appautomaton/openmoss-tts-local-mlx

OpenMOSS TTS Local Transformer — MLX 8-bit

This repository contains an MLX-native int8 conversion of OpenMOSS TTS Local Transformer for single-speaker text-to-speech on Apple Silicon.

It is intended for local speech generation with mlx-speech, without a PyTorch runtime at inference time.

Variants

Path	Precision
mlx-int8/	int8 quantized weights

Model Details

• Developed by: AppAutomaton
• Shared by: AppAutomaton on Hugging Face
• Upstream model: OpenMOSS-Team/MOSS-TTS-Local-Transformer
• Task: single-speaker text-to-speech and voice cloning
• Runtime: MLX on Apple Silicon

How to Get Started

Command-line generation with mlx-speech:

Generate speech:

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

Clone a voice:

python scripts/generate_moss_local.py \
  --mode clone \
  --text "This is a cloned voice." \
  --reference-audio reference.wav \
  --output outputs/clone.wav

Minimal Python usage:

from mlx_speech.generation import MossTTSLocalModel

model = MossTTSLocalModel.from_path("mlx-int8")

Notes

• This repo contains the quantized MLX runtime artifact only.
• The conversion keeps the original OpenMOSS local TTS architecture and remaps weights explicitly for MLX inference.
• The default runtime path uses W8Abf16 mixed precision with global and local KV cache enabled.

Links

• Source code: mlx-speech
• More examples: AppAutomaton

License

Apache 2.0 — following the upstream license published with OpenMOSS-Team/MOSS-TTS-Local-Transformer.