g0udurix/Chatterbox-TTS-French-MLX

Chatterbox TTS French (MLX)

MLX-converted weights of Thomcles/Chatterbox-TTS-French, with multiple quantized variants.

Variants

• fp16/
• mxfp8/
• mxfp4/
• int8/
• int4/

Usage (MLX)

from mlx_audio.tts.generate import load_model, generate_audio

model = load_model(model_path="./mxfp8")
text = "Jean-Paul Sartre laisse a la posterite une oeuvre considerable."

wav, sr = generate_audio(
    model,
    text=text,
    ref_audio=None,
    ref_text=None,
    cfg_weight=0.5,
    exaggeration=0.5,
    temperature=0.6,
)

Notes

• No reference audio is included.
• Tokenizer files are preserved from the base model.

License

Same license as the base model: CC-BY-4.0.

Base model and data

• Base model: Thomcles/Chatterbox-TTS-French
• Dataset: amphion/Emilia-Dataset (French)