NLP confidence drop on BYOC inbound legs

Architect & Flows
1

NLP Confidence Score: 0.42
Can anyone clarify why the entity extraction confidence plummets when the call originates from our SG BYOC trunks compared to native Genesys numbers? The audio quality is fine, but the intent mapping fails consistently in the Architect flow.

2

The root cause here is the speech_recognition_language setting in the BYOC trunk configuration not matching the NLP model’s expected locale. Verify that the trunk’s speech_recognition_language is set to en-GB or en-US specifically, rather than a generic en code, to ensure the acoustic model aligns with the intent classification engine.

3

Check your trunk config in Terraform.

  1. Set speech_recognition_language to en-AU instead of generic en.
  2. Apply changes to refresh the acoustic model binding.