Convert video/audio content into text, and further extract the core points.
Supports:
- Local files: MP4, MOV, MP3, M4A, etc.
- Online URL: platforms supported by yt-dlp such as YouTube, Bilibili, Xiaoyuzhou, Youku, etc.
- Extraction mode: after transcription is complete, a structured summary of key points can be output
🔊 Voice Notification (must be executed first)
curl -s -X POST http://localhost:8888/notify \
-H "Content-Type: application/json" \
-d '{"message": "Running VideoTranscribe skill"}' > /dev/null 2>&1 &
Workflow Routing
| What the user says | Routes to |
|---|---|
| Transcription / speech-to-text / subtitles / sending a video file or URL | Workflows/Transcribe.md |
| Extraction / summary / core points / highlights | Workflows/Extract.md |
| Both (transcribe first, then extract) | First Transcribe.md, then Extract.md |
Quick Reference
- Script path:
~/.shared-skills/VideoTranscribe/Tools/transcribe.py - Preferred strategy: URL → capture subtitles (fast) → if no subtitles, download audio and transcribe
- Transcription engine: Gemini 2.5 Flash (with punctuation) / mlx-whisper (offline fallback)
- Output saved by default to:
~/Downloads/transcript_YYYYMMDD_HHMMSS.txt