Responsiveness & Interruption Sensitivity
Published:
Overview
This guide explains how to configure the speech detection settings for an AI Voice Agent, including how quickly the agent responds after the caller stops speaking, how sensitive it is to interruptions, and how it detects turn-taking in conversation.
Steps to Configure Speech Detection
- Navigate to Call Flows and locate the agent you want to edit.
- Ensure you are in the Advanced View.
- Go to Voice and click Speech Detection.
Responsiveness
- Adjust the Responsiveness slider — this controls how quickly the agent starts speaking after the caller stops talking. A higher value means the agent responds faster.
Interruption Sensitivity
- Adjust the Interruption Sensitivity slider — this controls how many words the caller needs to say before the agent stops talking and yields the floor.
- A higher setting means the agent will stop talking even for brief acknowledgements like “yes” or “okay”
- A lower setting means the agent will continue speaking through short interjections and only yield when the caller says something more substantial — this reduces conversational overlap
Turn Detection Model
- Select the Turn Detection Model:
- STT-Based (Speech-to-Text) – uses speech-to-text processing to determine when the caller has finished speaking
- AI Detection – uses AI to predict turn-taking
Preemptive Generation
-
Toggle Preemptive Generation on or off. When enabled, the agent begins generating its response before the caller has finished speaking. This can reduce latency but may result in incorrect responses if the caller’s full intent was not yet captured. It is generally recommended to leave this off.
-
Click Save to Draft or Publish to Production.
Key Notes
| Item | Details |
|---|---|
| Responsiveness | Controls the delay between the caller stopping and the agent starting to speak |
| Interruption Sensitivity | Controls how reactive the agent is to short caller inputs — lower values reduce conversational overlap |
| STT-Based (Speech-to-Text) Turn Detection | Recommended default — more reliable for most use cases |
| Preemptive Generation | Can reduce latency but risks generating incorrect responses — leave off unless specifically needed |
| Advanced View | Must be enabled to access Speech Detection settings |
