Endpointing
Endpointing is the step in which a voice system decides that the caller has finished speaking and that it is its turn to reply.
Endpointing is how a voice agent decides that the caller has finished speaking and that it is now its turn to respond. People sense this intuitively from intonation, meaning and short pauses; software has to make the call with explicit rules or a model. The timing of that decision has a direct effect on whether a conversation feels natural. It is one of the less visible but more important parts of latency.
How it works
The simplest method is a silence threshold: voice activity detection (VAD) notices that the caller's voice has stopped, and after a set period of silence the turn passes to the agent. More advanced systems add turn-detection models that look at what has been said, so they keep waiting after an unfinished phrase such as "My address is Atatürk Avenue..." even if there is a pause. Speech-to-text also waits for this decision before finalising the last words.
Why it matters
Deciding too early cuts the caller off, which is especially frustrating for people who pause while reading out a phone number, an address or an ID number. Deciding too late leaves a long gap before every reply and makes the conversation drag. In Turkish, the verb usually comes at the end of the sentence and speakers often fill pauses with sounds like "şey" or "ııı" while thinking, so this balance needs to be tested with real callers.
In Speakline
In Speakline, timing settings such as how long the agent waits for the end of a turn are part of the agent settings. There you also decide whether callers can interrupt the agent (barge-in) and how many words it takes before the agent stops talking. Before going live, you can test the agent and listen to its pacing yourself.