OpenAI Whisper
Whisper runs OpenAI's speech-to-text model locally on your machine. Fast, accurate transcription across dozens of languages — audio never leaves your device.
Install:
npx clawhub@latest install openai-whisperWhat it does
Transcribes audio files to text using the Whisper model running locally. Supports .mp3, .mp4, .m4a, .wav, .webm, and most common audio formats.
| Use case | Example |
|---|---|
| Meeting notes | Transcribe a Zoom/Meet recording to a structured notes file |
| Voice memos | Convert voice recordings to searchable text in MEMORY.md |
| Podcast clips | Pull quotes from an episode without manual listening |
| Interviews | Transcribe a recorded interview for a written piece |
| Dictation | Speak your thoughts, get a draft document |
Basic usage
Transcribe the audio file at ~/recordings/meeting-2026-05-25.m4aTranscribe ~/voice-memo.m4a and save the result to ~/notes/2026-05-25-memo.mdTranscribe this meeting recording and format the output as structured notes with: attendees, key decisions, and action items.
~/recordings/team-standup.mp3Local processing
Whisper runs fully on your machine — no audio is sent to OpenAI's servers or any external service. This matters for:
- Confidential meetings and calls
- Legal or medical recordings
- Any audio you wouldn't want uploaded to a third party
The first run downloads the Whisper model weights to your machine (roughly 1–3 GB depending on model size). Subsequent runs are fast and fully offline.
Pair with other skills
- Morning brief — transcribe your morning voice note and include the summary in the brief
- Memory — save transcriptions to dated files in
~/memory/for searchable recall - Humanizer — transcriptions often sound natural already; use Humanizer to clean up transcribed text before publishing
- Nano PDF — transcribe a recorded presentation, then edit the accompanying PDF to match