SenseAudio ASR

Use this skill for all SenseAudio speech recognition tasks.

Credential source: read the API key from SENSEAUDIO_API_KEY and send it only in the Authorization: Bearer ... header. Do not place API keys in query parameters, logs, transcripts, or saved examples.

Read First

• references/asr.md

Workflow

1. Pick recognition mode:
2. HTTP file transcription for offline audio.
3. WebSocket for realtime streaming microphone/audio chunks.
4. Audio analysis for noise and quality checks before recognition.

5. Records query for recent recognition history lookup.

6. Choose model by feature needs:

7. Lite for low-cost basic transcription.
8. ASR for streaming, translation, diarization, sentiment, and timestamps.
9. Pro when diarization plus explicit max_speakers control is needed.

10. DeepThink for streaming, translation, and intelligent editing; do not send language, diarization, sentiment, timestamps, ITN, or punctuation controls.

11. Build minimal request:

12. Required auth, file/audio format, model.
13. Add optional controls only when needed.

14. Keep uploaded files at or below 10MB; split longer audio before sending.

15. Validate compatibility:

16. Check model-parameter support before sending.
17. Enforce WS pcm / 16000Hz / mono requirements.

18. For HTTP stream=true, expect SSE text deltas only, not structured verbose fields.

19. Parse robustly:

20. Handle JSON/text/verbose/SSE forms.
21. Handle WS terminal events and failures.
22. Treat returned audio URLs, api_key, session_id, and trace_id as sensitive operational data.