Barge-in

Without barge-in, a caller who already knows what they want has to sit through a full prompt before they can say anything. That is the single most recognisable trait of an old phone menu, and people hate it enough to press zero on reflex.

Implementing it means running speech detection while audio is still playing out, then cutting the outbound audio the moment real speech is detected rather than waiting for the sentence to finish. The hard part is not the cutting; it is deciding what counts as speech. A cough, a car horn or a second person talking in the room will all trigger a naive detector, and an agent that stops mid-sentence every time a door slams is worse than one that never stops at all.