd,
+        }
+        return (self.model.run(None, onnx_input)[0] * 32767).astype(np.int16)
+
+    def inference(
+        self,
+        raw_path,
+        sid,
+        f0_method="dio",
+        f0_up_key=0,
+        pad_time=0.5,
+        cr_threshold=0.02,
+    ):
+        f0_min = 50
+        f0_max = 1100
+        f0_mel_min = 1127 * np.log(1 + f0_min / 700)
+        f0_mel_max = 1127 * np.log(1 + f0_max / 700)
+        f0_predictor = get_f0_predictor(
+            f0_method,
+            hop_length=self.hop_size,
+            sampling_rate=self.sampling_rate,
+            threshold=cr_threshold,
+        )
+        wav, sr = librosa.load(raw_path, sr=self.sampling_rate)
+        org_length = len(wav)
+        if org_length / sr > 50.0:
+            raise RuntimeError("Reached Max Length")
+
+        wav16k = librosa.resample(wav, orig_sr=self.sampling_rate, target_sr=16000)
+        wav16k = wav16k
+
+        hubert = self.vec_model(wav16k)
+        hubert = np.repeat(hubert, 2, axis=2).transpose(0, 2, 1).astype(np.float32)
+        hubert_length = hubert.shape[1]
+
+        pitchf = f0_predictor.compute_f0(wav, hubert_length)
+        pitchf = pitchf * 2 ** (f0_up_key / 12)
+        pitch = pitchf.copy()
+        f0_mel = 1127 * np.log(1 + pitch / 700)
+        f0_mel[f0_mel > 0] = (f0_mel[f0_mel > 0] - f0_mel_min) * 254 / (
+            f0_mel_max - f0_mel_min
+        ) + 1
+        f0_mel[f0_mel <= 1] = 1
+        f0_mel[f0_mel > 255] = 255
+        pitch = np.rint(f0_mel).astype(np.int64)
+
+        pitchf = pitchf.reshape(1, len(pitchf)).astype(np.float32)
+        pitch = pitch.reshape(1, len(pitch))
+        ds = np.array([sid]).astype(np.int64)
+
+        rnd = np.random.randn(1, 192, hubert_length).astype(np.float32)
+        hubert_length = np.array([hubert_length]).astype(np.int64)
+
+        out_wav = self.forward(hubert, hubert_length, pitch, pitchf, ds, rnd).squeeze()
+        out_wav = np.pad(out_wav, (0, 2 * self.hop_size), "constant")
+        return out_wav[0:org_length]

rvc/lib/infer_pack/transforms.py

@@ -0,0 +1,207 @@
+import numpy as np
+import torch
+from torch.nn import functional as F
+
+DEFAULT_MIN_BIN_WIDTH = 1e-3
+DEFAULT_MIN_BIN_HEIGHT = 1e-3
+DEFAULT_MIN_DERIVATIVE = 1e-3
+
+
+def piecewise_rational_quadratic_transform(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails=None,
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if tails is None:
+        spline_fn = rational_quadratic_spline
+        spline_kwargs = {}
+    else:
+        spline_fn = unconstrained_rational_quadratic_spline
+        spline_kwargs = {"tails": tails, "tail_bound": tail_bound}
+
+    outputs, logabsdet = spline_fn(
+        inputs=inputs,
+        unnormalized_widths=unnormalized_widths,
+        unnormalized_heights=unnormalized_heights,
+        unnormalized_derivatives=unnormalized_derivatives,
+        inverse=inverse,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+        **spline_kwargs
+    )
+    return outputs, logabsdet
+
+
+def searchsorted(bin_locations, inputs, eps=1e-6):
+    bin_locations[..., -1] += eps
+    return torch.sum(inputs[..., None] >= bin_locations, dim=-1) - 1
+
+
+def unconstrained_rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    tails="linear",
+    tail_bound=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    inside_interval_mask = (inputs >= -tail_bound) & (inputs <= tail_bound)
+    outside_interval_mask = ~inside_interval_mask
+
+    outputs = torch.zeros_like(inputs)
+    logabsdet = torch.zeros_like(inputs)
+
+    if tails == "linear":
+        unnormalized_derivatives = F.pad(unnormalized_derivatives, pad=(1, 1))
+        constant = np.log(np.exp(1 - min_derivative) - 1)
+        unnormalized_derivatives[..., 0] = constant
+        unnormalized_derivatives[..., -1] = constant
+
+        outputs[outside_interval_mask] = inputs[outside_interval_mask]
+        logabsdet[outside_interval_mask] = 0
+    else:
+        raise RuntimeError("{} tails are not implemented.".format(tails))
+
+    (
+        outputs[inside_interval_mask],
+        logabsdet[inside_interval_mask],
+    ) = rational_quadratic_spline(
+        inputs=inputs[inside_interval_mask],
+        unnormalized_widths=unnormalized_widths[inside_interval_mask, :],
+        unnormalized_heights=unnormalized_heights[inside_interval_mask, :],
+        unnormalized_derivatives=unnormalized_derivatives[inside_interval_mask, :],
+        inverse=inverse,
+        left=-tail_bound,
+        right=tail_bound,
+        bottom=-tail_bound,
+        top=tail_bound,
+        min_bin_width=min_bin_width,
+        min_bin_height=min_bin_height,
+        min_derivative=min_derivative,
+    )
+
+    return outputs, logabsdet
+
+
+def rational_quadratic_spline(
+    inputs,
+    unnormalized_widths,
+    unnormalized_heights,
+    unnormalized_derivatives,
+    inverse=False,
+    left=0.0,
+    right=1.0,
+    bottom=0.0,
+    top=1.0,
+    min_bin_width=DEFAULT_MIN_BIN_WIDTH,
+    min_bin_height=DEFAULT_MIN_BIN_HEIGHT,
+    min_derivative=DEFAULT_MIN_DERIVATIVE,
+):
+    if torch.min(inputs) < left or torch.max(inputs) > right:
+        raise ValueError("Input to a transform is not within its domain")
+
+    num_bins = unnormalized_widths.shape[-1]
+
+    if min_bin_width * num_bins > 1.0:
+        raise ValueError("Minimal bin width too large for the number of bins")
+    if min_bin_height * num_bins > 1.0:
+        raise ValueError("Minimal bin height too large for the number of bins")
+
+    widths = F.softmax(unnormalized_widths, dim=-1)
+    widths = min_bin_width + (1 - min_bin_width * num_bins) * widths
+    cumwidths = torch.cumsum(widths, dim=-1)
+    cumwidths = F.pad(cumwidths, pad=(1, 0), mode="constant", value=0.0)
+    cumwidths = (right - left) * cumwidths + left
+    cumwidths[..., 0] = left
+    cumwidths[..., -1] = right
+    widths = cumwidths[..., 1:] - cumwidths[..., :-1]
+
+    derivatives = min_derivative + F.softplus(unnormalized_derivatives)
+
+    heights = F.softmax(unnormalized_heights, dim=-1)
+    heights = min_bin_height + (1 - min_bin_height * num_bins) * heights
+    cumheights = torch.cumsum(heights, dim=-1)
+    cumheights = F.pad(cumheights, pad=(1, 0), mode="constant", value=0.0)
+    cumheights = (top - bottom) * cumheights + bottom
+    cumheights[..., 0] = bottom
+    cumheights[..., -1] = top
+    heights = cumheights[..., 1:] - cumheights[..., :-1]
+
+    if inverse:
+        bin_idx = searchsorted(cumheights, inputs)[..., None]
+    else:
+        bin_idx = searchsorted(cumwidths, inputs)[..., None]
+
+    input_cumwidths = cumwidths.gather(-1, bin_idx)[..., 0]
+    input_bin_widths = widths.gather(-1, bin_idx)[..., 0]
+
+    input_cumheights = cumheights.gather(-1, bin_idx)[..., 0]
+    delta = heights / widths
+    input_delta = delta.gather(-1, bin_idx)[..., 0]
+
+    input_derivatives = derivatives.gather(-1, bin_idx)[..., 0]
+    input_derivatives_plus_one = derivatives[..., 1:].gather(-1, bin_idx)[..., 0]
+
+    input_heights = heights.gather(-1, bin_idx)[..., 0]
+
+    if inverse:
+        a = (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        ) + input_heights * (input_delta - input_derivatives)
+        b = input_heights * input_derivatives - (inputs - input_cumheights) * (
+            input_derivatives + input_derivatives_plus_one - 2 * input_delta
+        )
+        c = -input_delta * (inputs - input_cumheights)
+
+        discriminant = b.pow(2) - 4 * a * c
+        assert (discriminant >= 0).all()
+
+        root = (2 * c) / (-b - torch.sqrt(discriminant))
+        outputs = root * input_bin_widths + input_cumwidths
+
+        theta_one_minus_theta = root * (1 - root)
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * root.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - root).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, -logabsdet
+    else:
+        theta = (inputs - input_cumwidths) / input_bin_widths
+        theta_one_minus_theta = theta * (1 - theta)
+
+        numerator = input_heights * (
+            input_delta * theta.pow(2) + input_derivatives * theta_one_minus_theta
+        )
+        denominator = input_delta + (
+            (input_derivatives + input_derivatives_plus_one - 2 * input_delta)
+            * theta_one_minus_theta
+        )
+        outputs = input_cumheights + numerator / denominator
+
+        derivative_numerator = input_delta.pow(2) * (
+            input_derivatives_plus_one * theta.pow(2)
+            + 2 * input_delta * theta_one_minus_theta
+            + input_derivatives * (1 - theta).pow(2)
+        )
+        logabsdet = torch.log(derivative_numerator) - 2 * torch.log(denominator)
+
+        return outputs, logabsdet

rvc/lib/ipex/__init__.py

@@ -0,0 +1,182 @@
+import contextlib
+import os
+import sys
+
+import intel_extension_for_pytorch as ipex  # pylint: disable=import-error, unused-import
+import torch
+
+from .attention import attention_init
+from .hijacks import ipex_hijacks
+
+
+def ipex_init():
+    try:
+        # Replace cuda with xpu:
+        torch.cuda.current_device = torch.xpu.current_device
+        torch.cuda.current_stream = torch.xpu.current_stream
+        torch.cuda.device = torch.xpu.device
+        torch.cuda.device_count = torch.xpu.device_count
+        torch.cuda.device_of = torch.xpu.device_of
+        torch.cuda.get_device_name = torch.xpu.get_device_name
+        torch.cuda.get_device_properties = torch.xpu.get_device_properties
+        torch.cuda.init = torch.xpu.init
+        torch.cuda.is_available = torch.xpu.is_available
+        torch.cuda.is_initialized = torch.xpu.is_initialized
+        torch.cuda.is_current_stream_capturing = lambda: False
+        torch.cuda.set_device = torch.xpu.set_device
+        torch.cuda.stream = torch.xpu.stream
+        torch.cuda.synchronize = torch.xpu.synchronize
+        torch.cuda.Event = torch.xpu.Event
+        torch.cuda.Stream = torch.xpu.Stream
+        torch.cuda.FloatTensor = torch.xpu.FloatTensor
+        torch.Tensor.cuda = torch.Tensor.xpu
+        torch.Tensor.is_cuda = torch.Tensor.is_xpu
+        torch.cuda._initialization_lock = torch.xpu.lazy_init._initialization_lock
+        torch.cuda._initialized = torch.xpu.lazy_init._initialized
+        torch.cuda._lazy_seed_tracker = torch.xpu.lazy_init._lazy_seed_tracker
+        torch.cuda._queued_calls = torch.xpu.lazy_init._queued_calls
+        torch.cuda._tls = torch.xpu.lazy_init._tls
+        torch.cuda.threading = torch.xpu.lazy_init.threading
+        torch.cuda.traceback = torch.xpu.lazy_init.traceback
+        torch.cuda.Optional = torch.xpu.Optional
+        torch.cuda.__cached__ = torch.xpu.__cached__
+        torch.cuda.__loader__ = torch.xpu.__loader__
+        torch.cuda.ComplexFloatStorage = torch.xpu.ComplexFloatStorage
+        torch.cuda.Tuple = torch.xpu.Tuple
+        torch.cuda.streams = torch.xpu.streams
+        torch.cuda._lazy_new = torch.xpu._lazy_new
+        torch.cuda.FloatStorage = torch.xpu.FloatStorage
+        torch.cuda.Any = torch.xpu.Any
+        torch.cuda.__doc__ = torch.xpu.__doc__
+        torch.cuda.default_generators = torch.xpu.default_generators
+        torch.cuda.HalfTensor = torch.xpu.HalfTensor
+        torch.cuda._get_device_index = torch.xpu._get_device_index
+        torch.cuda.__path__ = torch.xpu.__path__
+        torch.cuda.Device = torch.xpu.Device
+        torch.cuda.IntTensor = torch.xpu.IntTensor
+        torch.cuda.ByteStorage = torch.xpu.ByteStorage
+        torch.cuda.set_stream = torch.xpu.set_stream
+        torch.cuda.BoolStorage = torch.xpu.BoolStorage
+        torch.cuda.os = torch.xpu.os
+        torch.cuda.torch = torch.xpu.torch
+        torch.cuda.BFloat16Storage = torch.xpu.BFloat16Storage
+        torch.cuda.Union = torch.xpu.Union
+        torch.cuda.DoubleTensor = torch.xpu.DoubleTensor
+        torch.cuda.ShortTensor = torch.xpu.ShortTensor
+        torch.cuda.LongTensor = torch.xpu.LongTensor
+        torch.cuda.IntStorage = torch.xpu.IntStorage
+        torch.cuda.LongStorage = torch.xpu.LongStorage
+        torch.cuda.__annotations__ = torch.xpu.__annotations__
+        torch.cuda.__package__ = torch.xpu.__package__
+        torch.cuda.__builtins__ = torch.xpu.__builtins__
+        torch.cuda.CharTensor = torch.xpu.CharTensor
+        torch.cuda.List = torch.xpu.List
+        torch.cuda._lazy_init = torch.xpu._lazy_init
+        torch.cuda.BFloat16Tensor = torch.xpu.BFloat16Tensor
+        torch.cuda.DoubleStorage = torch.xpu.DoubleStorage
+        torch.cuda.ByteTensor = torch.xpu.ByteTensor
+        torch.cuda.StreamContext = torch.xpu.StreamContext
+        torch.cuda.ComplexDoubleStorage = torch.xpu.ComplexDoubleStorage
+        torch.cuda.ShortStorage = torch.xpu.ShortStorage
+        torch.cuda._lazy_call = torch.xpu._lazy_call
+        torch.cuda.HalfStorage = torch.xpu.HalfStorage
+        torch.cuda.random = torch.xpu.random
+        torch.cuda._device = torch.xpu._device
+        torch.cuda.classproperty = torch.xpu.classproperty
+        torch.cuda.__name__ = torch.xpu.__name__
+        torch.cuda._device_t = torch.xpu._device_t
+        torch.cuda.warnings = torch.xpu.warnings
+        torch.cuda.__spec__ = torch.xpu.__spec__
+        torch.cuda.BoolTensor = torch.xpu.BoolTensor
+        torch.cuda.CharStorage = torch.xpu.CharStorage
+        torch.cuda.__file__ = torch.xpu.__file__
+        torch.cuda._is_in_bad_fork = torch.xpu.lazy_init._is_in_bad_fork
+        # torch.cuda.is_current_stream_capturing = torch.xpu.is_current_stream_capturing
+
+        # Memory:
+        torch.cuda.memory = torch.xpu.memory
+        if "linux" in sys.platform and "WSL2" in os.popen("uname -a").read():
+            torch.xpu.empty_cache = lambda: None
+        torch.cuda.empty_cache = torch.xpu.empty_cache
+        torch.cuda.memory_stats = torch.xpu.memory_stats
+        torch.cuda.memory_summary = torch.xpu.memory_summary
+        torch.cuda.memory_snapshot = torch.xpu.memory_snapshot
+        torch.cuda.memory_allocated = torch.xpu.memory_allocated
+        torch.cuda.max_memory_allocated = torch.xpu.max_memory_allocated
+        torch.cuda.memory_reserved = torch.xpu.memory_reserved
+        torch.cuda.memory_cached = torch.xpu.memory_reserved
+        torch.cuda.max_memory_reserved = torch.xpu.max_memory_reserved
+        torch.cuda.max_memory_cached = torch.xpu.max_memory_reserved
+        torch.cuda.reset_peak_memory_stats = torch.xpu.reset_peak_memory_stats
+        torch.cuda.reset_max_memory_cached = torch.xpu.reset_peak_memory_stats
+        torch.cuda.reset_max_memory_allocated = torch.xpu.reset_peak_memory_stats
+        torch.cuda.memory_stats_as_nested_dict = torch.xpu.memory_stats_as_nested_dict
+        torch.cuda.reset_accumulated_memory_stats = (
+            torch.xpu.reset_accumulated_memory_stats
+        )
+
+        # RNG:
+        torch.cuda.get_rng_state = torch.xpu.get_rng_state
+        torch.cuda.get_rng_state_all = torch.xpu.get_rng_state_all
+        torch.cuda.set_rng_state = torch.xpu.set_rng_state
+        torch.cuda.set_rng_state_all = torch.xpu.set_rng_state_all
+        torch.cuda.manual_seed = torch.xpu.manual_seed
+        torch.cuda.manual_seed_all = torch.xpu.manual_seed_all
+        torch.cuda.seed = torch.xpu.seed
+        torch.cuda.seed_all = torch.xpu.seed_all
+        torch.cuda.initial_seed = torch.xpu.initial_seed
+
+        # AMP:
+        torch.cuda.amp = torch.xpu.amp
+        if not hasattr(torch.cuda.amp, "common"):
+            torch.cuda.amp.common = contextlib.nullcontext()
+        torch.cuda.amp.common.amp_definitely_not_available = lambda: False
+        try:
+            torch.cuda.amp.GradScaler = torch.xpu.amp.GradScaler
+        except Exception:
+            try:
+                from .gradscaler import gradscaler_init
+
+                gradscaler_init()
+                torch.cuda.amp.GradScaler = torch.xpu.amp.GradScaler
+            except Exception:
+                torch.cuda.amp.GradScaler = ipex.cpu.autocast._grad_scaler.GradScaler
+
+        # C
+        torch._C._cuda_getCurrentRawStream = ipex._C._getCurrentStream
+        ipex._C._DeviceProperties.major = 2023
+        ipex._C._DeviceProperties.minor = 2
+
+        # Fix functions with ipex:
+        torch.cuda.mem_get_info = lambda device=None: [
+            (
+                torch.xpu.get_device_properties(device).total_memory
+                - torch.xpu.memory_allocated(device)
+            ),
+            torch.xpu.get_device_properties(device).total_memory,
+        ]
+        torch._utils._get_available_device_type = lambda: "xpu"
+        torch.has_cuda = True
+        torch.cuda.has_half = True
+        torch.cuda.is_bf16_supported = lambda *args, **kwargs: True
+        torch.cuda.is_fp16_supported = lambda *args, **kwargs: True
+        torch.version.cuda = "11.7"
+        torch.cuda.get_device_capability = lambda *args, **kwargs: [11, 7]
+        torch.cuda.get_device_properties.major = 11
+        torch.cuda.get_device_properties.minor = 7
+        torch.cuda.ipc_collect = lambda *args, **kwargs: None
+        torch.cuda.utilization = lambda *args, **kwargs: 0
+        if hasattr(torch.xpu, "getDeviceIdListForCard"):
+            torch.cuda.getDeviceIdListForCard = torch.xpu.getDeviceIdListForCard
+            torch.cuda.get_device_id_list_per_card = torch.xpu.getDeviceIdListForCard
+        else:
+            torch.cuda.getDeviceIdListForCard = torch.xpu.get_device_id_list_per_card
+            torch.cuda.get_device_id_list_per_card = (
+                torch.xpu.get_device_id_list_per_card
+            )
+
+        ipex_hijacks()
+        attention_init()
+    except Exception as e:
+        return False, e
+    return True, None

rvc/lib/ipex/attention.py

@@ -0,0 +1,206 @@
+import intel_extension_for_pytorch as ipex
+import torch
+
+original_torch_bmm = torch.bmm
+
+
+def torch_bmm(input, mat2, *, out=None):
+    if input.dtype != mat2.dtype:
+        mat2 = mat2.to(input.dtype)
+
+    # ARC GPUs can't allocate more than 4GB to a single block, Slice it:
+    batch_size_attention, input_tokens, mat2_shape = (
+        input.shape[0],
+        input.shape[1],
+        mat2.shape[2],
+    )
+    block_multiply = input.element_size()
+    slice_block_size = input_tokens * mat2_shape / 1024 / 1024 * block_multiply
+    block_size = batch_size_attention * slice_block_size
+
+    split_slice_size = batch_size_attention
+    if block_size > 4:
+        do_split = True
+        # Find something divisible with the input_tokens
+        while (split_slice_size * slice_block_size) > 4:
+            split_slice_size = split_slice_size // 2
+            if split_slice_size <= 1:
+                split_slice_size = 1
+                break
+    else:
+        do_split = False
+
+    split_2_slice_size = input_tokens
+    if split_slice_size * slice_block_size > 4:
+        slice_block_size2 = split_slice_size * mat2_shape / 1024 / 1024 * block_multiply
+        do_split_2 = True
+        # Find something divisible with the input_tokens
+        while (split_2_slice_size * slice_block_size2) > 4:
+            split_2_slice_size = split_2_slice_size // 2
+            if split_2_slice_size <= 1:
+                split_2_slice_size = 1
+                break
+    else:
+        do_split_2 = False
+
+    if do_split:
+        hidden_states = torch.zeros(
+            input.shape[0],
+            input.shape[1],
+            mat2.shape[2],
+            device=input.device,
+            dtype=input.dtype,
+        )
+        for i in range(batch_size_attention // split_slice_size):
+            start_idx = i * split_slice_size
+            end_idx = (i + 1) * split_slice_size
+            if do_split_2:
+                for i2 in range(input_tokens // split_2_slice_size):
+                    start_idx_2 = i2 * split_2_slice_size
+                    end_idx_2 = (i2 + 1) * split_2_slice_size
+                    hidden_states[
+                        start_idx:end_idx, start_idx_2:end_idx_2
+                    ] = original_torch_bmm(
+                        input[start_idx:end_idx, start_idx_2:end_idx_2],
+                        mat2[start_idx:end_idx, start_idx_2:end_idx_2],
+                        out=out,
+                    )
+            else:
+                hidden_states[start_idx:end_idx] = original_torch_bmm(
+                    input[start_idx:end_idx], mat2[start_idx:end_idx], out=out
+                )
+    else:
+        return original_torch_bmm(input, mat2, out=out)
+    return hidden_states
+
+
+original_scaled_dot_product_attention = torch.nn.functional.scaled_dot_product_attention
+
+
+def scaled_dot_product_attention(
+    query, key, value, attn_mask=None, dropout_p=0.0, is_causal=False
+):
+    # ARC GPUs can't allocate more than 4GB to a single block, Slice it:
+    if len(query.shape) == 3:
+        batch_size_attention, query_tokens, shape_four = query.shape
+        shape_one = 1
+        no_shape_one = True
+    else:
+        shape_one, batch_size_attention, query_tokens, shape_four = query.shape
+        no_shape_one = False
+
+    block_multiply = query.element_size()
+    slice_block_size = (
+        shape_one * query_tokens * shape_four / 1024 / 1024 * block_multiply
+    )
+    block_size = batch_size_attention * slice_block_size
+
+    split_slice_size = batch_size_attention
+    if block_size > 4:
+        do_split = True
+        # Find something divisible with the shape_one
+        while (split_slice_size * slice_block_size) > 4:
+            split_slice_size = split_slice_size // 2
+            if split_slice_size <= 1:
+                split_slice_size = 1
+                break
+    else:
+        do_split = False
+
+    split_2_slice_size = query_tokens
+    if split_slice_size * slice_block_size > 4:
+        slice_block_size2 = (
+            shape_one * split_slice_size * shape_four / 1024 / 1024 * block_multiply
+        )
+        do_split_2 = True
+        # Find something divisible with the batch_size_attention
+        while (split_2_slice_size * slice_block_size2) > 4:
+            split_2_slice_size = split_2_slice_size // 2
+            if split_2_slice_size <= 1:
+                split_2_slice_size = 1
+                break
+    else:
+        do_split_2 = False
+
+    if do_split:
+        hidden_states = torch.zeros(query.shape, device=query.device, dtype=query.dtype)
+        for i in range(batch_size_attention // split_slice_size):
+            start_idx = i * split_slice_size
+            end_idx = (i + 1) * split_slice_size
+            if do_split_2:
+                for i2 in range(query_tokens // split_2_slice_size):
+                    start_idx_2 = i2 * split_2_slice_size
+                    end_idx_2 = (i2 + 1) * split_2_slice_size
+                    if no_shape_one:
+                        hidden_states[
+                            start_idx:end_idx, start_idx_2:end_idx_2
+                        ] = original_scaled_dot_product_attention(
+                            query[start_idx:end_idx, start_idx_2:end_idx_2],
+                            key[start_idx:end_idx, start_idx_2:end_idx_2],
+                            value[start_idx:end_idx, start_idx_2:end_idx_2],
+                            attn_mask=attn_mask[
+                                start_idx:end_idx, start_idx_2:end_idx_2
+                            ]
+                            if attn_mask is not None
+                            else attn_mask,
+                            dropout_p=dropout_p,
+                            is_causal=is_causal,
+                        )
+                    else:
+                        hidden_states[
+                            :, start_idx:end_idx, start_idx_2:end_idx_2
+                        ] = original_scaled_dot_product_attention(
+                            query[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            key[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            value[:, start_idx:end_idx, start_idx_2:end_idx_2],
+                            attn_mask=attn_mask[
+                                :, start_idx:end_idx, start_idx_2:end_idx_2
+                            ]
+                            if attn_mask is not None
+                            else attn_mask,
+                            dropout_p=dropout_p,
+                            is_causal=is_causal,
+                        )
+            else:
+                if no_shape_one:
+                    hidden_states[
+                        start_idx:end_idx
+                    ] = original_scaled_dot_product_attention(
+                        query[start_idx:end_idx],
+                        key[start_idx:end_idx],
+                        value[start_idx:end_idx],
+                        attn_mask=attn_mask[start_idx:end_idx]
+                        if attn_mask is not None
+                        else attn_mask,
+                        dropout_p=dropout_p,
+                        is_causal=is_causal,
+                    )
+                else:
+                    hidden_states[
+                        :, start_idx:end_idx
+                    ] = original_scaled_dot_product_attention(
+                        query[:, start_idx:end_idx],
+                        key[:, start_idx:end_idx],
+                        value[:, start_idx:end_idx],
+                        attn_mask=attn_mask[:, start_idx:end_idx]
+                        if attn_mask is not None
+                        else attn_mask,
+                        dropout_p=dropout_p,
+                        is_causal=is_causal,
+                    )
+    else:
+        return original_scaled_dot_product_attention(
+            query,
+            key,
+            value,
+            attn_mask=attn_mask,
+            dropout_p=dropout_p,
+            is_causal=is_causal,
+        )
+    return hidden_states
+
+
+def attention_init():
+    # ARC GPUs can't allocate more than 4GB to a single block:
+    torch.bmm = torch_bmm
+    torch.nn.functional.scaled_dot_product_attention = scaled_dot_product_attention

rvc/lib/ipex/gradscaler.py

@@ -0,0 +1,184 @@
+from collections import defaultdict
+
+import intel_extension_for_pytorch as ipex
+import intel_extension_for_pytorch._C as core
+import torch
+
+OptState = ipex.cpu.autocast._grad_scaler.OptState
+_MultiDeviceReplicator = ipex.cpu.autocast._grad_scaler._MultiDeviceReplicator
+_refresh_per_optimizer_state = (
+    ipex.cpu.autocast._grad_scaler._refresh_per_optimizer_state
+)
+
+
+def _unscale_grads_(self, optimizer, inv_scale, found_inf, allow_fp16):
+    per_device_inv_scale = _MultiDeviceReplicator(inv_scale)
+    per_device_found_inf = _MultiDeviceReplicator(found_inf)
+
+    # To set up _amp_foreach_non_finite_check_and_unscale_, split grads by device and dtype.
+    # There could be hundreds of grads, so we'd like to iterate through them just once.
+    # However, we don't know their devices or dtypes in advance.
+
+    # https://stackoverflow.com/questions/5029934/defaultdict-of-defaultdict
+    # Google says mypy struggles with defaultdicts type annotations.
+    per_device_and_dtype_grads = defaultdict(lambda: defaultdict(list))  # type: ignore[var-annotated]
+    # sync grad to master weight
+    if hasattr(optimizer, "sync_grad"):
+        optimizer.sync_grad()
+    with torch.no_grad():
+        for group in optimizer.param_groups:
+            for param in group["params"]:
+                if param.grad is None:
+                    continue
+                if (not allow_fp16) and param.grad.dtype == torch.float16:
+                    raise ValueError("Attempting to unscale FP16 gradients.")
+                if param.grad.is_sparse:
+                    # is_coalesced() == False means the sparse grad has values with duplicate indices.
+                    # coalesce() deduplicates indices and adds all values that have the same index.
+                    # For scaled fp16 values, there's a good chance coalescing will cause overflow,
+                    # so we should check the coalesced _values().
+                    if param.grad.dtype is torch.float16:
+                        param.grad = param.grad.coalesce()
+                    to_unscale = param.grad._values()
+                else:
+                    to_unscale = param.grad
+
+                # -: is there a way to split by device and dtype without appending in the inner loop?
+                to_unscale = to_unscale.to("cpu")
+                per_device_and_dtype_grads[to_unscale.device][to_unscale.dtype].append(
+                    to_unscale
+                )
+
+        for _, per_dtype_grads in per_device_and_dtype_grads.items():
+            for grads in per_dtype_grads.values():
+                core._amp_foreach_non_finite_check_and_unscale_(
+                    grads,
+                    per_device_found_inf.get("cpu"),
+                    per_device_inv_scale.get("cpu"),
+                )
+
+    return per_device_found_inf._per_device_tensors
+
+
+def unscale_(self, optimizer):
+    """
+    Divides ("unscales") the optimizer's gradient tensors by the scale factor.
+    :meth:`unscale_` is optional, serving cases where you need to
+    :ref:`modify or inspect gradients<working-with-unscaled-gradients>`
+    between the backward pass(es) and :meth:`step`.
+    If :meth:`unscale_` is not called explicitly,  gradients will be unscaled  automatically during :meth:`step`.
+    Simple example, using :meth:`unscale_` to enable clipping of unscaled gradients::
+        ...
+        scaler.scale(loss).backward()
+        scaler.unscale_(optimizer)
+        torch.nn.utils.clip_grad_norm_(model.parameters(), max_norm)
+        scaler.step(optimizer)
+        scaler.update()
+    Args:
+        optimizer (torch.optim.Optimizer):  Optimizer that owns the gradients to be unscaled.
+    .. warning::
+        :meth:`unscale_` should only be called once per optimizer per :meth:`step` call,
+        and only after all gradients for that optimizer's assigned parameters have been accumulated.
+        Calling :meth:`unscale_` twice for a given optimizer between each :meth:`step` triggers a RuntimeError.
+    .. warning::
+        :meth:`unscale_` may unscale sparse gradients out of place, replacing the ``.grad`` attribute.
+    """
+    if not self._enabled:
+        return
+
+    self._check_scale_growth_tracker("unscale_")
+
+    optimizer_state = self._per_optimizer_states[id(optimizer)]
+
+    if optimizer_state["stage"] is OptState.UNSCALED:
+        raise RuntimeError(
+            "unscale_() has already been called on this optimizer since the last update()."
+        )
+    elif optimizer_state["stage"] is OptState.STEPPED:
+        raise RuntimeError("unscale_() is being called after step().")
+
+    # FP32 division can be imprecise for certain compile options, so we carry out the reciprocal in FP64.
+    assert self._scale is not None
+    inv_scale = (
+        self._scale.to("cpu").double().reciprocal().float().to(self._scale.device)
+    )
+    found_inf = torch.full((1,), 0.0, dtype=torch.float32, device=self._scale.device)
+
+    optimizer_state["found_inf_per_device"] = self._unscale_grads_(
+        optimizer, inv_scale, found_inf, False
+    )
+    optimizer_state["stage"] = OptState.UNSCALED
+
+
+def update(self, new_scale=None):
+    """
+    Updates the scale factor.
+    If any optimizer steps were skipped the scale is multiplied by ``backoff_factor``
+    to reduce it. If ``growth_interval`` unskipped iterations occurred consecutively,
+    the scale is multiplied by ``growth_factor`` to increase it.
+    Passing ``new_scale`` sets the new scale value manually. (``new_scale`` is not
+    used directly, it's used to fill GradScaler's internal scale tensor. So if
+    ``new_scale`` was a tensor, later in-place changes to that tensor will not further
+    affect the scale GradScaler uses internally.)
+    Args:
+        new_scale (float or :class:`torch.FloatTensor`, optional, default=None):  New scale factor.
+    .. warning::
+        :meth:`update` should only be called at the end of the iteration, after ``scaler.step(optimizer)`` has
+        been invoked for all optimizers used this iteration.
+    """
+    if not self._enabled:
+        return
+
+    _scale, _growth_tracker = self._check_scale_growth_tracker("update")
+
+    if new_scale is not None:
+        # Accept a new user-defined scale.
+        if isinstance(new_scale, float):
+            self._scale.fill_(new_scale)  # type: ignore[union-attr]
+        else:
+            reason = "new_scale should be a float or a 1-element torch.FloatTensor with requires_grad=False."
+            assert isinstance(new_scale, torch.FloatTensor), reason  # type: ignore[attr-defined]
+            assert new_scale.numel() == 1, reason
+            assert new_scale.requires_grad is False, reason
+            self._scale.copy_(new_scale)  # type: ignore[union-attr]
+    else:
+        # Consume shared inf/nan data collected from optimizers to update the scale.
+        # If all found_inf tensors are on the same device as self._scale, this operation is asynchronous.
+        found_infs = [
+            found_inf.to(device="cpu", non_blocking=True)
+            for state in self._per_optimizer_states.values()
+            for found_inf in state["found_inf_per_device"].values()
+        ]
+
+        assert len(found_infs) > 0, "No inf checks were recorded prior to update."
+
+        found_inf_combined = found_infs[0]
+        if len(found_infs) > 1:
+            for i in range(1, len(found_infs)):
+                found_inf_combined += found_infs[i]
+
+        to_device = _scale.device
+        _scale = _scale.to("cpu")
+        _growth_tracker = _growth_tracker.to("cpu")
+
+        core._amp_update_scale_(
+            _scale,
+            _growth_tracker,
+            found_inf_combined,
+            self._growth_factor,
+            self._backoff_factor,
+            self._growth_interval,
+        )
+
+        _scale = _scale.to(to_device)
+        _growth_tracker = _growth_tracker.to(to_device)
+    # To prepare for next iteration, clear the data collected from optimizers this iteration.
+    self._per_optimizer_states = defaultdict(_refresh_per_optimizer_state)
+
+
+def gradscaler_init():
+    torch.xpu.amp.GradScaler = ipex.cpu.autocast._grad_scaler.GradScaler
+    torch.xpu.amp.GradScaler._unscale_grads_ = _unscale_grads_
+    torch.xpu.amp.GradScaler.unscale_ = unscale_
+    torch.xpu.amp.GradScaler.update = update
+    return torch.xpu.amp.GradScaler

rvc/lib/ipex/hijacks.py

@@ -0,0 +1,352 @@
+import contextlib
+import importlib
+
+import intel_extension_for_pytorch as ipex
+import torch
+
+
+class CondFunc:
+    def __new__(cls, orig_func, sub_func, cond_func):
+        self = super(CondFunc, cls).__new__(cls)
+        if isinstance(orig_func, str):
+            func_path = orig_func.split(".")
+            for i in range(len(func_path) - 1, -1, -1):
+                try:
+                    resolved_obj = importlib.import_module(".".join(func_path[:i]))
+                    break
+                except ImportError:
+                    pass
+            for attr_name in func_path[i:-1]:
+                resolved_obj = getattr(resolved_obj, attr_name)
+            orig_func = getattr(resolved_obj, func_path[-1])
+            setattr(
+                resolved_obj,
+                func_path[-1],
+                lambda *args, **kwargs: self(*args, **kwargs),
+            )
+        self.__init__(orig_func, sub_func, cond_func)
+        return lambda *args, **kwargs: self(*args, **kwargs)
+
+    def __init__(self, orig_func, sub_func, cond_func):
+        self.__orig_func = orig_func
+        self.__sub_func = sub_func
+        self.__cond_func = cond_func
+
+    def __call__(self, *args, **kwargs):
+        if not self.__cond_func or self.__cond_func(self.__orig_func, *args, **kwargs):
+            return self.__sub_func(self.__orig_func, *args, **kwargs)
+        else:
+            return self.__orig_func(*args, **kwargs)
+
+
+_utils = torch.utils.data._utils
+
+
+def _shutdown_workers(self):
+    if (
+        torch.utils.data._utils is None
+        or torch.utils.data._utils.python_exit_status is True
+        or torch.utils.data._utils.python_exit_status is None
+    ):
+        return
+    if hasattr(self, "_shutdown") and not self._shutdown:
+        self._shutdown = True
+        try:
+            if hasattr(self, "_pin_memory_thread"):
+                self._pin_memory_thread_done_event.set()
+                self._worker_result_queue.put((None, None))
+                self._pin_memory_thread.join()
+                self._worker_result_queue.cancel_join_thread()
+                self._worker_result_queue.close()
+            self._workers_done_event.set()
+            for worker_id in range(len(self._workers)):
+                if self._persistent_workers or self._workers_status[worker_id]:
+                    self._mark_worker_as_unavailable(worker_id, shutdown=True)
+            for w in self._workers:
+                w.join(timeout=torch.utils.data._utils.MP_STATUS_CHECK_INTERVAL)
+            for q in self._index_queues:
+                q.cancel_join_thread()
+                q.close()
+        finally:
+            if self._worker_pids_set:
+                torch.utils.data._utils.signal_handling._remove_worker_pids(id(self))
+                self._worker_pids_set = False
+            for w in self._workers:
+                if w.is_alive():
+                    w.terminate()
+
+
+class DummyDataParallel(torch.nn.Module):
+    def __new__(cls, module, device_ids=None, output_device=None, dim=0):
+        if isinstance(device_ids, list) and len(device_ids) > 1:
+            print("IPEX backend doesn't support DataParallel on multiple XPU devices")
+        return module.to("xpu")
+
+
+def return_null_context(*args, **kwargs):
+    return contextlib.nullcontext()
+
+
+def check_device(device):
+    return bool(
+        (isinstance(device, torch.device) and device.type == "cuda")
+        or (isinstance(device, str) and "cuda" in device)
+        or isinstance(device, int)
+    )
+
+
+def return_xpu(device):
+    return (
+        f"xpu:{device[-1]}"
+        if isinstance(device, str) and ":" in device
+        else f"xpu:{device}"
+        if isinstance(device, int)
+        else torch.device("xpu")
+        if isinstance(device, torch.device)
+        else "xpu"
+    )
+
+
+def ipex_no_cuda(orig_func, *args, **kwargs):
+    torch.cuda.is_available = lambda: False
+    orig_func(*args, **kwargs)
+    torch.cuda.is_available = torch.xpu.is_available
+
+
+original_autocast = torch.autocast
+
+
+def ipex_autocast(*args, **kwargs):
+    if len(args) > 0 and args[0] == "cuda":
+        return original_autocast("xpu", *args[1:], **kwargs)
+    else:
+        return original_autocast(*args, **kwargs)
+
+
+original_torch_cat = torch.cat
+
+
+def torch_cat(tensor, *args, **kwargs):
+    if len(tensor) == 3 and (
+        tensor[0].dtype != tensor[1].dtype or tensor[2].dtype != tensor[1].dtype
+    ):
+        return original_torch_cat(
+            [tensor[0].to(tensor[1].dtype), tensor[1], tensor[2].to(tensor[1].dtype)],
+            *args,
+            **kwargs,
+        )
+    else:
+        return original_torch_cat(tensor, *args, **kwargs)
+
+
+original_interpolate = torch.nn.functional.interpolate
+
+
+def interpolate(
+    tensor,
+    size=None,
+    scale_factor=None,
+    mode="nearest",
+    align_corners=None,
+    recompute_scale_factor=None,
+    antialias=False,
+):
+    if antialias or align_corners is not None:
+        return_device = tensor.device
+        return_dtype = tensor.dtype
+        return original_interpolate(
+            tensor.to("cpu", dtype=torch.float32),
+            size=size,
+            scale_factor=scale_factor,
+            mode=mode,
+            align_corners=align_corners,
+            recompute_scale_factor=recompute_scale_factor,
+            antialias=antialias,
+        ).to(return_device, dtype=return_dtype)
+    else:
+        return original_interpolate(
+            tensor,
+            size=size,
+            scale_factor=scale_factor,
+            mode=mode,
+            align_corners=align_corners,
+            recompute_scale_factor=recompute_scale_factor,
+            antialias=antialias,
+        )
+
+
+original_linalg_solve = torch.linalg.solve
+
+
+def linalg_solve(A, B, *args, **kwargs):
+    if A.device != torch.device("cpu") or B.device != torch.device("cpu"):
+        return_device = A.device
+        return original_linalg_solve(A.to("cpu"), B.to("cpu"), *args, **kwargs).to(
+            return_device
+        )
+    else:
+        return original_linalg_solve(A, B, *args, **kwargs)
+
+
+def ipex_hijacks():
+    CondFunc(
+        "torch.Tensor.to",
+        lambda orig_func, self, device=None, *args, **kwargs: orig_func(
+            self, return_xpu(device), *args, **kwargs
+        ),
+        lambda orig_func, self, device=None, *args, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.Tensor.cuda",
+        lambda orig_func, self, device=None, *args, **kwargs: orig_func(
+            self, return_xpu(device), *args, **kwargs
+        ),
+        lambda orig_func, self, device=None, *args, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.empty",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.load",
+        lambda orig_func, *args, map_location=None, **kwargs: orig_func(
+            *args, return_xpu(map_location), **kwargs
+        ),
+        lambda orig_func, *args, map_location=None, **kwargs: map_location is None
+        or check_device(map_location),
+    )
+    CondFunc(
+        "torch.randn",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.ones",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.zeros",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.tensor",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+    CondFunc(
+        "torch.linspace",
+        lambda orig_func, *args, device=None, **kwargs: orig_func(
+            *args, device=return_xpu(device), **kwargs
+        ),
+        lambda orig_func, *args, device=None, **kwargs: check_device(device),
+    )
+
+    CondFunc(
+        "torch.Generator",
+        lambda orig_func, device=None: torch.xpu.Generator(device),
+        lambda orig_func, device=None: device is not None
+        and device != torch.device("cpu")
+        and device != "cpu",
+    )
+
+    CondFunc(
+        "torch.batch_norm",
+        lambda orig_func, input, weight, bias, *args, **kwargs: orig_func(
+            input,
+            weight
+            if weight is not None
+            else torch.ones(input.size()[1], device=input.device),
+            bias
+            if bias is not None
+            else torch.zeros(input.size()[1], device=input.device),
+            *args,
+            **kwargs,
+        ),
+        lambda orig_func, input, *args, **kwargs: input.device != torch.device("cpu"),
+    )
+    CondFunc(
+        "torch.instance_norm",
+        lambda orig_func, input, weight, bias, *args, **kwargs: orig_func(
+            input,
+            weight
+            if weight is not None
+            else torch.ones(input.size()[1], device=input.device),
+            bias
+            if bias is not None
+            else torch.zeros(input.size()[1], device=input.device),
+            *args,
+            **kwargs,
+        ),
+        lambda orig_func, input, *args, **kwargs: input.device != torch.device("cpu"),
+    )
+
+    # Functions with dtype errors:
+    CondFunc(
+        "torch.nn.modules.GroupNorm.forward",
+        lambda orig_func, self, input: orig_func(
+            self, input.to(self.weight.data.dtype)
+        ),
+        lambda orig_func, self, input: input.dtype != self.weight.data.dtype,
+    )
+    CondFunc(
+        "torch.nn.modules.linear.Linear.forward",
+        lambda orig_func, self, input: orig_func(
+            self, input.to(self.weight.data.dtype)
+        ),
+        lambda orig_func, self, input: input.dtype != self.weight.data.dtype,
+    )
+    CondFunc(
+        "torch.nn.modules.conv.Conv2d.forward",
+        lambda orig_func, self, input: orig_func(
+            self, input.to(self.weight.data.dtype)
+        ),
+        lambda orig_func, self, input: input.dtype != self.weight.data.dtype,
+    )
+    CondFunc(
+        "torch.nn.functional.layer_norm",
+        lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: orig_func(
+            input.to(weight.data.dtype), normalized_shape, weight, *args, **kwargs
+        ),
+        lambda orig_func, input, normalized_shape=None, weight=None, *args, **kwargs: weight
+        is not None
+        and input.dtype != weight.data.dtype,
+    )
+
+    # Diffusers Float64 (ARC GPUs doesn't support double or Float64):
+    if not torch.xpu.has_fp64_dtype():
+        CondFunc(
+            "torch.from_numpy",
+            lambda orig_func, ndarray: orig_func(ndarray.astype("float32")),
+            lambda orig_func, ndarray: ndarray.dtype == float,
+        )
+
+    # Broken functions when torch.cuda.is_available is True:
+    CondFunc(
+        "torch.utils.data.dataloader._BaseDataLoaderIter.__init__",
+        lambda orig_func, *args, **kwargs: ipex_no_cuda(orig_func, *args, **kwargs),
+        lambda orig_func, *args, **kwargs: True,
+    )
+
+    # Functions that make compile mad with CondFunc:
+    torch.utils.data.dataloader._MultiProcessingDataLoaderIter._shutdown_workers = (
+        _shutdown_workers
+    )
+    torch.nn.DataParallel = DummyDataParallel
+    torch.autocast = ipex_autocast
+    torch.cat = torch_cat
+    torch.linalg.solve = linalg_solve
+    torch.nn.functional.interpolate = interpolate
+    torch.backends.cuda.sdp_kernel = return_null_context

rvc/lib/jit/__init__.py

@@ -0,0 +1,164 @@
+import pickle
+import time
+from collections import OrderedDict
+from io import BytesIO
+
+import torch
+from tqdm import tqdm
+
+
+def load_inputs(path, device, is_half=False):
+    parm = torch.load(path, map_location=torch.device("cpu"))
+    for key in parm.keys():
+        parm[key] = parm[key].to(device)
+        if is_half and parm[key].dtype == torch.float32:
+            parm[key] = parm[key].half()
+        elif not is_half and parm[key].dtype == torch.float16:
+            parm[key] = parm[key].float()
+    return parm
+
+
+def benchmark(
+    model, inputs_path, device=torch.device("cpu"), epoch=1000, is_half=False
+):
+    parm = load_inputs(inputs_path, device, is_half)
+    total_ts = 0.0
+    bar = tqdm(range(epoch))
+    for i in bar:
+        start_time = time.perf_counter()
+        o = model(**parm)
+        total_ts += time.perf_counter() - start_time
+    print(f"num_epoch: {epoch} | avg time(ms): {(total_ts*1000)/epoch}")
+
+
+def jit_warm_up(model, inputs_path, device=torch.device("cpu"), epoch=5, is_half=False):
+    benchmark(model, inputs_path, device, epoch=epoch, is_half=is_half)
+
+
+def to_jit_model(
+    model_path,
+    model_type: str,
+    mode: str = "trace",
+    inputs_path: str = None,
+    device=torch.device("cpu"),
+    is_half=False,
+):
+    model = None
+    if model_type.lower() == "synthesizer":
+        from .get_synthesizer import get_synthesizer
+
+        model, _ = get_synthesizer(model_path, device)
+        model.forward = model.infer
+    elif model_type.lower() == "rmvpe":
+        from .get_rmvpe import get_rmvpe
+
+        model = get_rmvpe(model_path, device)
+    elif model_type.lower() == "hubert":
+        from .get_hubert import get_hubert_model
+
+        model = get_hubert_model(model_path, device)
+        model.forward = model.infer
+    else:
+        raise ValueError(f"No model type named {model_type}")
+    model = model.eval()
+    model = model.half() if is_half else model.float()
+    if mode == "trace":
+        assert not inputs_path
+        inputs = load_inputs(inputs_path, device, is_half)
+        model_jit = torch.jit.trace(model, example_kwarg_inputs=inputs)
+    elif mode == "script":
+        model_jit = torch.jit.script(model)
+    model_jit.to(device)
+    model_jit = model_jit.half() if is_half else model_jit.float()
+    # model = model.half() if is_half else model.float()
+    return (model, model_jit)
+
+
+def export(
+    model: torch.nn.Module,
+    mode: str = "trace",
+    inputs: dict = None,
+    device=torch.device("cpu"),
+    is_half: bool = False,
+) -> dict:
+    model = model.half() if is_half else model.float()
+    model.eval()
+    if mode == "trace":
+        assert inputs is not None
+        model_jit = torch.jit.trace(model, example_kwarg_inputs=inputs)
+    elif mode == "script":
+        model_jit = torch.jit.script(model)
+    model_jit.to(device)
+    model_jit = model_jit.half() if is_half else model_jit.float()
+    buffer = BytesIO()
+    # model_jit=model_jit.cpu()
+    torch.jit.save(model_jit, buffer)
+    del model_jit
+    cpt = OrderedDict()
+    cpt["model"] = buffer.getvalue()
+    cpt["is_half"] = is_half
+    return cpt
+
+
+def load(path: str):
+    with open(path, "rb") as f:
+        return pickle.load(f)
+
+
+def save(ckpt: dict, save_path: str):
+    with open(save_path, "wb") as f:
+        pickle.dump(ckpt, f)
+
+
+def rmvpe_jit_export(
+    model_path: str,
+    mode: str = "script",
+    inputs_path: str = None,
+    save_path: str = None,
+    device=torch.device("cpu"),
+    is_half=False,
+):
+    if not save_path:
+        save_path = model_path.rstrip(".pth")
+        save_path += ".half.jit" if is_half else ".jit"
+    if "cuda" in str(device) and ":" not in str(device):
+        device = torch.device("cuda:0")
+    from .get_rmvpe import get_rmvpe
+
+    model = get_rmvpe(model_path, device)
+    inputs = None
+    if mode == "trace":
+        inputs = load_inputs(inputs_path, device, is_half)
+    ckpt = export(model, mode, inputs, device, is_half)
+    ckpt["device"] = str(device)
+    save(ckpt, save_path)
+    return ckpt
+
+
+def synthesizer_jit_export(
+    model_path: str,
+    mode: str = "script",
+    inputs_path: str = None,
+    save_path: str = None,
+    device=torch.device("cpu"),
+    is_half=False,
+):
+    if not save_path:
+        save_path = model_path.rstrip(".pth")
+        save_path += ".half.jit" if is_half else ".jit"
+    if "cuda" in str(device) and ":" not in str(device):
+        device = torch.device("cuda:0")
+    from .get_synthesizer import get_synthesizer
+
+    model, cpt = get_synthesizer(model_path, device)
+    assert isinstance(cpt, dict)
+    model.forward = model.infer
+    inputs = None
+    if mode == "trace":
+        inputs = load_inputs(inputs_path, device, is_half)
+    ckpt = export(model, mode, inputs, device, is_half)
+    cpt.pop("weight")
+    cpt["model"] = ckpt["model"]
+    cpt["device"] = device
+    save(cpt, save_path)
+    return cpt

rvc/lib/jit/get_hubert.py

@@ -0,0 +1,343 @@
+import math
+import random
+from typing import Optional, Tuple
+
+import numpy as np
+import torch
+import torch.nn.functional as F
+from fairseq.checkpoint_utils import load_model_ensemble_and_task
+
+# from fairseq.data.data_utils import compute_mask_indices
+from fairseq.utils import index_put
+
+
+# @torch.jit.script
+def pad_to_multiple(x, multiple, dim=-1, value=0):
+    # Inspired from https://github.com/lucidrains/local-attention/blob/master/local_attention/local_attention.py#L41
+    if x is None:
+        return None, 0
+    tsz = x.size(dim)
+    m = tsz / multiple
+    remainder = math.ceil(m) * multiple - tsz
+    if int(tsz % multiple) == 0:
+        return x, 0
+    pad_offset = (0,) * (-1 - dim) * 2
+
+    return F.pad(x, (*pad_offset, 0, remainder), value=value), remainder
+
+
+def extract_features(
+    self,
+    x,
+    padding_mask=None,
+    tgt_layer=None,
+    min_layer=0,
+):
+    if padding_mask is not None:
+        x = index_put(x, padding_mask, 0)
+
+    x_conv = self.pos_conv(x.transpose(1, 2))
+    x_conv = x_conv.transpose(1, 2)
+    x = x + x_conv
+
+    if not self.layer_norm_first:
+        x = self.layer_norm(x)
+
+    # pad to the sequence length dimension
+    x, pad_length = pad_to_multiple(x, self.required_seq_len_multiple, dim=-2, value=0)
+    if pad_length > 0 and padding_mask is None:
+        padding_mask = x.new_zeros((x.size(0), x.size(1)), dtype=torch.bool)
+        padding_mask[:, -pad_length:] = True
+    else:
+        padding_mask, _ = pad_to_multiple(
+            padding_mask, self.required_seq_len_multiple, dim=-1, value=True
+        )
+    x = F.dropout(x, p=self.dropout, training=self.training)
+
+    # B x T x C -> T x B x C
+    x = x.transpose(0, 1)
+
+    layer_results = []
+    r = None
+    for i, layer in enumerate(self.layers):
+        dropout_probability = np.random.random() if self.layerdrop > 0 else 1
+        if not self.training or (dropout_probability > self.layerdrop):
+            x, (z, lr) = layer(
+                x, self_attn_padding_mask=padding_mask, need_weights=False
+            )
+            if i >= min_layer:
+                layer_results.append((x, z, lr))
+        if i == tgt_layer:
+            r = x
+            break
+
+    if r is not None:
+        x = r
+
+    # T x B x C -> B x T x C
+    x = x.transpose(0, 1)
+
+    # undo paddding
+    if pad_length > 0:
+        x = x[:, :-pad_length]
+
+        def undo_pad(a, b, c):
+            return (
+                a[:-pad_length],
+                b[:-pad_length] if b is not None else b,
+                c[:-pad_length],
+            )
+
+        layer_results = [undo_pad(*u) for u in layer_results]
+
+    return x, layer_results
+
+
+def compute_mask_indices(
+    shape: Tuple[int, int],
+    padding_mask: Optional[torch.Tensor],
+    mask_prob: float,
+    mask_length: int,
+    mask_type: str = "static",
+    mask_other: float = 0.0,
+    min_masks: int = 0,
+    no_overlap: bool = False,
+    min_space: int = 0,
+    require_same_masks: bool = True,
+    mask_dropout: float = 0.0,
+) -> torch.Tensor:
+    """
+    Computes random mask spans for a given shape
+
+    Args:
+        shape: the the shape for which to compute masks.
+            should be of size 2 where first element is batch size and 2nd is timesteps
+        padding_mask: optional padding mask of the same size as shape, which will prevent masking padded elements
+        mask_prob: probability for each token to be chosen as start of the span to be masked. this will be multiplied by
+            number of timesteps divided by length of mask span to mask approximately this percentage of all elements.
+            however due to overlaps, the actual number will be smaller (unless no_overlap is True)
+        mask_type: how to compute mask lengths
+            static = fixed size
+            uniform = sample from uniform distribution [mask_other, mask_length*2]
+            normal = sample from normal distribution with mean mask_length and stdev mask_other. mask is min 1 element
+            poisson = sample from possion distribution with lambda = mask length
+        min_masks: minimum number of masked spans
+        no_overlap: if false, will switch to an alternative recursive algorithm that prevents spans from overlapping
+        min_space: only used if no_overlap is True, this is how many elements to keep unmasked between spans
+        require_same_masks: if true, will randomly drop out masks until same amount of masks remains in each sample
+        mask_dropout: randomly dropout this percentage of masks in each example
+    """
+
+    bsz, all_sz = shape
+    mask = torch.full((bsz, all_sz), False)
+
+    all_num_mask = int(
+        # add a random number for probabilistic rounding
+        mask_prob * all_sz / float(mask_length)
+        + torch.rand([1]).item()
+    )
+
+    all_num_mask = max(min_masks, all_num_mask)
+
+    mask_idcs = []
+    for i in range(bsz):
+        if padding_mask is not None:
+            sz = all_sz - padding_mask[i].long().sum().item()
+            num_mask = int(mask_prob * sz / float(mask_length) + np.random.rand())
+            num_mask = max(min_masks, num_mask)
+        else:
+            sz = all_sz
+            num_mask = all_num_mask
+
+        if mask_type == "static":
+            lengths = torch.full([num_mask], mask_length)
+        elif mask_type == "uniform":
+            lengths = torch.randint(mask_other, mask_length * 2 + 1, size=[num_mask])
+        elif mask_type == "normal":
+            lengths = torch.normal(mask_length, mask_other, size=[num_mask])
+            lengths = [max(1, int(round(x))) for x in lengths]
+        else:
+            raise Exception("unknown mask selection " + mask_type)
+
+        if sum(lengths) == 0:
+            lengths[0] = min(mask_length, sz - 1)
+
+        if no_overlap:
+            mask_idc = []
+
+            def arrange(s, e, length, keep_length):
+                span_start = torch.randint(low=s, high=e - length, size=[1]).item()
+                mask_idc.extend(span_start + i for i in range(length))
+
+                new_parts = []
+                if span_start - s - min_space >= keep_length:
+                    new_parts.append((s, span_start - min_space + 1))
+                if e - span_start - length - min_space > keep_length:
+                    new_parts.append((span_start + length + min_space, e))
+                return new_parts
+
+            parts = [(0, sz)]
+            min_length = min(lengths)
+            for length in sorted(lengths, reverse=True):
+                t = [e - s if e - s >= length + min_space else 0 for s, e in parts]
+                lens = torch.asarray(t, dtype=torch.int)
+                l_sum = torch.sum(lens)
+                if l_sum == 0:
+                    break
+                probs = lens / torch.sum(lens)
+                c = torch.multinomial(probs.float(), len(parts)).item()
+                s, e = parts.pop(c)
+                parts.extend(arrange(s, e, length, min_length))
+            mask_idc = torch.asarray(mask_idc)
+        else:
+            min_len = min(lengths)
+            if sz - min_len <= num_mask:
+                min_len = sz - num_mask - 1
+            mask_idc = torch.asarray(
+                random.sample([i for i in range(sz - min_len)], num_mask)
+            )
+            mask_idc = torch.asarray(
+                [
+                    mask_idc[j] + offset
+                    for j in range(len(mask_idc))
+                    for offset in range(lengths[j])
+                ]
+            )
+
+        mask_idcs.append(torch.unique(mask_idc[mask_idc < sz]))
+
+    min_len = min([len(m) for m in mask_idcs])
+    for i, mask_idc in enumerate(mask_idcs):
+        if isinstance(mask_idc, torch.Tensor):
+            mask_idc = torch.asarray(mask_idc, dtype=torch.float)
+        if len(mask_idc) > min_len and require_same_masks:
+            mask_idc = torch.asarray(
+                random.sample([i for i in range(mask_idc)], min_len)
+            )
+        if mask_dropout > 0:
+            num_holes = int(round(len(mask_idc) * mask_dropout))
+            mask_idc = torch.asarray(
+                random.sample([i for i in range(mask_idc)], len(mask_idc) - num_holes)
+            )
+
+        mask[i, mask_idc.int()] = True
+
+    return mask
+
+
+def apply_mask(self, x, padding_mask, target_list):
+    B, T, C = x.shape
+    torch.zeros_like(x)
+    if self.mask_prob > 0:
+        mask_indices = compute_mask_indices(
+            (B, T),
+            padding_mask,
+            self.mask_prob,
+            self.mask_length,
+            self.mask_selection,
+            self.mask_other,
+            min_masks=2,
+            no_overlap=self.no_mask_overlap,
+            min_space=self.mask_min_space,
+        )
+        mask_indices = mask_indices.to(x.device)
+        x[mask_indices] = self.mask_emb
+    else:
+        mask_indices = None
+
+    if self.mask_channel_prob > 0:
+        mask_channel_indices = compute_mask_indices(
+            (B, C),
+            None,
+            self.mask_channel_prob,
+            self.mask_channel_length,
+            self.mask_channel_selection,
+            self.mask_channel_other,
+            no_overlap=self.no_mask_channel_overlap,
+            min_space=self.mask_channel_min_space,
+        )
+        mask_channel_indices = (
+            mask_channel_indices.to(x.device).unsqueeze(1).expand(-1, T, -1)
+        )
+        x[mask_channel_indices] = 0
+
+    return x, mask_indices
+
+
+def get_hubert_model(
+    model_path="assets/hubert/hubert_base.pt", device=torch.device("cpu")
+):
+    models, _, _ = load_model_ensemble_and_task(
+        [model_path],
+        suffix="",
+    )
+    hubert_model = models[0]
+    hubert_model = hubert_model.to(device)
+
+    def _apply_mask(x, padding_mask, target_list):
+        return apply_mask(hubert_model, x, padding_mask, target_list)
+
+    hubert_model.apply_mask = _apply_mask
+
+    def _extract_features(
+        x,
+        padding_mask=None,
+        tgt_layer=None,
+        min_layer=0,
+    ):
+        return extract_features(
+            hubert_model.encoder,
+            x,
+            padding_mask=padding_mask,
+            tgt_layer=tgt_layer,
+            min_layer=min_layer,
+        )
+
+    hubert_model.encoder.extract_features = _extract_features
+
+    hubert_model._forward = hubert_model.forward
+
+    def hubert_extract_features(
+        self,
+        source: torch.Tensor,
+        padding_mask: Optional[torch.Tensor] = None,
+        mask: bool = False,
+        ret_conv: bool = False,
+        output_layer: Optional[int] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        res = self._forward(
+            source,
+            padding_mask=padding_mask,
+            mask=mask,
+            features_only=True,
+            output_layer=output_layer,
+        )
+        feature = res["features"] if ret_conv else res["x"]
+        return feature, res["padding_mask"]
+
+    def _hubert_extract_features(
+        source: torch.Tensor,
+        padding_mask: Optional[torch.Tensor] = None,
+        mask: bool = False,
+        ret_conv: bool = False,
+        output_layer: Optional[int] = None,
+    ) -> Tuple[torch.Tensor, torch.Tensor]:
+        return hubert_extract_features(
+            hubert_model, source, padding_mask, mask, ret_conv, output_layer
+        )
+
+    hubert_model.extract_features = _hubert_extract_features
+
+    def infer(source, padding_mask, output_layer: torch.Tensor):
+        output_layer = output_layer.item()
+        logits = hubert_model.extract_features(
+            source=source, padding_mask=padding_mask, output_layer=output_layer
+        )
+        feats = hubert_model.final_proj(logits[0]) if output_layer == 9 else logits[0]
+        return feats
+
+    hubert_model.infer = infer
+    # hubert_model.forward=infer
+    # hubert_model.forward
+
+    return hubert_model

rvc/lib/jit/get_rmvpe.py

@@ -0,0 +1,12 @@
+import torch
+
+
+def get_rmvpe(model_path="assets/rmvpe/rmvpe.pt", device=torch.device("cpu")):
+    from infer.lib.rmvpe import E2E
+
+    model = E2E(4, 1, (2, 2))
+    ckpt = torch.load(model_path, map_location=device)
+    model.load_state_dict(ckpt)
+    model.eval()
+    model = model.to(device)
+    return model

rvc/lib/jit/get_synthesizer.py

@@ -0,0 +1,37 @@
+import torch
+
+
+def get_synthesizer(pth_path, device=torch.device("cpu")):
+    from infer.lib.infer_pack.models import (
+        SynthesizerTrnMs256NSFsid,
+        SynthesizerTrnMs256NSFsid_nono,
+        SynthesizerTrnMs768NSFsid,
+        SynthesizerTrnMs768NSFsid_nono,
+    )
+
+    cpt = torch.load(pth_path, map_location=torch.device("cpu"))
+    # tgt_sr = cpt["config"][-1]
+    cpt["config"][-3] = cpt["weight"]["emb_g.weight"].shape[0]
+    if_f0 = cpt.get("f0", 1)
+    version = cpt.get("version", "v1")
+    if version == "v1":
+        if if_f0 == 1:
+            net_g = SynthesizerTrnMs256NSFsid(*cpt["config"], is_half=False)
+        else:
+            net_g = SynthesizerTrnMs256NSFsid_nono(*cpt["config"])
+    elif version == "v2":
+        if if_f0 == 1:
+            net_g = SynthesizerTrnMs768NSFsid(*cpt["config"], is_half=False)
+        else:
+            net_g = SynthesizerTrnMs768NSFsid_nono(*cpt["config"])
+    del net_g.enc_q
+    # net_g.forward = net_g.infer
+    # ckpt = {}
+    # ckpt["config"] = cpt["config"]
+    # ckpt["f0"] = if_f0
+    # ckpt["version"] = version
+    # ckpt["info"] = cpt.get("info", "0epoch")
+    net_g.load_state_dict(cpt["weight"], strict=False)
+    net_g = net_g.float()
+    net_g.eval().to(device)
+    return net_g, cpt

rvc/lib/rmvpe.py

@@ -0,0 +1,665 @@
+import os
+from io import BytesIO
+from typing import List, Optional, Tuple
+
+import numpy as np
+import torch
+
+from rvc.lib import jit
+
+try:
+    # Fix "Torch not compiled with CUDA enabled"
+    import intel_extension_for_pytorch as ipex  # pylint: disable=import-error, unused-import
+
+    if torch.xpu.is_available():
+        from rvc.lib.ipex import ipex_init
+
+        ipex_init()
+except Exception:  # pylint: disable=broad-exception-caught
+    pass
+import logging
+from time import time as ttime
+
+import torch.nn as nn
+import torch.nn.functional as F
+from librosa.filters import mel
+from librosa.util import normalize, pad_center, tiny
+from scipy.signal import get_window
+
+logger = logging.getLogger(__name__)
+
+
+class STFT(torch.nn.Module):
+    def __init__(
+        self, filter_length=1024, hop_length=512, win_length=None, window="hann"
+    ):
+        """
+        This module implements an STFT using 1D convolution and 1D transpose convolutions.
+        This is a bit tricky so there are some cases that probably won't work as working
+        out the same sizes before and after in all overlap add setups is tough. Right now,
+        this code should work with hop lengths that are half the filter length (50% overlap
+        between frames).
+
+        Keyword Arguments:
+            filter_length {int} -- Length of filters used (default: {1024})
+            hop_length {int} -- Hop length of STFT (restrict to 50% overlap between frames) (default: {512})
+            win_length {[type]} -- Length of the window function applied to each frame (if not specified, it
+                equals the filter length). (default: {None})
+            window {str} -- Type of window to use (options are bartlett, hann, hamming, blackman, blackmanharris)
+                (default: {'hann'})
+        """
+        super(STFT, self).__init__()
+        self.filter_length = filter_length
+        self.hop_length = hop_length
+        self.win_length = win_length if win_length else filter_length
+        self.window = window
+        self.forward_transform = None
+        self.pad_amount = int(self.filter_length / 2)
+        fourier_basis = np.fft.fft(np.eye(self.filter_length))
+
+        cutoff = int((self.filter_length / 2 + 1))
+        fourier_basis = np.vstack(
+            [np.real(fourier_basis[:cutoff, :]), np.imag(fourier_basis[:cutoff, :])]
+        )
+        forward_basis = torch.FloatTensor(fourier_basis)
+        inverse_basis = torch.FloatTensor(np.linalg.pinv(fourier_basis))
+
+        assert filter_length >= self.win_length
+        # get window and zero center pad it to filter_length
+        fft_window = get_window(window, self.win_length, fftbins=True)
+        fft_window = pad_center(fft_window, size=filter_length)
+        fft_window = torch.from_numpy(fft_window).float()
+
+        # window the bases
+        forward_basis *= fft_window
+        inverse_basis = (inverse_basis.T * fft_window).T
+
+        self.register_buffer("forward_basis", forward_basis.float())
+        self.register_buffer("inverse_basis", inverse_basis.float())
+        self.register_buffer("fft_window", fft_window.float())
+
+    def transform(self, input_data, return_phase=False):
+        """Take input data (audio) to STFT domain.
+
+        Arguments:
+            input_data {tensor} -- Tensor of floats, with shape (num_batch, num_samples)
+
+        Returns:
+            magnitude {tensor} -- Magnitude of STFT with shape (num_batch,
+                num_frequencies, num_frames)
+            phase {tensor} -- Phase of STFT with shape (num_batch,
+                num_frequencies, num_frames)
+        """
+        input_data = F.pad(
+            input_data,
+            (self.pad_amount, self.pad_amount),
+            mode="reflect",
+        )
+        forward_transform = input_data.unfold(
+            1, self.filter_length, self.hop_length
+        ).permute(0, 2, 1)
+        forward_transform = torch.matmul(self.forward_basis, forward_transform)
+        cutoff = int((self.filter_length / 2) + 1)
+        real_part = forward_transform[:, :cutoff, :]
+        imag_part = forward_transform[:, cutoff:, :]
+        magnitude = torch.sqrt(real_part**2 + imag_part**2)
+        if return_phase:
+            phase = torch.atan2(imag_part.data, real_part.data)
+            return magnitude, phase
+        else:
+            return magnitude
+
+    def inverse(self, magnitude, phase):
+        """Call the inverse STFT (iSTFT), given magnitude and phase tensors produced
+        by the ```transform``` function.
+
+        Arguments:
+            magnitude {tensor} -- Magnitude of STFT with shape (num_batch,
+                num_frequencies, num_frames)
+            phase {tensor} -- Phase of STFT with shape (num_batch,
+                num_frequencies, num_frames)
+
+        Returns:
+            inverse_transform {tensor} -- Reconstructed audio given magnitude and phase. Of
+                shape (num_batch, num_samples)
+        """
+        cat = torch.cat(
+            [magnitude * torch.cos(phase), magnitude * torch.sin(phase)], dim=1
+        )
+        fold = torch.nn.Fold(
+            output_size=(1, (cat.size(-1) - 1) * self.hop_length + self.filter_length),
+            kernel_size=(1, self.filter_length),
+            stride=(1, self.hop_length),
+        )
+        inverse_transform = torch.matmul(self.inverse_basis, cat)
+        inverse_transform = fold(inverse_transform)[
+            :, 0, 0, self.pad_amount : -self.pad_amount
+        ]
+        window_square_sum = (
+            self.fft_window.pow(2).repeat(cat.size(-1), 1).T.unsqueeze(0)
+        )
+        window_square_sum = fold(window_square_sum)[
+            :, 0, 0, self.pad_amount : -self.pad_amount
+        ]
+        inverse_transform /= window_square_sum
+        return inverse_transform
+
+    def forward(self, input_data):
+        """Take input data (audio) to STFT domain and then back to audio.
+
+        Arguments:
+            input_data {tensor} -- Tensor of floats, with shape (num_batch, num_samples)
+
+        Returns:
+            reconstruction {tensor} -- Reconstructed audio given magnitude and phase. Of
+                shape (num_batch, num_samples)
+        """
+        self.magnitude, self.phase = self.transform(input_data, return_phase=True)
+        reconstruction = self.inverse(self.magnitude, self.phase)
+        return reconstruction
+
+
+class BiGRU(nn.Module):
+    def __init__(self, input_features, hidden_features, num_layers):
+        super(BiGRU, self).__init__()
+        self.gru = nn.GRU(
+            input_features,
+            hidden_features,
+            num_layers=num_layers,
+            batch_first=True,
+            bidirectional=True,
+        )
+
+    def forward(self, x):
+        return self.gru(x)[0]
+
+
+class ConvBlockRes(nn.Module):
+    def __init__(self, in_channels, out_channels, momentum=0.01):
+        super(ConvBlockRes, self).__init__()
+        self.conv = nn.Sequential(
+            nn.Conv2d(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=(1, 1),
+                padding=(1, 1),
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+            nn.Conv2d(
+                in_channels=out_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=(1, 1),
+                padding=(1, 1),
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+        )
+        # self.shortcut:Optional[nn.Module] = None
+        if in_channels != out_channels:
+            self.shortcut = nn.Conv2d(in_channels, out_channels, (1, 1))
+
+    def forward(self, x: torch.Tensor):
+        if not hasattr(self, "shortcut"):
+            return self.conv(x) + x
+        else:
+            return self.conv(x) + self.shortcut(x)
+
+
+class Encoder(nn.Module):
+    def __init__(
+        self,
+        in_channels,
+        in_size,
+        n_encoders,
+        kernel_size,
+        n_blocks,
+        out_channels=16,
+        momentum=0.01,
+    ):
+        super(Encoder, self).__init__()
+        self.n_encoders = n_encoders
+        self.bn = nn.BatchNorm2d(in_channels, momentum=momentum)
+        self.layers = nn.ModuleList()
+        self.latent_channels = []
+        for i in range(self.n_encoders):
+            self.layers.append(
+                ResEncoderBlock(
+                    in_channels, out_channels, kernel_size, n_blocks, momentum=momentum
+                )
+            )
+            self.latent_channels.append([out_channels, in_size])
+            in_channels = out_channels
+            out_channels *= 2
+            in_size //= 2
+        self.out_size = in_size
+        self.out_channel = out_channels
+
+    def forward(self, x: torch.Tensor):
+        concat_tensors: List[torch.Tensor] = []
+        x = self.bn(x)
+        for i, layer in enumerate(self.layers):
+            t, x = layer(x)
+            concat_tensors.append(t)
+        return x, concat_tensors
+
+
+class ResEncoderBlock(nn.Module):
+    def __init__(
+        self, in_channels, out_channels, kernel_size, n_blocks=1, momentum=0.01
+    ):
+        super(ResEncoderBlock, self).__init__()
+        self.n_blocks = n_blocks
+        self.conv = nn.ModuleList()
+        self.conv.append(ConvBlockRes(in_channels, out_channels, momentum))
+        for i in range(n_blocks - 1):
+            self.conv.append(ConvBlockRes(out_channels, out_channels, momentum))
+        self.kernel_size = kernel_size
+        if self.kernel_size is not None:
+            self.pool = nn.AvgPool2d(kernel_size=kernel_size)
+
+    def forward(self, x):
+        for i, conv in enumerate(self.conv):
+            x = conv(x)
+        if self.kernel_size is not None:
+            return x, self.pool(x)
+        else:
+            return x
+
+
+class Intermediate(nn.Module):  #
+    def __init__(self, in_channels, out_channels, n_inters, n_blocks, momentum=0.01):
+        super(Intermediate, self).__init__()
+        self.n_inters = n_inters
+        self.layers = nn.ModuleList()
+        self.layers.append(
+            ResEncoderBlock(in_channels, out_channels, None, n_blocks, momentum)
+        )
+        for i in range(self.n_inters - 1):
+            self.layers.append(
+                ResEncoderBlock(out_channels, out_channels, None, n_blocks, momentum)
+            )
+
+    def forward(self, x):
+        for i, layer in enumerate(self.layers):
+            x = layer(x)
+        return x
+
+
+class ResDecoderBlock(nn.Module):
+    def __init__(self, in_channels, out_channels, stride, n_blocks=1, momentum=0.01):
+        super(ResDecoderBlock, self).__init__()
+        out_padding = (0, 1) if stride == (1, 2) else (1, 1)
+        self.n_blocks = n_blocks
+        self.conv1 = nn.Sequential(
+            nn.ConvTranspose2d(
+                in_channels=in_channels,
+                out_channels=out_channels,
+                kernel_size=(3, 3),
+                stride=stride,
+                padding=(1, 1),
+                output_padding=out_padding,
+                bias=False,
+            ),
+            nn.BatchNorm2d(out_channels, momentum=momentum),
+            nn.ReLU(),
+        )
+        self.conv2 = nn.ModuleList()
+        self.conv2.append(ConvBlockRes(out_channels * 2, out_channels, momentum))
+        for i in range(n_blocks - 1):
+            self.conv2.append(ConvBlockRes(out_channels, out_channels, momentum))
+
+    def forward(self, x, concat_tensor):
+        x = self.conv1(x)
+        x = torch.cat((x, concat_tensor), dim=1)
+        for i, conv2 in enumerate(self.conv2):
+            x = conv2(x)
+        return x
+
+
+class Decoder(nn.Module):
+    def __init__(self, in_channels, n_decoders, stride, n_blocks, momentum=0.01):
+        super(Decoder, self).__init__()
+        self.layers = nn.ModuleList()
+        self.n_decoders = n_decoders
+        for i in range(self.n_decoders):
+            out_channels = in_channels // 2
+            self.layers.append(
+                ResDecoderBlock(in_channels, out_channels, stride, n_blocks, momentum)
+            )
+            in_channels = out_channels
+
+    def forward(self, x: torch.Tensor, concat_tensors: List[torch.Tensor]):
+        for i, layer in enumerate(self.layers):
+            x = layer(x, concat_tensors[-1 - i])
+        return x
+
+
+class DeepUnet(nn.Module):
+    def __init__(
+        self,
+        kernel_size,
+        n_blocks,
+        en_de_layers=5,
+        inter_layers=4,
+        in_channels=1,
+        en_out_channels=16,
+    ):
+        super(DeepUnet, self).__init__()
+        self.encoder = Encoder(
+            in_channels, 128, en_de_layers, kernel_size, n_blocks, en_out_channels
+        )
+        self.intermediate = Intermediate(
+            self.encoder.out_channel // 2,
+            self.encoder.out_channel,
+            inter_layers,
+            n_blocks,
+        )
+        self.decoder = Decoder(
+            self.encoder.out_channel, en_de_layers, kernel_size, n_blocks
+        )
+
+    def forward(self, x: torch.Tensor) -> torch.Tensor:
+        x, concat_tensors = self.encoder(x)
+        x = self.intermediate(x)
+        x = self.decoder(x, concat_tensors)
+        return x
+
+
+class E2E(nn.Module):
+    def __init__(
+        self,
+        n_blocks,
+        n_gru,
+        kernel_size,
+        en_de_layers=5,
+        inter_layers=4,
+        in_channels=1,
+        en_out_channels=16,
+    ):
+        super(E2E, self).__init__()
+        self.unet = DeepUnet(
+            kernel_size,
+            n_blocks,
+            en_de_layers,
+            inter_layers,
+            in_channels,
+            en_out_channels,
+        )
+        self.cnn = nn.Conv2d(en_out_channels, 3, (3, 3), padding=(1, 1))
+        if n_gru:
+            self.fc = nn.Sequential(
+                BiGRU(3 * 128, 256, n_gru),
+                nn.Linear(512, 360),
+                nn.Dropout(0.25),
+                nn.Sigmoid(),
+            )
+        else:
+            self.fc = nn.Sequential(
+                nn.Linear(3 * nn.N_MELS, nn.N_CLASS), nn.Dropout(0.25), nn.Sigmoid()
+            )
+
+    def forward(self, mel):
+        # print(mel.shape)
+        mel = mel.transpose(-1, -2).unsqueeze(1)
+        x = self.cnn(self.unet(mel)).transpose(1, 2).flatten(-2)
+        x = self.fc(x)
+        # print(x.shape)
+        return x
+
+
+class MelSpectrogram(torch.nn.Module):
+    def __init__(
+        self,
+        is_half,
+        n_mel_channels,
+        sampling_rate,
+        win_length,
+        hop_length,
+        n_fft=None,
+        mel_fmin=0,
+        mel_fmax=None,
+        clamp=1e-5,
+    ):
+        super().__init__()
+        n_fft = win_length if n_fft is None else n_fft
+        self.hann_window = {}
+        mel_basis = mel(
+            sr=sampling_rate,
+            n_fft=n_fft,
+            n_mels=n_mel_channels,
+            fmin=mel_fmin,
+            fmax=mel_fmax,
+            htk=True,
+        )
+        mel_basis = torch.from_numpy(mel_basis).float()
+        self.register_buffer("mel_basis", mel_basis)
+        self.n_fft = win_length if n_fft is None else n_fft
+        self.hop_length = hop_length
+        self.win_length = win_length
+        self.sampling_rate = sampling_rate
+        self.n_mel_channels = n_mel_channels
+        self.clamp = clamp
+        self.is_half = is_half
+
+    def forward(self, audio, keyshift=0, speed=1, center=True):
+        factor = 2 ** (keyshift / 12)
+        n_fft_new = int(np.round(self.n_fft * factor))
+        win_length_new = int(np.round(self.win_length * factor))
+        hop_length_new = int(np.round(self.hop_length * speed))
+        keyshift_key = str(keyshift) + "_" + str(audio.device)
+        if keyshift_key not in self.hann_window:
+            self.hann_window[keyshift_key] = torch.hann_window(win_length_new).to(
+                audio.device
+            )
+        if "privateuseone" in str(audio.device):
+            if not hasattr(self, "stft"):
+                self.stft = STFT(
+                    filter_length=n_fft_new,
+                    hop_length=hop_length_new,
+                    win_length=win_length_new,
+                    window="hann",
+                ).to(audio.device)
+            magnitude = self.stft.transform(audio)
+        else:
+            fft = torch.stft(
+                audio,
+                n_fft=n_fft_new,
+                hop_length=hop_length_new,
+                win_length=win_length_new,
+                window=self.hann_window[keyshift_key],
+                center=center,
+                return_complex=True,
+            )
+            magnitude = torch.sqrt(fft.real.pow(2) + fft.imag.pow(2))
+        if keyshift != 0:
+            size = self.n_fft // 2 + 1
+            resize = magnitude.size(1)
+            if resize < size:
+                magnitude = F.pad(magnitude, (0, 0, 0, size - resize))
+            magnitude = magnitude[:, :size, :] * self.win_length / win_length_new
+        mel_output = torch.matmul(self.mel_basis, magnitude)
+        if self.is_half == True:
+            mel_output = mel_output.half()
+        log_mel_spec = torch.log(torch.clamp(mel_output, min=self.clamp))
+        return log_mel_spec
+
+
+class RMVPE:
+    def __init__(self, model_path: str, is_half, device=None, use_jit=False):
+        self.resample_kernel = {}
+        self.resample_kernel = {}
+        self.is_half = is_half
+        if device is None:
+            device = "cuda:0" if torch.cuda.is_available() else "cpu"
+        self.device = device
+        self.mel_extractor = MelSpectrogram(
+            is_half, 128, 16000, 1024, 160, None, 30, 8000
+        ).to(device)
+        if "privateuseone" in str(device):
+            import onnxruntime as ort
+
+            ort_session = ort.InferenceSession(
+                "%s/rmvpe.onnx" % os.environ["rmvpe_root"],
+                providers=["DmlExecutionProvider"],
+            )
+            self.model = ort_session
+        else:
+            if str(self.device) == "cuda":
+                self.device = torch.device("cuda:0")
+
+            def get_jit_model():
+                jit_model_path = model_path.rstrip(".pth")
+                jit_model_path += ".half.jit" if is_half else ".jit"
+                reload = False
+                if os.path.exists(jit_model_path):
+                    ckpt = jit.load(jit_model_path)
+                    model_device = ckpt["device"]
+                    if model_device != str(self.device):
+                        reload = True
+                else:
+                    reload = True
+
+                if reload:
+                    ckpt = jit.rmvpe_jit_export(
+                        model_path=model_path,
+                        mode="script",
+                        inputs_path=None,
+                        save_path=jit_model_path,
+                        device=device,
+                        is_half=is_half,
+                    )
+                model = torch.jit.load(BytesIO(ckpt["model"]), map_location=device)
+                return model
+
+            def get_default_model():
+                model = E2E(4, 1, (2, 2))
+                ckpt = torch.load(model_path, map_location="cpu")
+                model.load_state_dict(ckpt)
+                model.eval()
+                if is_half:
+                    model = model.half()
+                else:
+                    model = model.float()
+                return model
+
+            if use_jit:
+                if is_half and "cpu" in str(self.device):
+                    logger.warning(
+                        "Use default rmvpe model. \
+                                 Jit is not supported on the CPU for half floating point"
+                    )
+                    self.model = get_default_model()
+                else:
+                    self.model = get_jit_model()
+            else:
+                self.model = get_default_model()
+
+            self.model = self.model.to(device)
+        cents_mapping = 20 * np.arange(360) + 1997.3794084376191
+        self.cents_mapping = np.pad(cents_mapping, (4, 4))  # 368
+
+    def mel2hidden(self, mel):
+        with torch.no_grad():
+            n_frames = mel.shape[-1]
+            n_pad = 32 * ((n_frames - 1) // 32 + 1) - n_frames
+            if n_pad > 0:
+                mel = F.pad(mel, (0, n_pad), mode="constant")
+            if "privateuseone" in str(self.device):
+                onnx_input_name = self.model.get_inputs()[0].name
+                onnx_outputs_names = self.model.get_outputs()[0].name
+                hidden = self.model.run(
+                    [onnx_outputs_names],
+                    input_feed={onnx_input_name: mel.cpu().numpy()},
+                )[0]
+            else:
+                mel = mel.half() if self.is_half else mel.float()
+                hidden = self.model(mel)
+            return hidden[:, :n_frames]
+
+    def decode(self, hidden, thred=0.03):
+        cents_pred = self.to_local_average_cents(hidden, thred=thred)
+        f0 = 10 * (2 ** (cents_pred / 1200))
+        f0[f0 == 10] = 0
+        # f0 = np.array([10 * (2 ** (cent_pred / 1200)) if cent_pred else 0 for cent_pred in cents_pred])
+        return f0
+
+    def infer_from_audio(self, audio, thred=0.03):
+        # torch.cuda.synchronize()
+        t0 = ttime()
+        mel = self.mel_extractor(
+            torch.from_numpy(audio).float().to(self.device).unsqueeze(0), center=True
+        )
+        # print(123123123,mel.device.type)
+        # torch.cuda.synchronize()
+        t1 = ttime()
+        hidden = self.mel2hidden(mel)
+        # torch.cuda.synchronize()
+        t2 = ttime()
+        # print(234234,hidden.device.type)
+        if "privateuseone" not in str(self.device):
+            hidden = hidden.squeeze(0).cpu().numpy()
+        else:
+            hidden = hidden[0]
+        if self.is_half == True:
+            hidden = hidden.astype("float32")
+
+        f0 = self.decode(hidden, thred=thred)
+        # torch.cuda.synchronize()
+        t3 = ttime()
+        # print("hmvpe:%s\t%s\t%s\t%s"%(t1-t0,t2-t1,t3-t2,t3-t0))
+        return f0
+
+    def to_local_average_cents(self, salience, thred=0.05):
+        # t0 = ttime()
+        center = np.argmax(salience, axis=1)  # 帧长#index
+        salience = np.pad(salience, ((0, 0), (4, 4)))  # 帧长,368
+        # t1 = ttime()
+        center += 4
+        todo_salience = []
+        todo_cents_mapping = []
+        starts = center - 4
+        ends = center + 5
+        for idx in range(salience.shape[0]):
+            todo_salience.append(salience[:, starts[idx] : ends[idx]][idx])
+            todo_cents_mapping.append(self.cents_mapping[starts[idx] : ends[idx]])
+        # t2 = ttime()
+        todo_salience = np.array(todo_salience)  # 帧长，9
+        todo_cents_mapping = np.array(todo_cents_mapping)  # 帧长，9
+        product_sum = np.sum(todo_salience * todo_cents_mapping, 1)
+        weight_sum = np.sum(todo_salience, 1)  # 帧长
+        devided = product_sum / weight_sum  # 帧长
+        # t3 = ttime()
+        maxx = np.max(salience, axis=1)  # 帧长
+        devided[maxx <= thred] = 0
+        # t4 = ttime()
+        # print("decode:%s\t%s\t%s\t%s" % (t1 - t0, t2 - t1, t3 - t2, t4 - t3))
+        return devided
+
+
+if __name__ == "__main__":
+    import librosa
+    import soundfile as sf
+
+    audio, sampling_rate = sf.read(r"C:\Users\liujing04\Desktop\Z\冬之花clip1.wav")
+    if len(audio.shape) > 1:
+        audio = librosa.to_mono(audio.transpose(1, 0))
+    audio_bak = audio.copy()
+    if sampling_rate != 16000:
+        audio = librosa.resample(audio, orig_sr=sampling_rate, target_sr=16000)
+    model_path = r"D:\BaiduNetdiskDownload\RVC-beta-v2-0727AMD_realtime\rmvpe.pt"
+    thred = 0.03  # 0.01
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    rmvpe = RMVPE(model_path, is_half=False, device=device)
+    t0 = ttime()
+    f0 = rmvpe.infer_from_audio(audio, thred=thred)
+    # f0 = rmvpe.infer_from_audio(audio, thred=thred)
+    # f0 = rmvpe.infer_from_audio(audio, thred=thred)
+    # f0 = rmvpe.infer_from_audio(audio, thred=thred)
+    # f0 = rmvpe.infer_from_audio(audio, thred=thred)
+    t1 = ttime()
+    logger.info("%s %.2f", f0.shape, t1 - t0)

rvc/lib/slicer2.py

@@ -0,0 +1,260 @@
+import numpy as np
+
+
+# This function is obtained from librosa.
+def get_rms(
+    y,
+    frame_length=2048,
+    hop_length=512,
+    pad_mode="constant",
+):
+    padding = (int(frame_length // 2), int(frame_length // 2))
+    y = np.pad(y, padding, mode=pad_mode)
+
+    axis = -1
+    # put our new within-frame axis at the end for now
+    out_strides = y.strides + tuple([y.strides[axis]])
+    # Reduce the shape on the framing axis
+    x_shape_trimmed = list(y.shape)
+    x_shape_trimmed[axis] -= frame_length - 1
+    out_shape = tuple(x_shape_trimmed) + tuple([frame_length])
+    xw = np.lib.stride_tricks.as_strided(y, shape=out_shape, strides=out_strides)
+    if axis < 0:
+        target_axis = axis - 1
+    else:
+        target_axis = axis + 1
+    xw = np.moveaxis(xw, -1, target_axis)
+    # Downsample along the target axis
+    slices = [slice(None)] * xw.ndim
+    slices[axis] = slice(0, None, hop_length)
+    x = xw[tuple(slices)]
+
+    # Calculate power
+    power = np.mean(np.abs(x) ** 2, axis=-2, keepdims=True)
+
+    return np.sqrt(power)
+
+
+class Slicer:
+    def __init__(
+        self,
+        sr: int,
+        threshold: float = -40.0,
+        min_length: int = 5000,
+        min_interval: int = 300,
+        hop_size: int = 20,
+        max_sil_kept: int = 5000,
+    ):
+        if not min_length >= min_interval >= hop_size:
+            raise ValueError(
+                "The following condition must be satisfied: min_length >= min_interval >= hop_size"
+            )
+        if not max_sil_kept >= hop_size:
+            raise ValueError(
+                "The following condition must be satisfied: max_sil_kept >= hop_size"
+            )
+        min_interval = sr * min_interval / 1000
+        self.threshold = 10 ** (threshold / 20.0)
+        self.hop_size = round(sr * hop_size / 1000)
+        self.win_size = min(round(min_interval), 4 * self.hop_size)
+        self.min_length = round(sr * min_length / 1000 / self.hop_size)
+        self.min_interval = round(min_interval / self.hop_size)
+        self.max_sil_kept = round(sr * max_sil_kept / 1000 / self.hop_size)
+
+    def _apply_slice(self, waveform, begin, end):
+        if len(waveform.shape) > 1:
+            return waveform[
+                :, begin * self.hop_size : min(waveform.shape[1], end * self.hop_size)
+            ]
+        else:
+            return waveform[
+                begin * self.hop_size : min(waveform.shape[0], end * self.hop_size)
+            ]
+
+    # @timeit
+    def slice(self, waveform):
+        if len(waveform.shape) > 1:
+            samples = waveform.mean(axis=0)
+        else:
+            samples = waveform
+        if samples.shape[0] <= self.min_length:
+            return [waveform]
+        rms_list = get_rms(
+            y=samples, frame_length=self.win_size, hop_length=self.hop_size
+        ).squeeze(0)
+        sil_tags = []
+        silence_start = None
+        clip_start = 0
+        for i, rms in enumerate(rms_list):
+            # Keep looping while frame is silent.
+            if rms < self.threshold:
+                # Record start of silent frames.
+                if silence_start is None:
+                    silence_start = i
+                continue
+            # Keep looping while frame is not silent and silence start has not been recorded.
+            if silence_start is None:
+                continue
+            # Clear recorded silence start if interval is not enough or clip is too short
+            is_leading_silence = silence_start == 0 and i > self.max_sil_kept
+            need_slice_middle = (
+                i - silence_start >= self.min_interval
+                and i - clip_start >= self.min_length
+            )
+            if not is_leading_silence and not need_slice_middle:
+                silence_start = None
+                continue
+            # Need slicing. Record the range of silent frames to be removed.
+            if i - silence_start <= self.max_sil_kept:
+                pos = rms_list[silence_start : i + 1].argmin() + silence_start
+                if silence_start == 0:
+                    sil_tags.append((0, pos))
+                else:
+                    sil_tags.append((pos, pos))
+                clip_start = pos
+            elif i - silence_start <= self.max_sil_kept * 2:
+                pos = rms_list[
+                    i - self.max_sil_kept : silence_start + self.max_sil_kept + 1
+                ].argmin()
+                pos += i - self.max_sil_kept
+                pos_l = (
+                    rms_list[
+                        silence_start : silence_start + self.max_sil_kept + 1
+                    ].argmin()
+                    + silence_start
+                )
+                pos_r = (
+                    rms_list[i - self.max_sil_kept : i + 1].argmin()
+                    + i
+                    - self.max_sil_kept
+                )
+                if silence_start == 0:
+                    sil_tags.append((0, pos_r))
+                    clip_start = pos_r
+                else:
+                    sil_tags.append((min(pos_l, pos), max(pos_r, pos)))
+                    clip_start = max(pos_r, pos)
+            else:
+                pos_l = (
+                    rms_list[
+                        silence_start : silence_start + self.max_sil_kept + 1
+                    ].argmin()
+                    + silence_start
+                )
+                pos_r = (
+                    rms_list[i - self.max_sil_kept : i + 1].argmin()
+                    + i
+                    - self.max_sil_kept
+                )
+                if silence_start == 0:
+                    sil_tags.append((0, pos_r))
+                else:
+                    sil_tags.append((pos_l, pos_r))
+                clip_start = pos_r
+            silence_start = None
+        # Deal with trailing silence.
+        total_frames = rms_list.shape[0]
+        if (
+            silence_start is not None
+            and total_frames - silence_start >= self.min_interval
+        ):
+            silence_end = min(total_frames, silence_start + self.max_sil_kept)
+            pos = rms_list[silence_start : silence_end + 1].argmin() + silence_start
+            sil_tags.append((pos, total_frames + 1))
+        # Apply and return slices.
+        if len(sil_tags) == 0:
+            return [waveform]
+        else:
+            chunks = []
+            if sil_tags[0][0] > 0:
+                chunks.append(self._apply_slice(waveform, 0, sil_tags[0][0]))
+            for i in range(len(sil_tags) - 1):
+                chunks.append(
+                    self._apply_slice(waveform, sil_tags[i][1], sil_tags[i + 1][0])
+                )
+            if sil_tags[-1][1] < total_frames:
+                chunks.append(
+                    self._apply_slice(waveform, sil_tags[-1][1], total_frames)
+                )
+            return chunks
+
+
+def main():
+    import os.path
+    from argparse import ArgumentParser
+
+    import librosa
+    import soundfile
+
+    parser = ArgumentParser()
+    parser.add_argument("audio", type=str, help="The audio to be sliced")
+    parser.add_argument(
+        "--out", type=str, help="Output directory of the sliced audio clips"
+    )
+    parser.add_argument(
+        "--db_thresh",
+        type=float,
+        required=False,
+        default=-40,
+        help="The dB threshold for silence detection",
+    )
+    parser.add_argument(
+        "--min_length",
+        type=int,
+        required=False,
+        default=5000,
+        help="The minimum milliseconds required for each sliced audio clip",
+    )
+    parser.add_argument(
+        "--min_interval",
+        type=int,
+        required=False,
+        default=300,
+        help="The minimum milliseconds for a silence part to be sliced",
+    )
+    parser.add_argument(
+        "--hop_size",
+        type=int,
+        required=False,
+        default=10,
+        help="Frame length in milliseconds",
+    )
+    parser.add_argument(
+        "--max_sil_kept",
+        type=int,
+        required=False,
+        default=500,
+        help="The maximum silence length kept around the sliced clip, presented in milliseconds",
+    )
+    args = parser.parse_args()
+    out = args.out
+    if out is None:
+        out = os.path.dirname(os.path.abspath(args.audio))
+    audio, sr = librosa.load(args.audio, sr=None, mono=False)
+    slicer = Slicer(
+        sr=sr,
+        threshold=args.db_thresh,
+        min_length=args.min_length,
+        min_interval=args.min_interval,
+        hop_size=args.hop_size,
+        max_sil_kept=args.max_sil_kept,
+    )
+    chunks = slicer.slice(audio)
+    if not os.path.exists(out):
+        os.makedirs(out)
+    for i, chunk in enumerate(chunks):
+        if len(chunk.shape) > 1:
+            chunk = chunk.T
+        soundfile.write(
+            os.path.join(
+                out,
+                f"%s_%d.wav"
+                % (os.path.basename(args.audio).rsplit(".", maxsplit=1)[0], i),
+            ),
+            chunk,
+            sr,
+        )
+
+
+if __name__ == "__main__":
+    main()

rvc/lib/train/architecture/v1.yml

@@ -0,0 +1,57 @@
+32k:
+    filter_length: 513,
+    a?: 32,
+    inter_channels: 192,
+    hidden_channels: 192,
+    filter_channels: 768,
+    n_heads: 2,
+    kernen_layersl_size: 6,
+    kernel_size: 3,
+    p_dropout: 0,
+    resblock: "1",
+    resblock_kernel_sizes: [3, 7, 11],
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+    upsample_rates:  [10, 4, 2, 2, 2],
+    upsample_initial_channel: 512,
+    upsample_kernel_sizes: [16, 16, 4, 4, 4],
+    spk_embed_dim: 109,
+    gin_channels: 256,
+    sampling_rate: 32000,
+40k:
+    filter_length: 1025,
+    a?: 32, # What?
+    inter_channels: 192,
+    hidden_channels: 192,
+    filter_channels: 768,
+    n_heads: 2,
+    kernen_layersl_size: 6,
+    kernel_size: 3,
+    p_dropout: 0,
+    resblock: "1",
+    resblock_kernel_sizes: [3, 7, 11],
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+    upsample_rates: [10, 10, 2, 2],
+    upsample_initial_channel: 512,
+    upsample_kernel_sizes: [16, 16, 4, 4],
+    spk_embed_dim: 109,
+    gin_channels: 256,
+    sampling_rate:  40000,
+48k:
+    filter_length: 1025,
+    a?: 32,
+    inter_channels: 192,
+    hidden_channels: 192,
+    filter_channels: 768,
+    n_heads: 2,
+    kernen_layersl_size: 6,
+    kernel_size: 3,
+    p_dropout: 0,
+    resblock: "1",
+    resblock_kernel_sizes: [3, 7, 11],
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+    upsample_rates: [10, 6, 2, 2, 2],
+    upsample_initial_channel: 512,
+    upsample_kernel_sizes: [16, 16, 4, 4, 4],
+    spk_embed_dim: 109,
+    gin_channels: 256,
+    sampling_rate: 48000,

rvc/lib/train/architecture/v2.yml

@@ -0,0 +1,38 @@
+32k:
+    filter_length: 513,
+    a?: 32,
+    inter_channels: 192,
+    hidden_channels: 192,
+    filter_channels: 768,
+    n_heads: 2,
+    kernen_layersl_size: 6,
+    kernel_size: 3,
+    p_dropout: 0,
+    resblock: "1",
+    resblock_kernel_sizes: [3, 7, 11],
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+    upsample_rates:  [10, 4, 2, 2, 2],
+    upsample_initial_channel: 512,
+    upsample_kernel_sizes: [16, 16, 4, 4, 4],
+    spk_embed_dim: 109,
+    gin_channels: 256,
+    sampling_rate: 32000,
+48k:
+    filter_length: 1025,
+    a?: 32,
+    inter_channels: 192,
+    hidden_channels: 192,
+    filter_channels: 768,
+    n_heads: 2,
+    kernen_layersl_size: 6,
+    kernel_size: 3,
+    p_dropout: 0,
+    resblock: "1",
+    resblock_kernel_sizes: [3, 7, 11],
+    resblock_dilation_sizes: [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+    upsample_rates: [12, 10, 2, 2],
+    upsample_initial_channel: 512,
+    upsample_kernel_sizes: [24, 20, 4, 4],
+    spk_embed_dim: 109,
+    gin_channels: 256,
+    sampling_rate: 48000,

rvc/lib/train/data_utils.py

@@ -0,0 +1,517 @@
+import logging
+import os
+import traceback
+
+logger = logging.getLogger(__name__)
+
+import numpy as np
+import torch
+import torch.utils.data
+
+from rvc.lib.train.mel_processing import spectrogram_torch
+from rvc.lib.train.utils import load_filepaths_and_text, load_wav_to_torch
+
+
+class TextAudioLoaderMultiNSFsid(torch.utils.data.Dataset):
+    """
+    1) loads audio, text pairs
+    2) normalizes text and converts them to sequences of integers
+    3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths_and_text, hparams):
+        self.audiopaths_and_text = load_filepaths_and_text(audiopaths_and_text)
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length = hparams.filter_length
+        self.hop_length = hparams.hop_length
+        self.win_length = hparams.win_length
+        self.sampling_rate = hparams.sampling_rate
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 5000)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+        audiopaths_and_text_new = []
+        lengths = []
+        for audiopath, text, pitch, pitchf, dv in self.audiopaths_and_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_and_text_new.append([audiopath, text, pitch, pitchf, dv])
+                lengths.append(os.path.getsize(audiopath) // (3 * self.hop_length))
+        self.audiopaths_and_text = audiopaths_and_text_new
+        self.lengths = lengths
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def get_audio_text_pair(self, audiopath_and_text):
+        # separate filename and text
+        file = audiopath_and_text[0]
+        phone = audiopath_and_text[1]
+        pitch = audiopath_and_text[2]
+        pitchf = audiopath_and_text[3]
+        dv = audiopath_and_text[4]
+
+        phone, pitch, pitchf = self.get_labels(phone, pitch, pitchf)
+        spec, wav = self.get_audio(file)
+        dv = self.get_sid(dv)
+
+        len_phone = phone.size()[0]
+        len_spec = spec.size()[-1]
+        # print(123,phone.shape,pitch.shape,spec.shape)
+        if len_phone != len_spec:
+            len_min = min(len_phone, len_spec)
+            # amor
+            len_wav = len_min * self.hop_length
+
+            spec = spec[:, :len_min]
+            wav = wav[:, :len_wav]
+
+            phone = phone[:len_min, :]
+            pitch = pitch[:len_min]
+            pitchf = pitchf[:len_min]
+
+        return (spec, wav, phone, pitch, pitchf, dv)
+
+    def get_labels(self, phone, pitch, pitchf):
+        phone = np.load(phone)
+        phone = np.repeat(phone, 2, axis=0)
+        pitch = np.load(pitch)
+        pitchf = np.load(pitchf)
+        n_num = min(phone.shape[0], 900)  # DistributedBucketSampler
+        # print(234,phone.shape,pitch.shape)
+        phone = phone[:n_num, :]
+        pitch = pitch[:n_num]
+        pitchf = pitchf[:n_num]
+        phone = torch.FloatTensor(phone)
+        pitch = torch.LongTensor(pitch)
+        pitchf = torch.FloatTensor(pitchf)
+        return phone, pitch, pitchf
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError(
+                "{} SR doesn't match target {} SR".format(
+                    sampling_rate, self.sampling_rate
+                )
+            )
+        audio_norm = audio
+        #        audio_norm = audio / self.max_wav_value
+        #        audio_norm = audio / np.abs(audio).max()
+
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            try:
+                spec = torch.load(spec_filename)
+            except:
+                logger.warning("%s %s", spec_filename, traceback.format_exc())
+                spec = spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    center=False,
+                )
+                spec = torch.squeeze(spec, 0)
+                torch.save(spec, spec_filename, _use_new_zipfile_serialization=False)
+        else:
+            spec = spectrogram_torch(
+                audio_norm,
+                self.filter_length,
+                self.sampling_rate,
+                self.hop_length,
+                self.win_length,
+                center=False,
+            )
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename, _use_new_zipfile_serialization=False)
+        return spec, audio_norm
+
+    def __getitem__(self, index):
+        return self.get_audio_text_pair(self.audiopaths_and_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_and_text)
+
+
+class TextAudioCollateMultiNSFsid:
+    """Zero-pads model inputs and targets"""
+
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text and aduio
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[0].size(1) for x in batch]), dim=0, descending=True
+        )
+
+        max_spec_len = max([x[0].size(1) for x in batch])
+        max_wave_len = max([x[1].size(1) for x in batch])
+        spec_lengths = torch.LongTensor(len(batch))
+        wave_lengths = torch.LongTensor(len(batch))
+        spec_padded = torch.FloatTensor(len(batch), batch[0][0].size(0), max_spec_len)
+        wave_padded = torch.FloatTensor(len(batch), 1, max_wave_len)
+        spec_padded.zero_()
+        wave_padded.zero_()
+
+        max_phone_len = max([x[2].size(0) for x in batch])
+        phone_lengths = torch.LongTensor(len(batch))
+        phone_padded = torch.FloatTensor(
+            len(batch), max_phone_len, batch[0][2].shape[1]
+        )  # (spec, wav, phone, pitch)
+        pitch_padded = torch.LongTensor(len(batch), max_phone_len)
+        pitchf_padded = torch.FloatTensor(len(batch), max_phone_len)
+        phone_padded.zero_()
+        pitch_padded.zero_()
+        pitchf_padded.zero_()
+        # dv = torch.FloatTensor(len(batch), 256)#gin=256
+        sid = torch.LongTensor(len(batch))
+
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            spec = row[0]
+            spec_padded[i, :, : spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wave = row[1]
+            wave_padded[i, :, : wave.size(1)] = wave
+            wave_lengths[i] = wave.size(1)
+
+            phone = row[2]
+            phone_padded[i, : phone.size(0), :] = phone
+            phone_lengths[i] = phone.size(0)
+
+            pitch = row[3]
+            pitch_padded[i, : pitch.size(0)] = pitch
+            pitchf = row[4]
+            pitchf_padded[i, : pitchf.size(0)] = pitchf
+
+            # dv[i] = row[5]
+            sid[i] = row[5]
+
+        return (
+            phone_padded,
+            phone_lengths,
+            pitch_padded,
+            pitchf_padded,
+            spec_padded,
+            spec_lengths,
+            wave_padded,
+            wave_lengths,
+            # dv
+            sid,
+        )
+
+
+class TextAudioLoader(torch.utils.data.Dataset):
+    """
+    1) loads audio, text pairs
+    2) normalizes text and converts them to sequences of integers
+    3) computes spectrograms from audio files.
+    """
+
+    def __init__(self, audiopaths_and_text, hparams):
+        self.audiopaths_and_text = load_filepaths_and_text(audiopaths_and_text)
+        self.max_wav_value = hparams.max_wav_value
+        self.sampling_rate = hparams.sampling_rate
+        self.filter_length = hparams.filter_length
+        self.hop_length = hparams.hop_length
+        self.win_length = hparams.win_length
+        self.sampling_rate = hparams.sampling_rate
+        self.min_text_len = getattr(hparams, "min_text_len", 1)
+        self.max_text_len = getattr(hparams, "max_text_len", 5000)
+        self._filter()
+
+    def _filter(self):
+        """
+        Filter text & store spec lengths
+        """
+        # Store spectrogram lengths for Bucketing
+        # wav_length ~= file_size / (wav_channels * Bytes per dim) = file_size / (1 * 2)
+        # spec_length = wav_length // hop_length
+        audiopaths_and_text_new = []
+        lengths = []
+        for audiopath, text, dv in self.audiopaths_and_text:
+            if self.min_text_len <= len(text) and len(text) <= self.max_text_len:
+                audiopaths_and_text_new.append([audiopath, text, dv])
+                lengths.append(os.path.getsize(audiopath) // (3 * self.hop_length))
+        self.audiopaths_and_text = audiopaths_and_text_new
+        self.lengths = lengths
+
+    def get_sid(self, sid):
+        sid = torch.LongTensor([int(sid)])
+        return sid
+
+    def get_audio_text_pair(self, audiopath_and_text):
+        # separate filename and text
+        file = audiopath_and_text[0]
+        phone = audiopath_and_text[1]
+        dv = audiopath_and_text[2]
+
+        phone = self.get_labels(phone)
+        spec, wav = self.get_audio(file)
+        dv = self.get_sid(dv)
+
+        len_phone = phone.size()[0]
+        len_spec = spec.size()[-1]
+        if len_phone != len_spec:
+            len_min = min(len_phone, len_spec)
+            len_wav = len_min * self.hop_length
+            spec = spec[:, :len_min]
+            wav = wav[:, :len_wav]
+            phone = phone[:len_min, :]
+        return (spec, wav, phone, dv)
+
+    def get_labels(self, phone):
+        phone = np.load(phone)
+        phone = np.repeat(phone, 2, axis=0)
+        n_num = min(phone.shape[0], 900)  # DistributedBucketSampler
+        phone = phone[:n_num, :]
+        phone = torch.FloatTensor(phone)
+        return phone
+
+    def get_audio(self, filename):
+        audio, sampling_rate = load_wav_to_torch(filename)
+        if sampling_rate != self.sampling_rate:
+            raise ValueError(
+                "{} SR doesn't match target {} SR".format(
+                    sampling_rate, self.sampling_rate
+                )
+            )
+        audio_norm = audio
+        #        audio_norm = audio / self.max_wav_value
+        #        audio_norm = audio / np.abs(audio).max()
+
+        audio_norm = audio_norm.unsqueeze(0)
+        spec_filename = filename.replace(".wav", ".spec.pt")
+        if os.path.exists(spec_filename):
+            try:
+                spec = torch.load(spec_filename)
+            except:
+                logger.warning("%s %s", spec_filename, traceback.format_exc())
+                spec = spectrogram_torch(
+                    audio_norm,
+                    self.filter_length,
+                    self.sampling_rate,
+                    self.hop_length,
+                    self.win_length,
+                    center=False,
+                )
+                spec = torch.squeeze(spec, 0)
+                torch.save(spec, spec_filename, _use_new_zipfile_serialization=False)
+        else:
+            spec = spectrogram_torch(
+                audio_norm,
+                self.filter_length,
+                self.sampling_rate,
+                self.hop_length,
+                self.win_length,
+                center=False,
+            )
+            spec = torch.squeeze(spec, 0)
+            torch.save(spec, spec_filename, _use_new_zipfile_serialization=False)
+        return spec, audio_norm
+
+    def __getitem__(self, index):
+        return self.get_audio_text_pair(self.audiopaths_and_text[index])
+
+    def __len__(self):
+        return len(self.audiopaths_and_text)
+
+
+class TextAudioCollate:
+    """Zero-pads model inputs and targets"""
+
+    def __init__(self, return_ids=False):
+        self.return_ids = return_ids
+
+    def __call__(self, batch):
+        """Collate's training batch from normalized text and aduio
+        PARAMS
+        ------
+        batch: [text_normalized, spec_normalized, wav_normalized]
+        """
+        # Right zero-pad all one-hot text sequences to max input length
+        _, ids_sorted_decreasing = torch.sort(
+            torch.LongTensor([x[0].size(1) for x in batch]), dim=0, descending=True
+        )
+
+        max_spec_len = max([x[0].size(1) for x in batch])
+        max_wave_len = max([x[1].size(1) for x in batch])
+        spec_lengths = torch.LongTensor(len(batch))
+        wave_lengths = torch.LongTensor(len(batch))
+        spec_padded = torch.FloatTensor(len(batch), batch[0][0].size(0), max_spec_len)
+        wave_padded = torch.FloatTensor(len(batch), 1, max_wave_len)
+        spec_padded.zero_()
+        wave_padded.zero_()
+
+        max_phone_len = max([x[2].size(0) for x in batch])
+        phone_lengths = torch.LongTensor(len(batch))
+        phone_padded = torch.FloatTensor(
+            len(batch), max_phone_len, batch[0][2].shape[1]
+        )
+        phone_padded.zero_()
+        sid = torch.LongTensor(len(batch))
+
+        for i in range(len(ids_sorted_decreasing)):
+            row = batch[ids_sorted_decreasing[i]]
+
+            spec = row[0]
+            spec_padded[i, :, : spec.size(1)] = spec
+            spec_lengths[i] = spec.size(1)
+
+            wave = row[1]
+            wave_padded[i, :, : wave.size(1)] = wave
+            wave_lengths[i] = wave.size(1)
+
+            phone = row[2]
+            phone_padded[i, : phone.size(0), :] = phone
+            phone_lengths[i] = phone.size(0)
+
+            sid[i] = row[3]
+
+        return (
+            phone_padded,
+            phone_lengths,
+            spec_padded,
+            spec_lengths,
+            wave_padded,
+            wave_lengths,
+            sid,
+        )
+
+
+class DistributedBucketSampler(torch.utils.data.distributed.DistributedSampler):
+    """
+    Maintain similar input lengths in a batch.
+    Length groups are specified by boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any batch is included either {x | b1 < length(x) <=b2} or {x | b2 < length(x) <= b3}.
+
+    It removes samples which are not included in the boundaries.
+    Ex) boundaries = [b1, b2, b3] -> any x s.t. length(x) <= b1 or length(x) > b3 are discarded.
+    """
+
+    def __init__(
+        self,
+        dataset,
+        batch_size,
+        boundaries,
+        num_replicas=None,
+        rank=None,
+        shuffle=True,
+    ):
+        super().__init__(dataset, num_replicas=num_replicas, rank=rank, shuffle=shuffle)
+        self.lengths = dataset.lengths
+        self.batch_size = batch_size
+        self.boundaries = boundaries
+
+        self.buckets, self.num_samples_per_bucket = self._create_buckets()
+        self.total_size = sum(self.num_samples_per_bucket)
+        self.num_samples = self.total_size // self.num_replicas
+
+    def _create_buckets(self):
+        buckets = [[] for _ in range(len(self.boundaries) - 1)]
+        for i in range(len(self.lengths)):
+            length = self.lengths[i]
+            idx_bucket = self._bisect(length)
+            if idx_bucket != -1:
+                buckets[idx_bucket].append(i)
+
+        for i in range(len(buckets) - 1, -1, -1):  #
+            if len(buckets[i]) == 0:
+                buckets.pop(i)
+                self.boundaries.pop(i + 1)
+
+        num_samples_per_bucket = []
+        for i in range(len(buckets)):
+            len_bucket = len(buckets[i])
+            total_batch_size = self.num_replicas * self.batch_size
+            rem = (
+                total_batch_size - (len_bucket % total_batch_size)
+            ) % total_batch_size
+            num_samples_per_bucket.append(len_bucket + rem)
+        return buckets, num_samples_per_bucket
+
+    def __iter__(self):
+        # deterministically shuffle based on epoch
+        g = torch.Generator()
+        g.manual_seed(self.epoch)
+
+        indices = []
+        if self.shuffle:
+            for bucket in self.buckets:
+                indices.append(torch.randperm(len(bucket), generator=g).tolist())
+        else:
+            for bucket in self.buckets:
+                indices.append(list(range(len(bucket))))
+
+        batches = []
+        for i in range(len(self.buckets)):
+            bucket = self.buckets[i]
+            len_bucket = len(bucket)
+            ids_bucket = indices[i]
+            num_samples_bucket = self.num_samples_per_bucket[i]
+
+            # add extra samples to make it evenly divisible
+            rem = num_samples_bucket - len_bucket
+            ids_bucket = (
+                ids_bucket
+                + ids_bucket * (rem // len_bucket)
+                + ids_bucket[: (rem % len_bucket)]
+            )
+
+            # subsample
+            ids_bucket = ids_bucket[self.rank :: self.num_replicas]
+
+            # batching
+            for j in range(len(ids_bucket) // self.batch_size):
+                batch = [
+                    bucket[idx]
+                    for idx in ids_bucket[
+                        j * self.batch_size : (j + 1) * self.batch_size
+                    ]
+                ]
+                batches.append(batch)
+
+        if self.shuffle:
+            batch_ids = torch.randperm(len(batches), generator=g).tolist()
+            batches = [batches[i] for i in batch_ids]
+        self.batches = batches
+
+        assert len(self.batches) * self.batch_size == self.num_samples
+        return iter(self.batches)
+
+    def _bisect(self, x, lo=0, hi=None):
+        if hi is None:
+            hi = len(self.boundaries) - 1
+
+        if hi > lo:
+            mid = (hi + lo) // 2
+            if self.boundaries[mid] < x and x <= self.boundaries[mid + 1]:
+                return mid
+            elif x <= self.boundaries[mid]:
+                return self._bisect(x, lo, mid)
+            else:
+                return self._bisect(x, mid + 1, hi)
+        else:
+            return -1
+
+    def __len__(self):
+        return self.num_samples // self.batch_size

rvc/lib/train/losses.py

@@ -0,0 +1,58 @@
+import torch
+
+
+def feature_loss(fmap_r, fmap_g):
+    loss = 0
+    for dr, dg in zip(fmap_r, fmap_g):
+        for rl, gl in zip(dr, dg):
+            rl = rl.float().detach()
+            gl = gl.float()
+            loss += torch.mean(torch.abs(rl - gl))
+
+    return loss * 2
+
+
+def discriminator_loss(disc_real_outputs, disc_generated_outputs):
+    loss = 0
+    r_losses = []
+    g_losses = []
+    for dr, dg in zip(disc_real_outputs, disc_generated_outputs):
+        dr = dr.float()
+        dg = dg.float()
+        r_loss = torch.mean((1 - dr) ** 2)
+        g_loss = torch.mean(dg**2)
+        loss += r_loss + g_loss
+        r_losses.append(r_loss.item())
+        g_losses.append(g_loss.item())
+
+    return loss, r_losses, g_losses
+
+
+def generator_loss(disc_outputs):
+    loss = 0
+    gen_losses = []
+    for dg in disc_outputs:
+        dg = dg.float()
+        l = torch.mean((1 - dg) ** 2)
+        gen_losses.append(l)
+        loss += l
+
+    return loss, gen_losses
+
+
+def kl_loss(z_p, logs_q, m_p, logs_p, z_mask):
+    """
+    z_p, logs_q: [b, h, t_t]
+    m_p, logs_p: [b, h, t_t]
+    """
+    z_p = z_p.float()
+    logs_q = logs_q.float()
+    m_p = m_p.float()
+    logs_p = logs_p.float()
+    z_mask = z_mask.float()
+
+    kl = logs_p - logs_q - 0.5
+    kl += 0.5 * ((z_p - m_p) ** 2) * torch.exp(-2.0 * logs_p)
+    kl = torch.sum(kl * z_mask)
+    l = kl / torch.sum(z_mask)
+    return l

rvc/lib/train/mel_processing.py

@@ -0,0 +1,133 @@
+import logging
+
+import torch
+import torch.utils.data
+from librosa.filters import mel as librosa_mel_fn
+
+logger = logging.getLogger(__name__)
+
+MAX_WAV_VALUE = 32768.0
+
+
+def dynamic_range_compression_torch(x, C=1, clip_val=1e-5):
+    """
+    PARAMS
+    ------
+    C: compression factor
+    """
+    return torch.log(torch.clamp(x, min=clip_val) * C)
+
+
+def dynamic_range_decompression_torch(x, C=1):
+    """
+    PARAMS
+    ------
+    C: compression factor used to compress
+    """
+    return torch.exp(x) / C
+
+
+def spectral_normalize_torch(magnitudes):
+    return dynamic_range_compression_torch(magnitudes)
+
+
+def spectral_de_normalize_torch(magnitudes):
+    return dynamic_range_decompression_torch(magnitudes)
+
+
+# Reusable banks
+mel_basis = {}
+hann_window = {}
+
+
+def spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center=False):
+    """Convert waveform into Linear-frequency Linear-amplitude spectrogram.
+
+    Args:
+        y             :: (B, T) - Audio waveforms
+        n_fft
+        sampling_rate
+        hop_size
+        win_size
+        center
+    Returns:
+        :: (B, Freq, Frame) - Linear-frequency Linear-amplitude spectrogram
+    """
+    # Validation
+    if torch.min(y) < -1.07:
+        logger.debug("min value is %s", str(torch.min(y)))
+    if torch.max(y) > 1.07:
+        logger.debug("max value is %s", str(torch.max(y)))
+
+    # Window - Cache if needed
+    global hann_window
+    dtype_device = str(y.dtype) + "_" + str(y.device)
+    wnsize_dtype_device = str(win_size) + "_" + dtype_device
+    if wnsize_dtype_device not in hann_window:
+        hann_window[wnsize_dtype_device] = torch.hann_window(win_size).to(
+            dtype=y.dtype, device=y.device
+        )
+
+    # Padding
+    y = torch.nn.functional.pad(
+        y.unsqueeze(1),
+        (int((n_fft - hop_size) / 2), int((n_fft - hop_size) / 2)),
+        mode="reflect",
+    )
+    y = y.squeeze(1)
+
+    # Complex Spectrogram :: (B, T) -> (B, Freq, Frame, RealComplex=2)
+    spec = torch.stft(
+        y,
+        n_fft,
+        hop_length=hop_size,
+        win_length=win_size,
+        window=hann_window[wnsize_dtype_device],
+        center=center,
+        pad_mode="reflect",
+        normalized=False,
+        onesided=True,
+        return_complex=False,
+    )
+
+    # Linear-frequency Linear-amplitude spectrogram :: (B, Freq, Frame, RealComplex=2) -> (B, Freq, Frame)
+    spec = torch.sqrt(spec.pow(2).sum(-1) + 1e-6)
+    return spec
+
+
+def spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax):
+    # MelBasis - Cache if needed
+    global mel_basis
+    dtype_device = str(spec.dtype) + "_" + str(spec.device)
+    fmax_dtype_device = str(fmax) + "_" + dtype_device
+    if fmax_dtype_device not in mel_basis:
+        mel = librosa_mel_fn(
+            sr=sampling_rate, n_fft=n_fft, n_mels=num_mels, fmin=fmin, fmax=fmax
+        )
+        mel_basis[fmax_dtype_device] = torch.from_numpy(mel).to(
+            dtype=spec.dtype, device=spec.device
+        )
+
+    # Mel-frequency Log-amplitude spectrogram :: (B, Freq=num_mels, Frame)
+    melspec = torch.matmul(mel_basis[fmax_dtype_device], spec)
+    melspec = spectral_normalize_torch(melspec)
+    return melspec
+
+
+def mel_spectrogram_torch(
+    y, n_fft, num_mels, sampling_rate, hop_size, win_size, fmin, fmax, center=False
+):
+    """Convert waveform into Mel-frequency Log-amplitude spectrogram.
+
+    Args:
+        y       :: (B, T)           - Waveforms
+    Returns:
+        melspec :: (B, Freq, Frame) - Mel-frequency Log-amplitude spectrogram
+    """
+    # Linear-frequency Linear-amplitude spectrogram :: (B, T) -> (B, Freq, Frame)
+    spec = spectrogram_torch(y, n_fft, sampling_rate, hop_size, win_size, center)
+
+    # Mel-frequency Log-amplitude spectrogram :: (B, Freq, Frame) -> (B, Freq=num_mels, Frame)
+    melspec = spec_to_mel_torch(spec, n_fft, num_mels, sampling_rate, fmin, fmax)
+
+    return melspec

rvc/lib/train/process_ckpt.py

@@ -0,0 +1,260 @@
+import os
+import sys
+import traceback
+from collections import OrderedDict
+
+import torch
+from i18n.i18n import I18nAuto
+
+i18n = I18nAuto()
+
+
+def savee(ckpt, sr, if_f0, name, epoch, version, hps):
+    try:
+        opt = OrderedDict()
+        opt["weight"] = {}
+        for key in ckpt.keys():
+            if "enc_q" in key:
+                continue
+            opt["weight"][key] = ckpt[key].half()
+        opt["config"] = [
+            hps.data.filter_length // 2 + 1,
+            32,
+            hps.model.inter_channels,
+            hps.model.hidden_channels,
+            hps.model.filter_channels,
+            hps.model.n_heads,
+            hps.model.n_layers,
+            hps.model.kernel_size,
+            hps.model.p_dropout,
+            hps.model.resblock,
+            hps.model.resblock_kernel_sizes,
+            hps.model.resblock_dilation_sizes,
+            hps.model.upsample_rates,
+            hps.model.upsample_initial_channel,
+            hps.model.upsample_kernel_sizes,
+            hps.model.spk_embed_dim,
+            hps.model.gin_channels,
+            hps.data.sampling_rate,
+        ]
+        opt["info"] = "%sepoch" % epoch
+        opt["sr"] = sr
+        opt["f0"] = if_f0
+        opt["version"] = version
+        torch.save(opt, "assets/weights/%s.pth" % name)
+        return "Success."
+    except:
+        return traceback.format_exc()
+
+
+def show_info(path):
+    try:
+        a = torch.load(path, map_location="cpu")
+        return "模型信息:%s\n采样率:%s\n模型是否输入音高引导:%s\n版本:%s" % (
+            a.get("info", "None"),
+            a.get("sr", "None"),
+            a.get("f0", "None"),
+            a.get("version", "None"),
+        )
+    except:
+        return traceback.format_exc()
+
+
+def extract_small_model(path, name, sr, if_f0, info, version):
+    try:
+        ckpt = torch.load(path, map_location="cpu")
+        if "model" in ckpt:
+            ckpt = ckpt["model"]
+        opt = OrderedDict()
+        opt["weight"] = {}
+        for key in ckpt.keys():
+            if "enc_q" in key:
+                continue
+            opt["weight"][key] = ckpt[key].half()
+        if sr == "40k":
+            opt["config"] = [
+                1025,
+                32,
+                192,
+                192,
+                768,
+                2,
+                6,
+                3,
+                0,
+                "1",
+                [3, 7, 11],
+                [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+                [10, 10, 2, 2],
+                512,
+                [16, 16, 4, 4],
+                109,
+                256,
+                40000,
+            ]
+        elif sr == "48k":
+            if version == "v1":
+                opt["config"] = [
+                    1025,
+                    32,
+                    192,
+                    192,
+                    768,
+                    2,
+                    6,
+                    3,
+                    0,
+                    "1",
+                    [3, 7, 11],
+                    [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+                    [10, 6, 2, 2, 2],
+                    512,
+                    [16, 16, 4, 4, 4],
+                    109,
+                    256,
+                    48000,
+                ]
+            else:
+                opt["config"] = [
+                    1025,
+                    32,
+                    192,
+                    192,
+                    768,
+                    2,
+                    6,
+                    3,
+                    0,
+                    "1",
+                    [3, 7, 11],
+                    [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+                    [12, 10, 2, 2],
+                    512,
+                    [24, 20, 4, 4],
+                    109,
+                    256,
+                    48000,
+                ]
+        elif sr == "32k":
+            if version == "v1":
+                opt["config"] = [
+                    513,
+                    32,
+                    192,
+                    192,
+                    768,
+                    2,
+                    6,
+                    3,
+                    0,
+                    "1",
+                    [3, 7, 11],
+                    [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+                    [10, 4, 2, 2, 2],
+                    512,
+                    [16, 16, 4, 4, 4],
+                    109,
+                    256,
+                    32000,
+                ]
+            else:
+                opt["config"] = [
+                    513,
+                    32,
+                    192,
+                    192,
+                    768,
+                    2,
+                    6,
+                    3,
+                    0,
+                    "1",
+                    [3, 7, 11],
+                    [[1, 3, 5], [1, 3, 5], [1, 3, 5]],
+                    [10, 8, 2, 2],
+                    512,
+                    [20, 16, 4, 4],
+                    109,
+                    256,
+                    32000,
+                ]
+        if info == "":
+            info = "Extracted model."
+        opt["info"] = info
+        opt["version"] = version
+        opt["sr"] = sr
+        opt["f0"] = int(if_f0)
+        torch.save(opt, "assets/weights/%s.pth" % name)
+        return "Success."
+    except:
+        return traceback.format_exc()
+
+
+def change_info(path, info, name):
+    try:
+        ckpt = torch.load(path, map_location="cpu")
+        ckpt["info"] = info
+        if name == "":
+            name = os.path.basename(path)
+        torch.save(ckpt, "assets/weights/%s" % name)
+        return "Success."
+    except:
+        return traceback.format_exc()
+
+
+def merge(path1, path2, alpha1, sr, f0, info, name, version):
+    try:
+
+        def extract(ckpt):
+            a = ckpt["model"]
+            opt = OrderedDict()
+            opt["weight"] = {}
+            for key in a.keys():
+                if "enc_q" in key:
+                    continue
+                opt["weight"][key] = a[key]
+            return opt
+
+        ckpt1 = torch.load(path1, map_location="cpu")
+        ckpt2 = torch.load(path2, map_location="cpu")
+        cfg = ckpt1["config"]
+        if "model" in ckpt1:
+            ckpt1 = extract(ckpt1)
+        else:
+            ckpt1 = ckpt1["weight"]
+        if "model" in ckpt2:
+            ckpt2 = extract(ckpt2)
+        else:
+            ckpt2 = ckpt2["weight"]
+        if sorted(list(ckpt1.keys())) != sorted(list(ckpt2.keys())):
+            return "Fail to merge the models. The model architectures are not the same."
+        opt = OrderedDict()
+        opt["weight"] = {}
+        for key in ckpt1.keys():
+            # try:
+            if key == "emb_g.weight" and ckpt1[key].shape != ckpt2[key].shape:
+                min_shape0 = min(ckpt1[key].shape[0], ckpt2[key].shape[0])
+                opt["weight"][key] = (
+                    alpha1 * (ckpt1[key][:min_shape0].float())
+                    + (1 - alpha1) * (ckpt2[key][:min_shape0].float())
+                ).half()
+            else:
+                opt["weight"][key] = (
+                    alpha1 * (ckpt1[key].float()) + (1 - alpha1) * (ckpt2[key].float())
+                ).half()
+        # except:
+        #     pdb.set_trace()
+        opt["config"] = cfg
+        """
+        if(sr=="40k"):opt["config"] = [1025, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10, 10, 2, 2], 512, [16, 16, 4, 4,4], 109, 256, 40000]
+        elif(sr=="48k"):opt["config"] = [1025, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10,6,2,2,2], 512, [16, 16, 4, 4], 109, 256, 48000]
+        elif(sr=="32k"):opt["config"] = [513, 32, 192, 192, 768, 2, 6, 3, 0, "1", [3, 7, 11], [[1, 3, 5], [1, 3, 5], [1, 3, 5]], [10, 4, 2, 2, 2], 512, [16, 16, 4, 4,4], 109, 256, 32000]
+        """
+        opt["sr"] = sr
+        opt["f0"] = 1 if f0 == i18n("是") else 0
+        opt["version"] = version
+        opt["info"] = info
+        torch.save(opt, "assets/weights/%s.pth" % name)
+        return "Success."
+    except:
+        return traceback.format_exc()

rvc/lib/train/utils.py

@@ -0,0 +1,478 @@
+import argparse
+import glob
+import json
+import logging
+import os
+import shutil
+import subprocess
+import sys
+
+import numpy as np
+import torch
+from scipy.io.wavfile import read
+
+MATPLOTLIB_FLAG = False
+
+logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
+logger = logging
+
+
+def load_checkpoint_d(checkpoint_path, combd, sbd, optimizer=None, load_opt=1):
+    assert os.path.isfile(checkpoint_path)
+    checkpoint_dict = torch.load(checkpoint_path, map_location="cpu")
+
+    ##################
+    def go(model, bkey):
+        saved_state_dict = checkpoint_dict[bkey]
+        if hasattr(model, "module"):
+            state_dict = model.module.state_dict()
+        else:
+            state_dict = model.state_dict()
+        new_state_dict = {}
+        for k, v in state_dict.items():  # 模型需要的shape
+            try:
+                new_state_dict[k] = saved_state_dict[k]
+                if saved_state_dict[k].shape != state_dict[k].shape:
+                    logger.warning(
+                        "shape-%s-mismatch. need: %s, get: %s",
+                        k,
+                        state_dict[k].shape,
+                        saved_state_dict[k].shape,
+                    )  #
+                    raise KeyError
+            except:
+                # logger.info(traceback.format_exc())
+                logger.info("%s is not in the checkpoint", k)  # pretrain缺失的
+                new_state_dict[k] = v  # 模型自带的随机值
+        if hasattr(model, "module"):
+            model.module.load_state_dict(new_state_dict, strict=False)
+        else:
+            model.load_state_dict(new_state_dict, strict=False)
+        return model
+
+    go(combd, "combd")
+    model = go(sbd, "sbd")
+    #############
+    logger.info("Loaded model weights")
+
+    iteration = checkpoint_dict["iteration"]
+    learning_rate = checkpoint_dict["learning_rate"]
+    if (
+        optimizer is not None and load_opt == 1
+    ):  ###加载不了，如果是空的的话，重新初始化，可能还会影响lr时间表的更新，因此在train文件最外围catch
+        #   try:
+        optimizer.load_state_dict(checkpoint_dict["optimizer"])
+    #   except:
+    #     traceback.print_exc()
+    logger.info("Loaded checkpoint '{}' (epoch {})".format(checkpoint_path, iteration))
+    return model, optimizer, learning_rate, iteration
+
+
+# def load_checkpoint(checkpoint_path, model, optimizer=None):
+#   assert os.path.isfile(checkpoint_path)
+#   checkpoint_dict = torch.load(checkpoint_path, map_location='cpu')
+#   iteration = checkpoint_dict['iteration']
+#   learning_rate = checkpoint_dict['learning_rate']
+#   if optimizer is not None:
+#     optimizer.load_state_dict(checkpoint_dict['optimizer'])
+#   # print(1111)
+#   saved_state_dict = checkpoint_dict['model']
+#   # print(1111)
+#
+#   if hasattr(model, 'module'):
+#     state_dict = model.module.state_dict()
+#   else:
+#     state_dict = model.state_dict()
+#   new_state_dict= {}
+#   for k, v in state_dict.items():
+#     try:
+#       new_state_dict[k] = saved_state_dict[k]
+#     except:
+#       logger.info("%s is not in the checkpoint" % k)
+#       new_state_dict[k] = v
+#   if hasattr(model, 'module'):
+#     model.module.load_state_dict(new_state_dict)
+#   else:
+#     model.load_state_dict(new_state_dict)
+#   logger.info("Loaded checkpoint '{}' (epoch {})" .format(
+#     checkpoint_path, iteration))
+#   return model, optimizer, learning_rate, iteration
+def load_checkpoint(checkpoint_path, model, optimizer=None, load_opt=1):
+    assert os.path.isfile(checkpoint_path)
+    checkpoint_dict = torch.load(checkpoint_path, map_location="cpu")
+
+    saved_state_dict = checkpoint_dict["model"]
+    if hasattr(model, "module"):
+        state_dict = model.module.state_dict()
+    else:
+        state_dict = model.state_dict()
+    new_state_dict = {}
+    for k, v in state_dict.items():  # 模型需要的shape
+        try:
+            new_state_dict[k] = saved_state_dict[k]
+            if saved_state_dict[k].shape != state_dict[k].shape:
+                logger.warning(
+                    "shape-%s-mismatch|need-%s|get-%s",
+                    k,
+                    state_dict[k].shape,
+                    saved_state_dict[k].shape,
+                )  #
+                raise KeyError
+        except:
+            # logger.info(traceback.format_exc())
+            logger.info("%s is not in the checkpoint", k)  # pretrain缺失的
+            new_state_dict[k] = v  # 模型自带的随机值
+    if hasattr(model, "module"):
+        model.module.load_state_dict(new_state_dict, strict=False)
+    else:
+        model.load_state_dict(new_state_dict, strict=False)
+    logger.info("Loaded model weights")
+
+    iteration = checkpoint_dict["iteration"]
+    learning_rate = checkpoint_dict["learning_rate"]
+    if (
+        optimizer is not None and load_opt == 1
+    ):  ###加载不了，如果是空的的话，重新初始化，可能还会影响lr时间表的更新，因此在train文件最外围catch
+        #   try:
+        optimizer.load_state_dict(checkpoint_dict["optimizer"])
+    #   except:
+    #     traceback.print_exc()
+    logger.info("Loaded checkpoint '{}' (epoch {})".format(checkpoint_path, iteration))
+    return model, optimizer, learning_rate, iteration
+
+
+def save_checkpoint(model, optimizer, learning_rate, iteration, checkpoint_path):
+    logger.info(
+        "Saving model and optimizer state at epoch {} to {}".format(
+            iteration, checkpoint_path
+        )
+    )
+    if hasattr(model, "module"):
+        state_dict = model.module.state_dict()
+    else:
+        state_dict = model.state_dict()
+    torch.save(
+        {
+            "model": state_dict,
+            "iteration": iteration,
+            "optimizer": optimizer.state_dict(),
+            "learning_rate": learning_rate,
+        },
+        checkpoint_path,
+    )
+
+
+def save_checkpoint_d(combd, sbd, optimizer, learning_rate, iteration, checkpoint_path):
+    logger.info(
+        "Saving model and optimizer state at epoch {} to {}".format(
+            iteration, checkpoint_path
+        )
+    )
+    if hasattr(combd, "module"):
+        state_dict_combd = combd.module.state_dict()
+    else:
+        state_dict_combd = combd.state_dict()
+    if hasattr(sbd, "module"):
+        state_dict_sbd = sbd.module.state_dict()
+    else:
+        state_dict_sbd = sbd.state_dict()
+    torch.save(
+        {
+            "combd": state_dict_combd,
+            "sbd": state_dict_sbd,
+            "iteration": iteration,
+            "optimizer": optimizer.state_dict(),
+            "learning_rate": learning_rate,
+        },
+        checkpoint_path,
+    )
+
+
+def summarize(
+    writer,
+    global_step,
+    scalars={},
+    histograms={},
+    images={},
+    audios={},
+    audio_sampling_rate=22050,
+):
+    for k, v in scalars.items():
+        writer.add_scalar(k, v, global_step)
+    for k, v in histograms.items():
+        writer.add_histogram(k, v, global_step)
+    for k, v in images.items():
+        writer.add_image(k, v, global_step, dataformats="HWC")
+    for k, v in audios.items():
+        writer.add_audio(k, v, global_step, audio_sampling_rate)
+
+
+def latest_checkpoint_path(dir_path, regex="G_*.pth"):
+    f_list = glob.glob(os.path.join(dir_path, regex))
+    f_list.sort(key=lambda f: int("".join(filter(str.isdigit, f))))
+    x = f_list[-1]
+    logger.debug(x)
+    return x
+
+
+def plot_spectrogram_to_numpy(spectrogram):
+    global MATPLOTLIB_FLAG
+    if not MATPLOTLIB_FLAG:
+        import matplotlib
+
+        matplotlib.use("Agg")
+        MATPLOTLIB_FLAG = True
+        mpl_logger = logging.getLogger("matplotlib")
+        mpl_logger.setLevel(logging.WARNING)
+    import matplotlib.pylab as plt
+    import numpy as np
+
+    fig, ax = plt.subplots(figsize=(10, 2))
+    im = ax.imshow(spectrogram, aspect="auto", origin="lower", interpolation="none")
+    plt.colorbar(im, ax=ax)
+    plt.xlabel("Frames")
+    plt.ylabel("Channels")
+    plt.tight_layout()
+
+    fig.canvas.draw()
+    data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep="")
+    data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+    plt.close()
+    return data
+
+
+def plot_alignment_to_numpy(alignment, info=None):
+    global MATPLOTLIB_FLAG
+    if not MATPLOTLIB_FLAG:
+        import matplotlib
+
+        matplotlib.use("Agg")
+        MATPLOTLIB_FLAG = True
+        mpl_logger = logging.getLogger("matplotlib")
+        mpl_logger.setLevel(logging.WARNING)
+    import matplotlib.pylab as plt
+    import numpy as np
+
+    fig, ax = plt.subplots(figsize=(6, 4))
+    im = ax.imshow(
+        alignment.transpose(), aspect="auto", origin="lower", interpolation="none"
+    )
+    fig.colorbar(im, ax=ax)
+    xlabel = "Decoder timestep"
+    if info is not None:
+        xlabel += "\n\n" + info
+    plt.xlabel(xlabel)
+    plt.ylabel("Encoder timestep")
+    plt.tight_layout()
+
+    fig.canvas.draw()
+    data = np.fromstring(fig.canvas.tostring_rgb(), dtype=np.uint8, sep="")
+    data = data.reshape(fig.canvas.get_width_height()[::-1] + (3,))
+    plt.close()
+    return data
+
+
+def load_wav_to_torch(full_path):
+    sampling_rate, data = read(full_path)
+    return torch.FloatTensor(data.astype(np.float32)), sampling_rate
+
+
+def load_filepaths_and_text(filename, split="|"):
+    with open(filename, encoding="utf-8") as f:
+        filepaths_and_text = [line.strip().split(split) for line in f]
+    return filepaths_and_text
+
+
+def get_hparams(init=True):
+    """
+    todo:
+      结尾七人组：
+        保存频率、总epoch                     done
+        bs                                    done
+        pretrainG、pretrainD                  done
+        卡号：os.en["CUDA_VISIBLE_DEVICES"]   done
+        if_latest                             done
+      模型：if_f0                             done
+      采样率：自动选择config                  done
+      是否缓存数据集进GPU:if_cache_data_in_gpu done
+
+      -m:
+        自动决定training_files路径,改掉train_nsf_load_pretrain.py里的hps.data.training_files    done
+      -c不要了
+    """
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "-se",
+        "--save_every_epoch",
+        type=int,
+        required=True,
+        help="checkpoint save frequency (epoch)",
+    )
+    parser.add_argument(
+        "-te", "--total_epoch", type=int, required=True, help="total_epoch"
+    )
+    parser.add_argument(
+        "-pg", "--pretrainG", type=str, default="", help="Pretrained Discriminator path"
+    )
+    parser.add_argument(
+        "-pd", "--pretrainD", type=str, default="", help="Pretrained Generator path"
+    )
+    parser.add_argument("-g", "--gpus", type=str, default="0", help="split by -")
+    parser.add_argument(
+        "-bs", "--batch_size", type=int, required=True, help="batch size"
+    )
+    parser.add_argument(
+        "-e", "--experiment_dir", type=str, required=True, help="experiment dir"
+    )  # -m
+    parser.add_argument(
+        "-sr", "--sample_rate", type=str, required=True, help="sample rate, 32k/40k/48k"
+    )
+    parser.add_argument(
+        "-sw",
+        "--save_every_weights",
+        type=str,
+        default="0",
+        help="save the extracted model in weights directory when saving checkpoints",
+    )
+    parser.add_argument(
+        "-v", "--version", type=str, required=True, help="model version"
+    )
+    parser.add_argument(
+        "-f0",
+        "--if_f0",
+        type=int,
+        required=True,
+        help="use f0 as one of the inputs of the model, 1 or 0",
+    )
+    parser.add_argument(
+        "-l",
+        "--if_latest",
+        type=int,
+        required=True,
+        help="if only save the latest G/D pth file, 1 or 0",
+    )
+    parser.add_argument(
+        "-c",
+        "--if_cache_data_in_gpu",
+        type=int,
+        required=True,
+        help="if caching the dataset in GPU memory, 1 or 0",
+    )
+
+    args = parser.parse_args()
+    name = args.experiment_dir
+    experiment_dir = os.path.join("./logs", args.experiment_dir)
+
+    config_save_path = os.path.join(experiment_dir, "config.json")
+    with open(config_save_path, "r") as f:
+        config = json.load(f)
+
+    hparams = HParams(**config)
+    hparams.model_dir = hparams.experiment_dir = experiment_dir
+    hparams.save_every_epoch = args.save_every_epoch
+    hparams.name = name
+    hparams.total_epoch = args.total_epoch
+    hparams.pretrainG = args.pretrainG
+    hparams.pretrainD = args.pretrainD
+    hparams.version = args.version
+    hparams.gpus = args.gpus
+    hparams.train.batch_size = args.batch_size
+    hparams.sample_rate = args.sample_rate
+    hparams.if_f0 = args.if_f0
+    hparams.if_latest = args.if_latest
+    hparams.save_every_weights = args.save_every_weights
+    hparams.if_cache_data_in_gpu = args.if_cache_data_in_gpu
+    hparams.data.training_files = "%s/filelist.txt" % experiment_dir
+    return hparams
+
+
+def get_hparams_from_dir(model_dir):
+    config_save_path = os.path.join(model_dir, "config.json")
+    with open(config_save_path, "r") as f:
+        data = f.read()
+    config = json.loads(data)
+
+    hparams = HParams(**config)
+    hparams.model_dir = model_dir
+    return hparams
+
+
+def get_hparams_from_file(config_path):
+    with open(config_path, "r") as f:
+        data = f.read()
+    config = json.loads(data)
+
+    hparams = HParams(**config)
+    return hparams
+
+
+def check_git_hash(model_dir):
+    source_dir = os.path.dirname(os.path.realpath(__file__))
+    if not os.path.exists(os.path.join(source_dir, ".git")):
+        logger.warning(
+            "{} is not a git repository, therefore hash value comparison will be ignored.".format(
+                source_dir
+            )
+        )
+        return
+
+    cur_hash = subprocess.getoutput("git rev-parse HEAD")
+
+    path = os.path.join(model_dir, "githash")
+    if os.path.exists(path):
+        saved_hash = open(path).read()
+        if saved_hash != cur_hash:
+            logger.warning(
+                "git hash values are different. {}(saved) != {}(current)".format(
+                    saved_hash[:8], cur_hash[:8]
+                )
+            )
+    else:
+        open(path, "w").write(cur_hash)
+
+
+def get_logger(model_dir, filename="train.log"):
+    global logger
+    logger = logging.getLogger(os.path.basename(model_dir))
+    logger.setLevel(logging.DEBUG)
+
+    formatter = logging.Formatter("%(asctime)s\t%(name)s\t%(levelname)s\t%(message)s")
+    if not os.path.exists(model_dir):
+        os.makedirs(model_dir)
+    h = logging.FileHandler(os.path.join(model_dir, filename))
+    h.setLevel(logging.DEBUG)
+    h.setFormatter(formatter)
+    logger.addHandler(h)
+    return logger
+
+
+class HParams:
+    def __init__(self, **kwargs):
+        for k, v in kwargs.items():
+            if type(v) == dict:
+                v = HParams(**v)
+            self[k] = v
+
+    def keys(self):
+        return self.__dict__.keys()
+
+    def items(self):
+        return self.__dict__.items()
+
+    def values(self):
+        return self.__dict__.values()
+
+    def __len__(self):
+        return len(self.__dict__)
+
+    def __getitem__(self, key):
+        return getattr(self, key)
+
+    def __setitem__(self, key, value):
+        return setattr(self, key, value)
+
+    def __contains__(self, key):
+        return key in self.__dict__
+
+    def __repr__(self):
+        return self.__dict__.__repr__()

rvc/lib/uvr5_pack/lib_v5/dataset.py

@@ -0,0 +1,183 @@
+import os
+import random
+
+import numpy as np
+import torch
+import torch.utils.data
+from tqdm import tqdm
+
+from . import spec_utils
+
+
+class VocalRemoverValidationSet(torch.utils.data.Dataset):
+    def __init__(self, patch_list):
+        self.patch_list = patch_list
+
+    def __len__(self):
+        return len(self.patch_list)
+
+    def __getitem__(self, idx):
+        path = self.patch_list[idx]
+        data = np.load(path)
+
+        X, y = data["X"], data["y"]
+
+        X_mag = np.abs(X)
+        y_mag = np.abs(y)
+
+        return X_mag, y_mag
+
+
+def make_pair(mix_dir, inst_dir):
+    input_exts = [".wav", ".m4a", ".mp3", ".mp4", ".flac"]
+
+    X_list = sorted(
+        [
+            os.path.join(mix_dir, fname)
+            for fname in os.listdir(mix_dir)
+            if os.path.splitext(fname)[1] in input_exts
+        ]
+    )
+    y_list = sorted(
+        [
+            os.path.join(inst_dir, fname)
+            for fname in os.listdir(inst_dir)
+            if os.path.splitext(fname)[1] in input_exts
+        ]
+    )
+
+    filelist = list(zip(X_list, y_list))
+
+    return filelist
+
+
+def train_val_split(dataset_dir, split_mode, val_rate, val_filelist):
+    if split_mode == "random":
+        filelist = make_pair(
+            os.path.join(dataset_dir, "mixtures"),
+            os.path.join(dataset_dir, "instruments"),
+        )
+
+        random.shuffle(filelist)
+
+        if len(val_filelist) == 0:
+            val_size = int(len(filelist) * val_rate)
+            train_filelist = filelist[:-val_size]
+            val_filelist = filelist[-val_size:]
+        else:
+            train_filelist = [
+                pair for pair in filelist if list(pair) not in val_filelist
+            ]
+    elif split_mode == "subdirs":
+        if len(val_filelist) != 0:
+            raise ValueError(
+                "The `val_filelist` option is not available in `subdirs` mode"
+            )
+
+        train_filelist = make_pair(
+            os.path.join(dataset_dir, "training/mixtures"),
+            os.path.join(dataset_dir, "training/instruments"),
+        )
+
+        val_filelist = make_pair(
+            os.path.join(dataset_dir, "validation/mixtures"),
+            os.path.join(dataset_dir, "validation/instruments"),
+        )
+
+    return train_filelist, val_filelist
+
+
+def augment(X, y, reduction_rate, reduction_mask, mixup_rate, mixup_alpha):
+    perm = np.random.permutation(len(X))
+    for i, idx in enumerate(tqdm(perm)):
+        if np.random.uniform() < reduction_rate:
+            y[idx] = spec_utils.reduce_vocal_aggressively(
+                X[idx], y[idx], reduction_mask
+            )
+
+        if np.random.uniform() < 0.5:
+            # swap channel
+            X[idx] = X[idx, ::-1]
+            y[idx] = y[idx, ::-1]
+        if np.random.uniform() < 0.02:
+            # mono
+            X[idx] = X[idx].mean(axis=0, keepdims=True)
+            y[idx] = y[idx].mean(axis=0, keepdims=True)
+        if np.random.uniform() < 0.02:
+            # inst
+            X[idx] = y[idx]
+
+        if np.random.uniform() < mixup_rate and i < len(perm) - 1:
+            lam = np.random.beta(mixup_alpha, mixup_alpha)
+            X[idx] = lam * X[idx] + (1 - lam) * X[perm[i + 1]]
+            y[idx] = lam * y[idx] + (1 - lam) * y[perm[i + 1]]
+
+    return X, y
+
+
+def make_padding(width, cropsize, offset):
+    left = offset
+    roi_size = cropsize - left * 2
+    if roi_size == 0:
+        roi_size = cropsize
+    right = roi_size - (width % roi_size) + left
+
+    return left, right, roi_size
+
+
+def make_training_set(filelist, cropsize, patches, sr, hop_length, n_fft, offset):
+    len_dataset = patches * len(filelist)
+
+    X_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
+    y_dataset = np.zeros((len_dataset, 2, n_fft // 2 + 1, cropsize), dtype=np.complex64)
+
+    for i, (X_path, y_path) in enumerate(tqdm(filelist)):
+        X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
+        coef = np.max([np.abs(X).max(), np.abs(y).max()])
+        X, y = X / coef, y / coef
+
+        l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
+        X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
+        y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
+
+        starts = np.random.randint(0, X_pad.shape[2] - cropsize, patches)
+        ends = starts + cropsize
+        for j in range(patches):
+            idx = i * patches + j
+            X_dataset[idx] = X_pad[:, :, starts[j] : ends[j]]
+            y_dataset[idx] = y_pad[:, :, starts[j] : ends[j]]
+
+    return X_dataset, y_dataset
+
+
+def make_validation_set(filelist, cropsize, sr, hop_length, n_fft, offset):
+    patch_list = []
+    patch_dir = "cs{}_sr{}_hl{}_nf{}_of{}".format(
+        cropsize, sr, hop_length, n_fft, offset
+    )
+    os.makedirs(patch_dir, exist_ok=True)
+
+    for i, (X_path, y_path) in enumerate(tqdm(filelist)):
+        basename = os.path.splitext(os.path.basename(X_path))[0]
+
+        X, y = spec_utils.cache_or_load(X_path, y_path, sr, hop_length, n_fft)
+        coef = np.max([np.abs(X).max(), np.abs(y).max()])
+        X, y = X / coef, y / coef
+
+        l, r, roi_size = make_padding(X.shape[2], cropsize, offset)
+        X_pad = np.pad(X, ((0, 0), (0, 0), (l, r)), mode="constant")
+        y_pad = np.pad(y, ((0, 0), (0, 0), (l, r)), mode="constant")
+
+        len_dataset = int(np.ceil(X.shape[2] / roi_size))
+        for j in range(len_dataset):
+            outpath = os.path.join(patch_dir, "{}_p{}.npz".format(basename, j))
+            start = j * roi_size
+            if not os.path.exists(outpath):
+                np.savez(
+                    outpath,
+                    X=X_pad[:, :, start : start + cropsize],
+                    y=y_pad[:, :, start : start + cropsize],
+                )
+            patch_list.append(outpath)
+
+    return VocalRemoverValidationSet(patch_list)