🇧🇬 BG-TTS V5 — MLX (Apple Silicon)

Native MLX port of beleata74/bg-tts-v5 for Apple Silicon (M1/M2/M3/M4).

No CUDA, no NeMo, no PyTorch required. Runs fully on Apple Silicon via MLX.

Requirements

pip install mlx soundfile numpy
pip install "nanocodec-mlx @ git+https://github.com/nineninesix-ai/nanocodec-mlx.git"

Quick Start

from tts_mlx.inference import synthesize

synthesize(
    checkpoint=".",  # path to this repo
    text="Здравейте, аз съм българска система за синтез на реч.",
    output="output.wav",
    speaker_id=0,      # 0 = AI voice, 1 = audiobook narrator
    temperature=0.25,
    top_k=50,
    top_p=0.8,
)

Speakers

Speaker	Description	Best text length
0	AI-generated, clear & fast	Any (20–500+ chars)
1	Real female, audiobook narrator	250–320 chars

Credits

Original model by beleata74, created with Claude. MLX port by Radi Totev. NanoCodec MLX by nineninesix-ai.

Downloads last month: 68

MLX

Hardware compatibility

Quantized