Speech technology
Voice latency
Also called: response latency · turn latency
Voice latency is the round-trip delay between a caller finishing their sentence and the agent beginning to respond. It's the sum of speech-to-text, language-model, and text-to-speech time. Sub-second latency is the bar for natural conversation.
Humans notice pauses longer than roughly 500–800 ms. High latency makes an agent feel robotic and causes callers to talk over it. Minimizing latency is a core engineering challenge in real-time voice AI.
Related terms
Speech-to-textTechnology that transcribes spoken audio into text in real time. STT is how a voice agent understands what the caller is saying.Text-to-speechTechnology that converts written text into natural spoken audio. TTS is what gives an AI voice agent its voice.EndpointingDetecting when the caller has finished speaking so the agent knows it's its turn to respond. Bad endpointing causes awkward pauses or interruptions.Barge-inThe ability for a caller to interrupt the agent mid-sentence and have it stop and listen — essential for natural, non-frustrating conversation.
See Voice latency in a real call.
Book a 30-minute demo and watch Finn handle inbound and outbound calls end to end — no stack to assemble.
Try Finn for free