Voice latency

Voice latency is the total time between the caller stopping speaking and the agent starting to answer. It adds up the chain's stages: end-of-speech detection, transcription, model reasoning, synthesis. Human perception is unforgiving: under roughly a second, the exchange feels fluid; beyond 1.5 seconds, it feels sluggish; beyond two, the caller talks over it and the conversation desynchronizes.

Latency is thus the nerve center of voice AI: a brilliant but slow agent loses to an average but snappy one. The techniques that reduce it are known — streaming transcription, fast-response models, synthesized speech started on the first words. The buying test is simple: call the demo agent and stopwatch the reaction time on an ordinary question. If you have time to look at your watch, your customers have time to hang up.