Local Whisper

🎙️ Local Whisper Transcription

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

Documentation

• Installation Guide
• Troubleshooting
• GitHub Repository