Latency in voice AI
Latency in voice AI is the delay between a caller finishing speaking and the agent starting to reply, the gap that decides whether a conversation feels natural.
In conversation people take turns quickly: across ten languages the average gap between one speaker finishing and the next starting is about 200 milliseconds (Stivers et al., PNAS 2009). Gaps much longer than that make an agent feel slow or make callers repeat themselves.
A voice agent’s latency is the sum of its chain: the phone network, turn detection, speech-to-text, the language model, text-to-speech and the network again on the way back. Teams reduce it with streaming at every step, shorter model responses, servers close to the caller, and realtime models that combine steps.
VocalStack Voice publishes its own end-to-end figures, phone to phone: median 247 ms and P95 under 300 ms, measured at five capture points. Vendors measure latency differently, so compare on your own calls.
