https://retellai.com/?utm_source=discord&utm_medium=seo logo
how to detect transcription issues or transcriptio...
# ❓・ai-evy-support
e
@AI Evy how to detect transcription issues or transcription accuracy in custom llm implementation? We want to transfer if users audio is low or not understandable. can we do that?
a
Thanks for tagging me @Erkam. I'll look through relevant sources and get back to you shortly.
### Answer Retell AI provides ASR (Automatic Speech Recognition) latency metrics that can help you detect transcription issues [(1) ]() . After a call ends, you can retrieve detailed latency breakdowns using the Get Call API, which includes an
asr
field showing transcription latency—the difference between the duration of audio chunks streamed and the duration of the transcribed portion . The latency object contains statistical fields including
p50
,
p90
,
p95
,
p99
,
min
,
max
,
num
, and
values
for the ASR component . Monitoring these metrics can help identify when transcription is struggling due to audio quality issues. Retell's ASR system includes noise robustness capabilities designed to filter out background noise, accents, or speech variability for clean transcriptions [(2) ]() . The platform offers two transcription modes: optimize for speed (Fast & Accurate modes) and optimize for accuracy (Accurate mode) [(3) ]() . For your specific use case of transferring calls when audio is low or not understandable, you could: 1. Monitor ASR latency metrics in real-time through the API [(1) ]() 2. Use the TransferCallTool to transfer calls when transcription quality indicators suggest poor audio [(4) ]() 3. Consider implementing custom logic in your LLM WebSocket interface to detect when transcription confidence is low The platform also supports Voice Activity Detection (VAD) which detects when the caller starts and stops speaking to avoid interruptions [(5) ]() , which can help identify audio quality issues. For more detail
ed implementation guidance specific to your custom LLM setup, I'd recommend checking out these resources: - [Check actual latency]() - [Register Phone Call]() Would you like a general answer based on my broader knowledge of how transcription quality detection typically works in voice AI systems?
​ To learn more, see the sources I considered: [[1] Check actual latency](https://docs.retellai.com/reliability/check-actual-latency) [[2] Automatic Speech Recognition (ASR)](https://www.retellai.com/glossary/automatic-speech-recognition-asr) [[3] Smarter Understanding, Clearer Calls: Retell’s ASR Just Got a Major Upgrade | Retell AI](https://www.retellai.com/blog/retell-asr) [[4] Register Phone Call](https://docs.retellai.com/api-references/register-phone-call) [[5] Speech Processing](https://www.retellai.com/glossary/speech-processing) ​
Feel free to tag @AI Evy with additional questions. I also appreciate feedback, it helps me improve.