STT Transcriptions Stop Mid-Call (LiveKit Inference + Deepgram) Hi all, We're experiencing an issue where STT transcriptions stop arriving mid-call, even though audio continues to flow (confirmed via call recordings). Issue: - Calls work normally for 2-5 minutes (multiple successful turns) - After agent finishes speaking, STT stops producing transcriptions - No interim transcripts, no final transcripts - complete silence from STT - Audio IS being recorded (waveform shows user speaking) - User can be heard saying "Are you still there?" in recordings, but agent never received transcription Setup: - LiveKit Agents with LiveKit Inference - STT: Deepgram nova-2 - Turn detection: VAD or STT Affected Rooms (today, Dec 11): - call_freeswitch_uz_g8258LGcuwYm (~11:05 UTC) - call_freeswitch_uz_quDmMibWS2t3 (~11:58 UTC) - call_freeswitch_uz_HeT4mCTLUbRC (~12:15 UTC) Pattern: 1. Agent speaks response 2. User responds (audio visible in recording) 3. Zero transcription events received This suggests the audio pipeline to STT breaks after agent speech, while the recording pipeline continues working. Any ideas what could cause this?

Hi Taha, this is addressed in our email, hope the issue is resolved

Hi Karlson, unfortunately even with the FallbackAdapter, the issue is not resolved

DM'ed