100% private voice transcription using OpenAI Whisper. Audio never leaves your machine.
⚠️ Important: OpenClaw Configuration Required
This skill must be configured in OpenClaw's tools.media.audio to handle incoming voice messages. Without this config, voice messages may cause token overflow errors or be sent to cloud APIs.
Add to ~/.openclaw/openclaw.json:
{
tools: {
media: {
audio: {
enabled: true,
models: [
{
type: "cli",
command: "node",
args: ["<skill-path>/transcribe.js", "{{MediaPath}}"]
}
]
}
}
}
}
Replace <skill-path> with the path to the checked-out skill repository. For a mirrored OpenClaw install, use ~/.openclaw/skills/local-whisper.
Quick Start
# Install dependencies
pip install openai-whisper
# Transcribe audio
node transcribe.js voice.ogg
CLI Options
--model <tiny|base|small|medium|large> Model size (default: small)
--language <lang> Language code (default: auto)
--output-dir <dir> Output directory
--smart-model Auto-select model by file size
--check Verify dependencies
Model Sizes
| Model | Size | Speed | RAM |
|---|---|---|---|
| tiny | 39 MB | ⚡⚡⚡⚡ | ~1GB |
| base | 74 MB | ⚡⚡⚡ | ~1GB |
| small | 244 MB | ⚡⚡ | ~2GB |
| medium | 769 MB | ⚡ | ~5GB |
| large | 1550 MB | 🐢 | ~10GB |