What is Streaming TTS?

Streaming TTS is text to speech that starts sending audio while the rest of the reply is still being generated, instead of waiting for the full text before speaking. In a voice agent the language model produces its reply a few words at a time. Streaming TTS begins synthesising after the first phrase, so the agent starts talking long before the full reply exists. The difficulty is prosody: the synthesiser has to commit to an intonation before it knows how the sentence ends, which can make a question sound like a statement, and mixed-language sentences complicate where to break the text into chunks.

Without streaming, a long reply would add seconds of silence before the agent speaks, and callers read silence as a dropped line. Streaming is one of the standard techniques that bring voice agents into the one-to-two-second response range that feels conversational; Outpero's end-to-end latency is 0.8 to 1.3 seconds on average from the caller stopping to the agent speaking.