Start real-time microphone transcription. The script captures audio, streams to ElevenLabs Scribe v2 Realtime via WebSocket, and writes committed text to a temp file continuously.
Start command
source ~/.claude/skills/transcribe/scripts/.env && \
ELEVENLABS_API_KEY="$ELEVENLABS_API_KEY" \
nohup python3 ~/.claude/scripts/realtime-transcribe.py > /tmp/realtime-transcribe.log 2>&1 &
Wait ~4 seconds, then read the first line of /tmp/realtime-transcribe.log for the startup JSON:
{"status": "started", "pid": 12345, "output_file": "/tmp/transcribe-20260526-143022.txt", ...}
Before starting
Check if one is already running:
test -f /tmp/realtime-transcribe.pid && kill -0 $(cat /tmp/realtime-transcribe.pid) 2>/dev/null && echo "ALREADY RUNNING"
If already running, tell the user and offer to stop first.
What to tell the user
- Transcription is running
- The output file path
- Ways to stop: say "Okay, that's enough, let's stop the transcription", or ask the agent to stop, or
/live-transcribe-stop - They can ask to read the transcript anytime
Audio cues
Pre-recorded sounds play automatically:
- Start: "I started transcription!" (before recording begins)
- Stop: "I finished the transcription." (after recording ends)
- Reminder: "I am still transcribing." (every 30 minutes)