stgreenman/sesame-colab-v1

Uploaded model

• Developed by: stgreenman
• License: apache-2.0
• Finetuned from model : unsloth/csm-1b

Usage

from transformers import CsmForConditionalGeneration, AutoProcessor
from IPython.display import Audio, display
import soundfile as sf
import torch

# Load with the correct model class
model = CsmForConditionalGeneration.from_pretrained("stgreenman/sesame-colab-v1")
processor = AutoProcessor.from_pretrained("stgreenman/sesame-colab-v1")

# Move model to CUDA
model = model.to("cuda")

# TTS function (exactly from your original)
def generate_speech(text, speaker_id=0):
    inputs = processor(f"[{speaker_id}]{text}", add_special_tokens=True).to("cuda")
    audio_values = model.generate(
        **inputs,
        max_new_tokens=125,
        depth_decoder_temperature=0.6,
        depth_decoder_top_k=0,
        depth_decoder_top_p=0.9,
        temperature=0.8,
        top_k=50,
        top_p=1.0,
        output_audio=True
    )
    audio = audio_values[0].to(torch.float32).cpu().numpy()
    return audio

# Test
text = "Hello, this is my fine-tuned voice!"
audio = generate_speech(text)
sf.write("test_output.wav", audio, 24000)
display(Audio(audio, rate=24000))

This csm model was trained 2x faster with Unsloth and Huggingface's TRL library.