mlx-audio-server

# MLX Audio Server

Local 24x7 OpenAI-compatible API server for STT/TTS, powered by MLX on your Mac.

[mlx-audio](https://github.com/Blaizzy/mlx-audio): The best audio processing library built on Apple's MLX framework, providing fast and efficient text-to-speech (TTS), speech-to-text (STT), and speech-to-speech (STS) on Apple Silicon.

[guoqiao/tap/mlx-audio-server](https://github.com/guoqiao/homebrew-tap/blob/main/Formula/mlx-audio-server.rb): Homebrew Formula to install `mlx-audio` with `brew`, and run `mlx_audio.server` as a LaunchAgent service on macOS.

## Requirements

- `mlx`: macOS with Apple Silicon
- `brew`: used to install deps if not available

## Installation

```bash
bash ${baseDir}/install.sh
```
This script will:
- install ffmpeg/jq with brew if missing.
- install homebrew formula `mlx-audio-server` from `guoqiao/tap`
- start brew service for `mlx-audio-server`

## Usage

STT/Speech-To-Text(default model: **mlx-community/glm-asr-nano-2512-8bit**):
```bash
# input will be converted to wav with ffmpeg, if not yet.
# output will be transcript text only.
bash ${baseDir}/run_stt.sh <audio_or_video_path>
```

TTS/Text-To-Speech(default model: **mlx-community/Qwen3-TTS-12Hz-1.7B-VoiceDesign-bf16**):
```bash
# audio will be saved into a tmp dir, with default name `speech.wav`, and print to stdout.
bash ${baseDir}/run_tts.sh "Hello, Human!"
# or you can specify a output dir
bash ${baseDir}/run_tts.sh "Hello, Human!" ./output
# output will be audio path only.
```
You can use both scripts directly, or as example/reference.

# Homebrew Formula mlx-audio-server

[mlx-audio](https://github.com/Blaizzy/mlx-audio) is an audio processing library built on Apple's MLX framework for text-to-speech (TTS), speech-to-text (STT), and speech-to-speech (STS) on Apple Silicon.

Currently, it is a python package with a few cli tools such as `mlx_audio.server`. It's dependencies for server is not working properly (yet), and has no builtin support to run it as system service in background.

This Homebrew Formula re-packages it for homebrew with missing deps, and the main focus is to add LaunchAgent service support for the `mlx_audio.server` on macOS, so you can use it as an 24x7 local OpenAI-compatible API server.

## Installation

Tap and install:

```bash
# install the formula with cli tools, not only server
brew install --HEAD guoqiao/tap/mlx-audio-server

# start the mlx-audio.server as a LaunchAgent service
brew services start mlx-audio-server

# verify
curl -Ss http://localhost:8899/v1/models
```

## Usage
Here are examples to use `GLM-ASR-Nano-2512` model for STT with different ways:

with curl:
```
bash stt_glmasr.sh  path/to/audio
```

with openai python sdk:
```
uv run openai_api_transcribe.py path/to/audio
```

with Spokenly iOS/macOS app, as OpenAI Compatible API:
```
URL: http://<IP>:8899
Model: mlx-community/glm-asr-nano-2512-8bit
API KEY: <blank>
```
NOTE: `/v1` is not needed in URL here.


## About

This is a Homebrew tap for the mlx-audio Python package, providing fast and efficient audio processing on Apple Silicon using MLX framework.

For more information, visit the [mlx-audio repository](https://github.com/Blaizzy/mlx-audio).