TTS (text-to-speech)

TTS (text-to-speech) is the technology that converts written text into audible speech. The historical synthetic voices — robotic, flat, instantly identifiable — have given way to neural voices, generated by models trained on hours of human speech: they respect rhythm, intonation, and pauses. In a voice chain, TTS is the output component: speech recognition (ASR) listens, the language model (LLM) decides, TTS speaks.

For a phone agent, TTS is judged on three criteria that are rarely combined: naturalness (a voice you stop noticing), latency (the gap between the end of the caller's sentence and the start of the reply — past one second, the conversation turns uncomfortable), and behavior under interruption (the caller talks over it, the voice must stop — barge-in). One watchpoint that is becoming a legal one: explicitly disclosing that the caller is talking to an automated system is required in several jurisdictions, and imitating a specific human voice without authorization is prohibited.