transcribe

# Transcribe - Audio to Markdown Convert audio files to clean Markdown transcripts. Auto-selects best backend. ## Usage ``` /transcribe <audio-file> /transcribe <audio-file> --azure # Force Azure gpt-4o-mini-transcribe (best quality) /transcribe <audio-file> --groq # Force Groq whisper-large-v3-turbo (fastest) /transcribe <audio-file> --local # Force local whisper.cpp (free) ``` ## Auto-Selection Priority 1. **Azure** — if `AZURE_OPENAI_KEY` + `AZURE_OPENAI_ENDPOINT` set (gpt-4o-mini-transcribe, lowest WER) 2. **Groq** — if `GROQ_API_KEY` set (216x realtime, cheapest API) 3. **Local** — whisper.cpp fallback (free, ~1x realtime) ## Supported Formats - M4A, WAV, MP3, FLAC, OGG, WEBM ## Workflow When invoked, run: ```bash ~/.claude/skills/Transcribe/Tools/Transcribe.sh "<audio-file>" [--groq|--azure|--local] ``` Output saved to same directory as source: `{name}-transcript.md` ## Infrastructure | Component | Path / Config | |-----------|---------------| | Transcribe.sh | `~/.claude/skills/Transcribe/Tools/Transcribe.sh` | | Azure API | `AZURE_OPENAI_KEY`, `AZURE_OPENAI_ENDPOINT`, deployment: `gpt-4o-mini-transcribe` (default) with `whisper` fallback | | Groq API | `GROQ_API_KEY` env var, model: whisper-large-v3-turbo | | whisper-cli | `~/tools/whisper.cpp/build/bin/whisper-cli` | | large-v3 model | `~/tools/whisper.cpp/models/ggml-large-v3.bin` | ## Cost Comparison | Method | Cost | Speed | Quality | Best For | |--------|------|-------|---------|----------| | Azure (gpt-4o-mini-transcribe) | ~$0.30/hr | Fast | **Best** (lowest WER) | Default — best quality | | Groq (turbo) | $0.04/hr | 216x realtime | Good | Bulk/fast transcription | | Azure (whisper) | ~$0.36/hr | Fast | Good | Fallback | | Local (large-v3) | FREE | ~1x realtime | Good | Offline, full privacy | | Local (small.en) | FREE | ~3x faster | OK | Quick English drafts | ## Override Deployment Set `AZURE_TRANSCRIBE_DEPLOYMENT` env var to use a different Azure deployment (e.g., `whisper` for old behavior).