Skip to main content

STT — Read What the Mic Heard

STT runs continuously as part of the voice pipeline — you generally don't call it to transcribe an ad-hoc WAV. Instead:

  • Live turn stream (VAD → STT → LLM → TTS): subscribe to the OS Server flow monitor SSE endpoint — see docs/flow-monitor.md.
  • Current voice-pipeline state (STT provider, language, VAD, is speaking): GET /voice/status.
  • Live mic RMS level (SSE, useful for a "am I hearing you?" indicator): GET /voice/mic-level.
curl http://127.0.0.1:5001/voice/status
curl -N http://127.0.0.1:5001/voice/mic-level   # SSE stream

The STT drivers (Autonomous, Deepgram) live under hal/drivers/voice/stt/ — provider.py is the interface to implement if you need to plug a new one in.