One-command YouTube video transcription for Claude Code. Works even when YouTube subtitles are disabled.
- Fast path: Tries to fetch existing YouTube subtitles (free, instant)
- Fallback: Downloads audio → compresses → transcribes via OpenAI Whisper API
YouTube URL → [subtitles available?] → YES → clean transcript ✓
→ NO → download audio → Whisper API → transcript ✓
npx clawhub install youtube-transcriberOr install directly from GitHub:
# Clone to your Claude Code skills directory
git clone https://github.com/EdisonChenAI/youtube-transcriber.git ~/.claude/skills/youtube-transcriber- yt-dlp:
brew install yt-dlporpip install yt-dlp - ffmpeg:
brew install ffmpeg - OPENAI_API_KEY: Required only when subtitles are unavailable
# Basic usage
./scripts/transcribe.sh "https://www.youtube.com/watch?v=VIDEO_ID"
# Chinese video
./scripts/transcribe.sh "https://www.youtube.com/watch?v=VIDEO_ID" --lang zh
# Custom output path
./scripts/transcribe.sh "https://www.youtube.com/watch?v=VIDEO_ID" --out ./transcript.txt
# Force Whisper API (skip subtitle check)
./scripts/transcribe.sh "https://www.youtube.com/watch?v=VIDEO_ID" --force-whisper
# Keep the downloaded audio file
./scripts/transcribe.sh "https://www.youtube.com/watch?v=VIDEO_ID" --keep-audio| Option | Description | Default |
|---|---|---|
--lang <code> |
Language hint (en, zh, ja, ko, etc.) | auto-detect |
--out <path> |
Output transcript file path | /tmp/yt_transcript_<ID>.txt |
--force-whisper |
Always use Whisper API | off |
--keep-audio |
Keep downloaded audio | off |
--audio-bitrate <kbps> |
Compression bitrate | 64 |
| Method | Cost | Speed |
|---|---|---|
| Subtitles available | Free | Instant |
| Whisper API | ~$0.006/min | ~30s for 20min video |
A typical 20-minute video costs about $0.12 via Whisper API.
YouTube lets creators disable subtitles/captions. When that happens, tools that rely on YouTube's transcript API simply fail. This skill solves that by downloading the audio and running it through OpenAI's Whisper speech-to-text model.
Built with Claude Code via vibe coding.
MIT