Text-to-Speech

What is text-to-speech?

Speech recognition turns sound into text; text-to-speech (TTS) runs it in reverse. Hand it some text and it hands back natural-sounding speech. The voice in your phone's navigation, audiobooks and AI customer service is very often TTS doing the talking.

How does it make sound?

The old way: snapping blocks together
Early systems stitched recorded syllables and words like building blocks. It worked, but it was stiff — clearly a machine.
The new way: end-to-end generation
Modern neural networks generate the waveform straight from text, no stitching. The model reads the text, predicts pitch, rhythm and pauses, then synthesizes continuous speech.

How does "natural" happen?

Prosody
Humans speak with ups and downs. Good TTS reads punctuation and meaning to decide where to pause and where to stress.
Voice cloning
Feed it a recording and the model can mimic that person's voice. That's voice cloning — and the source of deepfake audio.
Emotion
More models now read with joy, surprise or seriousness instead of a flat monotone.

Where does it show up?

Spoken content
Articles and e-books read aloud to you.
Accessibility
Screen readers give vision-impaired people access to on-screen text.
Voice assistants
The voice that answers you back is generated by TTS.

What's the risk?

Voice cloning makes it easy to fake someone speaking, which scammers have already used. That's why more platforms now watermark synthetic voices or require disclosure.

Bottom line: text-to-speech takes machines from reading to actually talking.