Realtime audio
Phase 3 — push-to-talk via MediaRecorder → /audio/chunk OR live via WebRTC to OpenAI Realtime. Both paths feed transcripts into /cue. Latency badge validates the senior_translation <1000ms target.
Mode
Push-to-talk (recorded path)
Live (Realtime / WebRTC)
connect to start
Latest cue
Capture audio (push-to-talk or live) to generate a cue.