Create inference_cli.toml

inference_cli.toml

@@ -0,0 +1,76 @@
+# Configuration for inference-cli.py
+
+# --- Input Files and Text ---
+
+# Path or Hugging Face Hub ID (e.g., "hf://user/repo/model.safetensors") to the TTS model checkpoint.
+# This is the primary required setting. The script infers model type (DiT/UNetT) from this path.
+ckpt_path = "hf://Gregniuki/F5-tts_English_German_Polish/multi3/model_1100000.pt" # Default used in script
+
+# Path to the reference audio file (WAV, MP3, etc.). Recommended < 10 seconds.
+ref_audio = "tests/ref_audio/test_en_1_ref_short.wav"
+
+# Text transcription of the reference audio.
+# If set to "", the script will attempt to transcribe ref_audio using Whisper.
+ref_text = "Some call me nature, others call me mother nature."
+
+# Text to be synthesized by the TTS model.
+gen_text = "I don't really care what you call me. I've been a silent spectator, watching species evolve, empires rise and fall. But always remember, I am mighty and enduring. Respect me and I'll nurture you; ignore me and you shall face the consequences."
+
+# Optional: Path to a UTF-8 encoded text file containing the text to synthesize.
+# If provided, this overrides the gen_text setting above.
+gen_file = ""
+
+# Path to the tokenizer.json file required by the model.
+tokenizer_path = "data/Emilia_ZH_EN_pinyin/tokenizer.json" # Default used in script
+
+# --- Output Settings ---
+
+# Directory where the output audio (.wav) and spectrogram (.png) will be saved.
+output_dir = "tests"
+
+# Base name for the output files (e.g., "my_speech" -> my_speech.wav, my_speech.png).
+output_name = "out" # Default: "out"
+
+# --- Language Settings ---
+
+# Language code for phonemizing the *reference* text (e.g., en-us, en-gb, de, pl, fr-fr).
+# Needs to match the language spoken in ref_audio / ref_text. See phonemizer docs for codes.
+ref_language = "en-us" # Default: "en-us"
+
+# Language code for phonemizing the *generated* text (gen_text / gen_file).
+# Needs to match the language you want the model to speak.
+language = "en-us" # Default: "en-us"
+
+# --- Inference Parameters ---
+
+# Speech speed multiplier. > 1.0 is faster, < 1.0 is slower.
+speed = 1.0 # Default: 1.0
+
+# Number of Function Evaluations (sampling steps). Higher values may improve quality but increase time.
+nfe = 32 # Default: 32
+
+# Classifier-Free Guidance strength. Higher values increase adherence to reference timbre but can reduce naturalness.
+cfg = 2.0 # Default: 2.0
+
+# Sway sampling coefficient (experimental). Often -1.0 or disabled.
+sway = -1.0 # Default: -1.0
+
+# --- Postprocessing ---
+
+# Duration (in seconds) for cross-fading between generated audio batches. 0 disables cross-fading.
+cross_fade = 0.15 # Default: 0.15
+
+# Apply silence removal to the final generated audio using pydub.
+remove_silence = false # Default: false
+
+# --- System Settings ---
+
+# Optional: Hugging Face API token for downloading private models or high-rate limiting.
+# Can also be set via environment variable HUGGING_FACE_HUB_TOKEN.
+hf_token = "" # Default: "" (uses cached credentials or public access)
+
+# Optional: Specify the device ('cuda', 'cpu', 'mps'). If commented out or empty, defaults to auto-detection.
+# device = "cuda"
+
+# Optional: Specify the data type ('float16', 'bfloat16', 'float32'). If commented out or empty, defaults based on device.
+# dtype = "float16"