Glossary · Conversation
What is barge-in on a voice call?
Barge-in
Barge-in is when the caller starts talking while the voice agent is still speaking, and the agent stops to listen. Without it, people have to sit through every sentence before they can answer.
On a call
The agent is reading out an order: "आपका order कल शाम तक…" The caller cuts in with "नहीं, मुझे cancel करना है." The agent stops mid-sentence and asks which order to cancel.
AI agents from ₹1 a minute.¹
Explained
Here is how it works, in plain words.
People talk over each other on the phone all the time. A caller hears half a sentence, knows the answer, and starts speaking. If the agent keeps going, the caller repeats themselves or hangs up.
Barge-in lets the agent notice and stop. It keeps listening while it speaks. When the caller’s speech looks like a real attempt to take the turn, playback stops and the rest of the sentence is thrown away. Then the agent answers what the caller actually said.
The hard part is deciding what counts. Phone lines carry traffic, television and, on speakerphone, the agent’s own voice coming back. Callers also make small listening sounds that don’t mean stop. If every sound interrupts, the agent stutters and restarts. If nothing does, it sounds like a recording.
So most systems use a threshold: how many words, or how long the caller must speak, before the agent yields. A low threshold suits a support line, where callers want to steer. A higher one suits a payment reminder, where the amount and the due date have to land.
There’s a cleanup job too. Audio already sent to the phone network has to be flushed. Otherwise the caller hears the old sentence finish after the agent has moved on. The agent’s memory needs trimming as well, so the model doesn’t believe the caller heard words that were never played.
The idea is older than AI agents. VoiceXML, the W3C standard behind many IVR menus, lets each prompt be marked interruptible or not. Its own advice is to switch barge-in off for a legal notice the caller must hear in full.
In Voice Infra
This is what Voice Infra does about it.
You choose how much a caller has to say before the agent gives way: easy, normal, firm or never. A short "haan" or "ji" never stops it, while a clear "ruko" does at every setting except never. The agent also learns how loud the caller is, so a quieter voice in the room can't cut it off.
Questions
People often ask us these questions.
Does saying "haan" interrupt the agent?
No. Short sounds like "haan", "ji" and "ok" mean the caller is listening, so the agent keeps talking. A clear stop word such as "ruko" does interrupt it.
Can I stop callers from interrupting one part of the call?
Yes. On a node-based agent, a node with allow_interrupt set to false always finishes its line, then answers what the caller said. It suits a closing summary the caller has to hear in full.
Still deciding? Ask us.
Fifteen minutes with the team that builds it. Bring your script, or your hardest call.
Book 15 minutes