How does barge-in work?
The system keeps listening while it plays audio. When it detects speech, it stops playback and treats what follows as the caller’s next turn. The W3C’s VoiceXML 2.0 standard defines two kinds: speech barge-in stops the prompt as soon as speech is detected, and hotword barge-in stops it only when the caller says something the system is listening for.
Speed matters. VoiceXML says a prompt should stop as early as possible to avoid a “stutter” effect, where callers stop mid-sentence and start again because they think the system hasn’t heard them. The IETF’s MRCPv2 protocol has a matching kill-on-barge-in setting that stops speech synthesis when the caller starts talking.
Barge-in isn’t always wanted. VoiceXML suggests turning it off when a caller must hear all of a warning or legal notice.
Example: the agent starts listing three ways to export data, the caller says “the CSV one”, and the agent stops and explains CSV export.
Why does barge-in matter for an AI voice agent?
AI agents talk more than menu prompts did, so a caller who can’t interrupt waits through answers they don’t need. Good barge-in also has to ignore sounds that aren’t interruptions: a cough, a door, or the agent’s own voice coming back through the caller’s speakers.
That last case is why echo cancellation matters in the browser. When a page requests it, the browser tries to remove the call’s incoming audio from what the microphone picks up. OpenAI’s prompting guide for GPT-Live recommends an interruption policy that tells the assistant to stop speaking when the user interrupts and listen to what they say.
How does Kawl Now handle barge-in?
Callers can talk over a Kawl Now agent at any time. There’s no push-to-talk and no setting that makes part of an answer uninterruptible. Calls run on OpenAI’s GPT-Live, which OpenAI says can respond to interruptions, and Kawl Now gives every agent the same interruption policy: stop speaking when the caller interrupts, and listen.
The widget asks the browser for echo cancellation and noise suppression, which helps stop the agent’s own voice from counting as an interruption.
Interruption handling hasn’t yet been verified in a recorded test with a person speaking into a browser microphone, so include it in your own tests: interrupt mid-answer, change the question halfway and check that the transcript shows what you said.
Common questions
What does barge-in mean in a call centre?
In contact centres, barge-in also means a supervisor joining a live call between an agent and a customer, usually after listening in. In voice systems and IVR, it means the caller interrupting a prompt.
Should a voice agent allow barge-in?
Usually, yes, because people interrupt to correct or redirect, and waiting feels slow. The usual exception is a short passage a caller must hear in full, such as a legal notice. Kawl Now has no setting that makes an answer uninterruptible, and every agent is told to stop and listen when the caller talks over it. Interruption handling hasn’t been verified in a recorded test yet, so check it, along with noise and echo, before you switch on customer calls.