A high-performance local speech-to-text utility utilizing Parakeet and Whisper backends for ultra-fast, offline audio transcription.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install local-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 local-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 Local STT (Parakeet / Whisper)?
Local STT is a specialized module for Openclaw Skills designed to provide fast, private, and accurate speech-to-text capabilities without relying on cloud APIs. By utilizing ONNX Runtime with int8 quantization, it ensures minimal resource usage while maintaining high fidelity. Users can toggle between Parakeet for superior English accuracy and Whisper for broad multilingual support, making it a versatile choice for any local AI agent workflow powered by Openclaw Skills.
This skill is particularly valuable for developers who prioritize data sovereignty and low-latency processing. As part of the Openclaw Skills library, it offers a unified interface for multiple STT engines, allowing for seamless backend switching depending on whether your priority is capturing nuances in English dialogue or transcribing one of 99 different languages.
Local STT (Parakeet / Whisper) Use Cases
- Transcribing voice notes or audio files directly from the command line without sending data to external servers.
- Providing real-time audio processing for AI agents within the Openclaw Skills framework.
- Integrating private transcription into Matrix chat rooms via room-id tagging.
- Building multilingual voice interfaces that require offline processing for security and speed.
- Benchmarking different STT models to find the optimal balance of accuracy and inference time.
How Local STT (Parakeet / Whisper) Works
- The user provides an audio file path to the local-stt script via the command line or an automated agent.
- The system checks for the required FFmpeg binary to process and normalize the audio stream.
- The skill loads the selected model (Parakeet or Whisper) using the optimized ONNX Runtime engine.
- Audio is transcribed locally with optional int8 quantization to maximize inference speed and reduce memory footprint.
- The resulting text is returned as a string to the console or sent to a designated Matrix room for further use.
Local STT (Parakeet / Whisper) Setup
To use this within your Openclaw Skills environment, ensure you have FFmpeg installed on your system. You can then configure your agent to call the script.
# Install FFmpeg (Ubuntu example)
sudo apt update && sudo apt install ffmpeg
# Run a test transcription with the default Parakeet model
~/.openclaw/skills/local-stt/scripts/local-stt.py path/to/audio.ogg
# Run using the Whisper backend for multilingual support
~/.openclaw/skills/local-stt/scripts/local-stt.py path/to/audio.ogg -b whisper
Add the tool to your openclaw.json to enable it for your AI agent:
{
"tools": {
"media": {
"audio": {
"enabled": true,
"models": [
{
"type": "cli",
"command": "~/.openclaw/skills/local-stt/scripts/local-stt.py",
"args": ["--quiet", "{{MediaPath}}"],
"timeoutSeconds": 30
}
]
}
}
}
}
Local STT (Parakeet / Whisper) Data Schema & Taxonomy
The skill processes various audio formats and outputs plain text. Model selection influences the output quality and language support within the Openclaw Skills ecosystem.
| Feature | Details |
|---|---|
| Supported Backends | Parakeet (Best English accuracy), Whisper (Fastest, 99 languages) |
| Quantization | int8 (enabled by default, can be disabled with --no-int8) |
| Model Variants | Parakeet (v2, v3); Whisper (tiny, base, small, large-v3-turbo) |
| Metadata Requirements | Requires ffmpeg for audio processing |
| Output | Plain text transcription via STDOUT or Matrix API |
Local STT (Parakeet / Whisper) Advanced Features
- Choose between different model sizes, from 'tiny' for speed to 'large-v3-turbo' for maximum quality.
- Native support for int8 quantization, allowing for significant speedups (e.g., 0.43s transcription for 24s audio).
- Seamless integration with Matrix for direct-to-chat messaging using the
--room-idflag. - Quiet mode for clean CLI output, ideal for piping into other Openclaw Skills.
- Flexible backend selection to switch between English-optimized Parakeet and multilingual Whisper on the fly.
SKILL.md
Loading
Related Openclaw Skills
Gemini Speech-to-Text
A high-performance transcription utility that leverages Google's Gemini models to convert audio files into text with zero external dependencies.
GhostBot ACLM
An AI-powered hook system for Uniswap v4 that automates liquidity rebalancing and dynamic fee optimization.
GhostBot ACLM: Automated Concentrated Liquidity Manager
An AI-powered Uniswap v4 hook system that automates liquidity provision, rebalancing, and fee optimization on Ethereum Sepolia.
Resend Email Sender
A streamlined email integration for Openclaw Skills that enables sending professional text and HTML emails via the Resend API without managing SMTP servers.
Local Whisper STT
A private, high-accuracy offline speech-to-text engine for the Openclaw Skills ecosystem.
Markdown to PDF (md-to-pdf)
A streamlined utility to transform markdown files into professional, high-quality PDF documents using reportlab.
METADATA
Featured*
