What is a Voice Agent?

A voice agent pipeline typically chains three components: speech-to-text that transcribes the caller's audio, a language model that interprets intent and decides on a response or tool call, and text-to-speech that renders the reply as audio, all tuned for low enough latency that the exchange feels like a real conversation rather than a series of delayed messages. More advanced systems handle interruptions, where the user talks over the agent, turn-taking cues, and streaming partial responses so the agent can start speaking before it has finished formulating a full answer. Voice agents matter for use cases where typing isn't practical or a phone call is the expected channel, such as customer support lines, appointment scheduling, or intake calls, and they raise their own set of concerns beyond text agents, including accent and background-noise robustness, latency budgets tight enough to avoid awkward silence, and clear disclosure that the caller is speaking with an AI system where that's required by law.