Occasional caller speech (full sentences) not transcribed — agent misses lines and repeats questions

Support Help
1

Occasionally, full sentences spoken by the caller are audibly present in the call recording but are completely absent from the transcript. This causes the agent to miss caller input, repeat questions, and appear slow/unresponsive.

Current Realtime Transcription Settings:

  • Denoising Mode: Remove noise

  • Transcription Mode: Optimize for accuracy

  • Vocabulary Specialization: Medical

  • Boosted Keywords: (none set)

Observation:

  • Audio is clearly audible in the recording — the issue is at the transcription layer, not audio capture

  • Dropped content is full sentences, with no consistent pattern identified yet

  • Occurs occasionally, not on every call

Impact: Agent repeats questions already answered by caller; caller perceives agent as slow or unresponsive.

2

I couldn’t find the exact call_id for this one
but i found another one with slight differently issue but same impact of looping . where it didn’t skip entire line but captured part of response in transcription .
call_id call_fc3f5b5b0aafa2ac867d6b6942f
check starting from mark 1:25

3

Hey @phonavar

I’ve shared your details with our team for review.

We’ll get back to you as soon as we have an update.

Best regards

4

Hey @phonavar the audio was a bit low, so the ASR couldn’t transcribe it fully. What you can do is:

  • Change the Transcription Mode to Custom and increase the endpointing

  • Add boosted keywords for your key terms (e.g., refill, oncolytic, ezetimibe).

  • Optionally experiment with denoising modes per our docs: (Transcription tuning guide)

Thank You