Audio8-TTS-Realistic-0.6b-bf16

MLX conversion of Edge0/Audio8-TTS-Realistic-0.6b. Audio8 generates mono 44.1 kHz speech with optional zero-shot voice cloning. The codec is included; runtime inference does not require PyTorch or remote model code.

from mlx_audio.audio_io import write
from mlx_audio.tts.utils import load

model = load("nativ-community/Audio8-TTS-Realistic-0.6b-bf16")
result = next(model.generate(text="Welcome to Audio8 on Apple Silicon."))
write("speech.wav", result.audio, result.sample_rate)

For voice cloning, pass ref_audio="reference.wav" and ref_text="Exact transcript of the reference." to generate().

python -m mlx_audio.tts.generate \
  --model nativ-community/Audio8-TTS-Realistic-0.6b-bf16 \
  --text "Welcome to Audio8 on Apple Silicon."

The codec retains its source precision when converting or quantizing the DualAR model. Generation returns one complete waveform per call. Text, reference audio, and generated frames share a 2,048-position context budget.