Text-to-speech model
Reads text aloud in a natural voice, and some can copy a voice from a short sample.
What it is
Text-to-speech (TTS) models turn written text into spoken audio. Current models sound close to human, carry emotion and pacing, and some can clone a voice from a few seconds of reference audio.
How it works
Text is converted to phonemes or tokens, a model predicts acoustic features or audio tokens for them, and a vocoder turns those into a waveform. Newer models generate audio tokens directly with transformer or flow-matching methods, conditioned on a reference voice.
What it is good for
- Voice assistants and phone agents
- Narration for videos and audiobooks
- Accessibility and read-aloud features
- Dubbing and localization
Limits to know
- Voice cloning can be misused for fraud
- Unusual names and numbers can be mispronounced
- Expressive control is still limited
