A high-performance, local speech-to-text engine powered by mlx-audio and Whisper models, optimized specifically for Apple Silicon hardware.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install openclaw-mlx-stt
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install openclaw-mlx-stt using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is MLX-STT: Local Whisper Speech-to-Text?
MLX-STT is a specialized tool within the ecosystem of Openclaw Skills that enables private, offline speech-to-text processing. By leveraging the mlx-audio framework and OpenAI's Whisper models, it provides near-instant transcription and translation directly on your Mac. Because it runs entirely on Apple Silicon, it eliminates the need for expensive API keys or external cloud services, ensuring your audio data never leaves your machine.
This skill is designed for developers and power users who require reliable transcription for various audio formats, including MP3, WAV, and even audio extraction from MP4 files. It supports a vast array of models, from the lightning-fast whisper-large-v3-turbo to high-precision Qwen3-ASR models, making it one of the most versatile Openclaw Skills for audio processing.
MLX-STT: Local Whisper Speech-to-Text Use Cases
- Transcribing meeting recordings or voice memos into structured text.
- Translating foreign language audio files directly into English text.
- Generating timestamps for video subtitles or podcast indexing.
- Automating batch transcription of large audio archives locally.
- Extracting dialogue from video files for content repurposing.
How MLX-STT: Local Whisper Speech-to-Text Works
- The user provides an audio file path and triggers the transcribe action via a natural language request or CLI.
- The skill initializes the configured model (e.g., Whisper v3) using the mlx-audio backend on the local GPU.
- The system automatically detects the language (if not specified) and processes the audio stream.
- It generates a detailed JSON response containing the full text, language metadata, and segment-level timestamps.
- The result is returned to the agent for further processing or saved directly to the workspace.
MLX-STT: Local Whisper Speech-to-Text Setup
First, ensure you have the mlx-audio library installed on your system:
uv tool install mlx-audio --prerelease=allow
Then, enable the skill in your openclaw.json configuration file:
{
"plugins": {
"entries": {
"openclaw-mlx-audio": {
"config": {
"stt": {
"enabled": true,
"model": "mlx-community/whisper-large-v3-turbo",
"port": 19290,
"language": "zh",
"pythonEnvMode": "managed"
}
}
}
}
}
}
MLX-STT: Local Whisper Speech-to-Text Data Schema & Taxonomy
The skill utilizes a structured JSON schema for both inputs and outputs to ensure compatibility within Openclaw Skills.
Input Parameters
| Parameter | Type | Description |
|---|---|---|
action |
string | Must be "transcribe", "status", or "reload" |
audioPath |
string | The absolute path to the audio file |
language |
string | Optional ISO language code (e.g., "en", "zh") |
task |
string | Optional: "transcribe" (default) or "translate" |
Output Structure
| Field | Type | Description |
|---|---|---|
success |
boolean | Indicates if the transcription succeeded |
text |
string | The full transcribed text content |
duration |
float | Total duration of the audio in seconds |
segments |
array | List of objects containing start, end, and text |
MLX-STT: Local Whisper Speech-to-Text Advanced Features
- Support for 1000+ languages via Meta's MMS model integration.
- Speaker Diarization using VibeVoice-ASR-9B for long-form audio (up to 60 minutes).
- Word-level time alignment using Qwen3-ForcedAligner.
- Real-time status monitoring and hot-reloading of configurations without restarting the agent.
- Multi-format support including MP3, WAV, M4A, FLAC, OGG, WebM, and MP4 extraction.
SKILL.md
Loading
Related Openclaw Skills
Tmall Hot Trend
A specialized automation skill for analyzing Tmall product data, pricing, and brand flagship store performance.
revibe
revibe transforms any GitHub repository into a structured architecture map with design patterns, file roles, and persistent agent context.
Tencent Hunyuan Hot Trend
A specialized tool for extracting and organizing public information, pricing, and documentation from the Tencent Hunyuan AI ecosystem.
GTM ICP Definition
A structured AI workshop skill for defining, refining, and versioning your Ideal Customer Profile to align sales and marketing teams.
Utility Payment
A comprehensive automation and guidance tool for aggregating household utility payments, managing auto-debits, and auditing digital bills.
GTM Outbound Strategy
A strategic AI agent skill designed to architect high-conversion outbound sales campaigns and craft personalized messaging sequences based on real-time triggers.
Featured*
