Voice
Openmeow
By A native macOS menu bar app that serves as a local, OpenAI-compatible voice API gateway. Ready for OpenClaw.
Configuration Example
"messages": {
"tts": {
"auto": "always",
"provider": "openai",
"providers": {
"openai": {
"apiKey": "dummy-key",
"baseUrl": "http://127.0.0.1:23333/v1",
"model": "qwen3-tts-1.7b-mlx",
"voice": "Vivian"
}
},
"timeoutMs": 60000
}
}README
<p align="center">
<img src="assets/icon_readme.png" width="128" height="128" alt="OpenMeow">
</p>
<h1 align="center">OpenMeow</h1>
<p align="center">
A native macOS menu bar app that serves as a local, OpenAI-compatible voice API gateway.
</p>
<p align="center">
<a href="LICENSE"><img src="https://img.shields.io/badge/license-MIT-blue.svg" alt="License"></a>
<img src="https://img.shields.io/badge/platform-macOS-lightgrey.svg" alt="Platform">
<img src="https://img.shields.io/badge/Apple_Silicon-M1%2B-black.svg?logo=apple" alt="Apple Silicon">
<img src="https://img.shields.io/badge/Swift-6.3-F05138.svg?logo=swift&logoColor=white" alt="Swift">
<img src="https://img.shields.io/badge/API-OpenAI_Compatible-10a37f.svg" alt="OpenAI Compatible">
</p>
<p align="center">
<a href="#features">Features</a> •
<a href="#installation">Installation</a> •
<a href="#api">API</a> •
<a href="#use-with-openclaw">OpenClaw</a> •
<a href="#models">Models</a> •
<a href="#building">Building</a> •
<a href="README_CN.md">δΈζ</a>
</p>
---
## Features
- **Menu bar app** β runs quietly in the background on macOS (Apple Silicon)
- **OpenAI-compatible API** β drop-in replacement for `/v1/audio/speech` and `/v1/audio/transcriptions`
- **Multiple engines** β sherpa-onnx, WhisperKit, speech-swift (Qwen3-TTS/ASR)
- **Audio format support** β WAV, MP3, Opus (OGG/WebM), PCM, FLAC, AAC
- **Model store** β download and manage models from the built-in registry
- **Works with OpenClaw** β give [OpenClaw](https://github.com/openclaw/openclaw) local voice capabilities in one line of config
- **Privacy first** β everything runs locally, no data leaves your machine
## Requirements
- macOS 15.0+
- Apple Silicon (M1/M2/M3/M4/M5)
## Installation
Download the latest `.dmg` or `.zip` from [Releases](../../releases), drag to Applications, and launch. No additional tools required.
## API
OpenMeow listens on `http://127.0.0.1:23333` by default.
| Endpoint | Method | Description |
|----------|--------|-------------|
| `/v1/audio/speech` | POST | Text-to-Speech |
| `/v1/audio/transcriptions` | POST | Speech-to-Text |
| `/v1/models` | GET | List available models |
| `/v1/voices` | GET | List available voices |
| `/health` | GET | Health check |
### TTS Example
```bash
curl -X POST http://127.0.0.1:23333/v1/audio/speech \
-H "Content-Type: application/json" \
-d '{"model": "kokoro-multi-lang-v1_0", "input": "Hello from OpenMeow!", "voice": "af_heart"}' \
--output speech.mp3
```
### ASR Example
```bash
curl -X POST http://127.0.0.1:23333/v1/audio/transcriptions \
-F "[email protected]" \
-F "model=whisper-large-v3-turbo"
```
## Use with OpenClaw
OpenMeow speaks the same API as OpenAI, so [OpenClaw](https://github.com/openclaw/openclaw) can use it as a local voice backend with zero cloud dependency.
**TTS** β add to your OpenClaw config:
```jsonc
"messages": {
"tts": {
"auto": "always",
"provider": "openai",
"providers": {
"openai": {
"apiKey": "dummy-key",
"baseUrl": "http://127.0.0.1:23333/v1",
"model": "qwen3-tts-1.7b-mlx",
"voice": "Vivian"
}
},
"timeoutMs": 60000
}
}
```
**ASR** β add to `tools.media.audio`:
```jsonc
"tools": {
"media": {
"audio": {
"enabled": true,
"models": [
{
"type": "cli",
"command": "/bin/sh",
"args": [
"-c",
"curl -s http://127.0.0.1:23333/v1/audio/transcriptions -F file=@{{MediaPath}} -F model=qwen3-asr-0.6b-mlx | jq -r .text"
],
"timeoutSeconds": 60
}
]
}
}
}
```
## Models
### TTS
| Model | Size | Languages |
|-------|------|-----------|
| Kokoro Multi-lang v1.0 | 360 MB | Chinese, English |
| Kitten Nano v0.1 | 26 MB | English |
| Kitten Mini v0.1 | 18 MB | English |
| Qwen3-TTS 0.6B (MLX) | 1.7 GB | Multilingual |
| Qwen3-TTS 1.7B (MLX) | 3.2 GB | Multilingual |
### ASR
| Model | Size | Languages |
|-------|------|-----------|
| FireRedASR v2 | 200 MB | Chinese + 20 dialects |
| Qwen3-ASR 0.6B (MLX) | 680 MB | 30+ languages |
| Whisper Large v3 Turbo | 600 MB | 92+ languages |
| Whisper Base | 150 MB | 99 languages |
## Building
```bash
# 1. Clone
git clone https://github.com/user/openmeow.git
cd openmeow
# 2. Download frameworks
Scripts/download-sherpa-onnx.sh
Scripts/download-opus.sh
Scripts/download-lame.sh
# 3. Open in Xcode and build
open openmeow/openmeow.xcodeproj
```
> Frameworks (sherpa-onnx, opus, lame) are not included in the repo due to size. Build scripts download and compile them locally.
## License
[MIT](LICENSE)
### Third-Party Components
| Component | License | Link |
|-----------|---------|------|
| [sherpa-onnx](https://github.com/k2-fsa/sherpa-onnx) | Apache-2.0 | On-device speech (TTS/ASR) |
| [WhisperKit](https://github.com/argmaxinc/WhisperKit) | MIT | CoreML-based speech recognition |
| [speech-swift](https://github.com/soniqo/speech-swift) | Apache-2.0 | Qwen3-TTS/ASR via MLX |
| [Hummingbird](https://github.com/hummingbird-project/hummingbird) | Apache-2.0 | Lightweight HTTP server |
| [Opus](https://github.com/xiph/opus) / [Ogg](https://github.com/xiph/ogg) | BSD-3-Clause | Audio codec |
| [LAME](https://lame.sourceforge.io/) | LGPL-2.0 | MP3 encoder (dynamically linked) |
LAME is the only LGPL component and is dynamically linked as `libmp3lame.dylib`. You may replace it with your own build. See [THIRD-PARTY-LICENSES](THIRD-PARTY-LICENSES) for details.
voice
Comments
Sign in to leave a comment