AI Voice Agent
Only at roughly a hundred milliseconds of latency does speech start to feel like conversation. This covers which conversions end-to-end speech-to-speech skips compared with the ASR to TTS chain, and how turn-taking plus tool use move voice from talking to getting things done.
The old chain transcribes, reasons, then speaks: slow and broken by interruption. End-to-end speech-to-speech folds those stages into one; Chroma 1.0 answers inside 150 ms, Step-Audio-R1.1 reaches first audio in about 1.51 seconds. Two things then decide the experience: turn-taking, which judges when to reply or keep listening, and tool use, so voice can search, filter and book. The measure shifts from a human-sounding voice to a finished task.
