Audio8-TTS-Preview-0.6b-bf16 (MLX)

MLX conversion of Audio8/Audio8-TTS-Preview-0.6b (revision 1b17c91) for Apple Silicon — a 0.6B-parameter multilingual text-to-speech model with zero-shot voice cloning and a bundled 44.1 kHz neural codec.

• Language model: bf16 (as published upstream)
• Codec: fp32, converted from codec.pth to safetensors with weight-norm folded and conv weights in MLX (channels-last) layout
• Architecture: arktts — a DualAR transformer (24-layer slow AR predicting one semantic token per frame, 4-layer fast AR predicting 10 codec codebooks per frame), inspired by Fish Audio S2 Pro

Conversion parity vs the PyTorch reference (fp32, CPU): unit/block outputs within 1e-4, reference-audio codec encoding 100% code-exact, greedy generation 100% token-exact over the validation utterance, decoded waveform max-abs 7.5e-6.

Usage (mlx-audio)

Requires an mlx-audio version with arktts support.

import soundfile as sf
from mlx_audio.tts.utils import load

model = load("mlx-community/Audio8-TTS-Preview-0.6b-bf16")

# zero-shot voice cloning: the reference transcript must match the reference audio
for result in model.generate(
    text="Welcome to Audio8 TTS, running on Apple Silicon.",
    ref_audio="reference.wav",
    ref_text="Transcript of the reference clip.",
    temperature=0.7, top_p=0.9, top_k=50,
):
    sf.write("output.wav", result.audio, result.sample_rate)

Without ref_audio, the model synthesizes with a default voice.

Supported languages

Cantonese, Chinese, Dutch, English, French, German, Italian, Japanese, Korean, Polish, Spanish (per the upstream preview release).

License

Apache-2.0, following the upstream model. The bundled codec weights are part of the upstream repository and carry the same license.