Text-to-speech (TTS)
Text-to-speech (TTS) is technology that converts written text into spoken audio. It gives software a voice by turning words, punctuation, and pronunciation cues into an audible response.
How text-to-speech works
A TTS system first prepares the input so it can be spoken. This may include expanding abbreviations, interpreting numbers, separating sentences, and determining how unfamiliar terms should be pronounced. The system then represents the text as sounds and predicts timing, pitch, emphasis, and other acoustic details. Finally, it produces an audio waveform for playback.
The exact process varies by system. Some voices offer controls for speaking rate, pauses, emphasis, or pronunciation, while others accept plain text only. A pronunciation dictionary or markup such as SSML can provide additional direction when the speech engine supports it.
Why TTS matters for AI phone calls
In a typical AI phone agent, speech recognition turns the caller's audio into text, the conversation system decides what to say, and TTS speaks that response back to the caller. The generated audio therefore shapes how the caller experiences every answer, clarification, and handoff.
Voice quality is only part of the evaluation. A phone agent also needs speech that begins promptly, remains understandable over telephone audio, pronounces important names correctly, and stops cleanly when a caller interrupts. Pacing matters as well: long pauses can sound like a dropped call, while speech with no pauses can make instructions difficult to follow.
Teams should test TTS with the language, phone connection, and vocabulary their callers will actually use. Product names, addresses, abbreviations, dates, and confirmation numbers often expose pronunciation or pacing problems that ordinary sample sentences do not.
In practice on ThunderPhone
ThunderPhone's generated catalog contains 123 voices across 47 voice languages. An agent begins in its primary language and can switch automatically when a caller uses a configured additional language. The selected voice must support the agent's configured language set, so voice choice and language configuration should be tested together.