MLX-Audio

MLX-Audio is an audio framework built on Apple MLX for Apple Silicon Macs. It supports VoxCPM as one of its TTS backends, and gives you a CLI, an OpenAI-compatible FastAPI server, and a separate Next.js web UI.

Note

This is a good fit if you want to run VoxCPM on an Apple Silicon Mac through the MLX stack, especially when you also want a local API server or a browser UI.

Supported VoxCPM Versions

VoxCPM 1.0 (0.5B)

✅ Supported

VoxCPM 1.5

✅ Supported

VoxCPM 2

⚠️ This page focuses on the current VoxCPM 1.x path on MLX-Audio

Prerequisites

  • Apple Silicon Mac (M1 or later)

  • Python 3.10+

  • mlx and mlx-audio installed

Installation

pip install mlx-audio

Or from source:

git clone https://github.com/Blaizzy/mlx-audio.git
cd mlx-audio
pip install -e .

CLI Usage

Generate speech directly from the command line:

mlx_audio.tts.generate --text "Hello from MLX-Audio!" --model voxcpm

The --model flag accepts voxcpm (maps to VoxCPM 1.0) or voxcpm1.5.

API Server

MLX-Audio includes an OpenAI-compatible FastAPI server:

mlx_audio.server --model voxcpm

This starts a local server that accepts /v1/audio/speech requests, compatible with the OpenAI TTS API format. You can point any OpenAI-compatible client at it.

Web UI

A separate Next.js web UI is included under mlx_audio/ui/. See the MLX-Audio README for setup instructions.

Limitations

  • VoxCPM 2 support is not yet available in MLX-Audio. The current backend covers VoxCPM 1.0 and 1.5.

  • Performance depends on your Apple Silicon chip. M1 Pro / M1 Max and later provide the best experience.

  • The MLX-Audio project is community-maintained and not officially affiliated with OpenBMB.