mlx-speech 0.1.0

MLX-native speech library for Apple Silicon.

mlx-speech

Local speech synthesis on Apple Silicon, running pure MLX. No cloud, no PyTorch.

Model	Best for
MossTTSLocal	shorter TTS, voice cloning, continuation
MOSS-TTSD	multi-speaker dialogue
MOSS-SoundEffect	text-to-sound-effect
VibeVoice	long-form speech, voice-conditioned generation

Requirements

• Apple Silicon Mac (M1 or later)
• Python 3.13+
• uv

Installation

git clone https://github.com/appautomaton/mlx-speech.git
cd mlx-speech
uv sync

PyPI package (pip install mlx-speech) coming soon.

Convert the checkpoints you want to use — each model family has a scripts/convert_*.py entry point:

python scripts/convert_moss_local.py
python scripts/convert_moss_audio_tokenizer.py
python scripts/convert_moss_ttsd.py
python scripts/convert_moss_sound_effect.py
python scripts/convert_vibevoice.py

Usage

Generate speech:

python scripts/generate_moss_local.py \
  --text "Hello, this is a test." \
  --output outputs/moss_local.wav

Clone a voice:

python scripts/generate_moss_local.py \
  --mode clone \
  --text "Hello, this is a cloned sample." \
  --reference-audio reference.wav \
  --output outputs/moss_local_clone.wav

Multi-speaker dialogue:

python scripts/generate_moss_ttsd.py \
  --text "[S1] Watson, we should go now." \
  --output outputs/ttsd.wav

Sound effect:

python scripts/generate_moss_sound_effect.py \
  --ambient-sound "rolling thunder with steady rainfall on a metal roof" \
  --duration-seconds 8 \
  --output outputs/thunder.wav

VibeVoice:

python scripts/generate_vibevoice.py \
  --text "Hello from VibeVoice." \
  --output outputs/vibevoice.wav

Exploring the Codebase

The PyPI package is still in progress. The best way to explore right now is to drop the repo into an agentic coding tool like Claude Code or Cursor — the codebase is structured and self-describing, and an agent can walk you through it quickly.

Model Guides

Each family has a doc covering behavior, flags, and known limitations:

• MossTTSLocal
• MOSS-TTSD
• MOSS-SoundEffect
• VibeVoice

Development

uv run pytest
uv run ruff check .
uv build --no-sources

mlx-speech/
  src/mlx_speech/    library code
  scripts/          conversion and generation entry points
  models/           local checkpoints (not in git)
  tests/            unit and integration tests
  docs/             model-family behavior guides

License

MIT — see LICENSE