What is text to speech, and how does it work?

Updated October 31, 2026

Text to speech, often shortened to TTS, is technology that reads written text aloud in a synthetic voice. You give it words; it gives you audio. Here is what happens in between.

The short answer

A text to speech system works in two main stages. First, it works out what to say: how each word is pronounced, where the sentences break, and which words to stress. Then it works out how to say it: it generates the actual sound wave of a voice speaking those sounds with a natural rhythm and melody.

Stage 1: understanding the text

Written text is full of things that aren't spelled the way they are spoken. Before a computer can speak, it has to clean that up. This is usually called text normalization:

• Numbers: "1999" can be "nineteen ninety-nine" (a year) or "one thousand nine hundred ninety-nine" (a quantity).
• Abbreviations: "Dr." might be "doctor" or "drive". "St." might be "saint" or "street".
• Symbols: "$5", "50%", "3:30" and "&" all need to become words.
• Dates and times: "04/05" means April 5 in some countries and May 4 in others.

Next, the system converts words into phonemes, the individual sounds of a language. It looks most words up in a pronunciation dictionary, and uses rules or a trained model to guess the rest, such as new names and invented words. This step is called grapheme-to-phoneme conversion. You can read more in Phonemes: the building blocks of spoken language.

Finally, it predicts prosody: the rhythm, stress and intonation that make speech sound like a person and not a list of sounds. A question usually rises in pitch at the end; important words get more stress; commas and full stops create pauses.

Stage 2: making the sound

With a sequence of sounds and a plan for how to say them, the system generates audio. Over the years there have been three main ways to do this:

• Formant synthesis builds sound from a mathematical model of the human vocal tract. It is fast, tiny and very clear, but sounds robotic. eSpeak NG, which covers more than 100 languages, works this way.
• Concatenative synthesis stitches together short pieces of recordings of a real person. It can sound natural, but joins are sometimes audible.
• Neural synthesis uses a neural network trained on hours of recorded speech to produce the sound directly. This is how today's most natural voices work, including the Piper and Kokoro voices on our site.

We go into more detail in Robotic vs natural voices.

Where text to speech is used

• Accessibility: screen readers for blind and low-vision people, and reading support for people with dyslexia. See who text to speech helps.
• Websites: "listen to this article" buttons. See why your site should have one.
• Learning: hearing the pronunciation of a new language.
• Everyday devices: GPS directions, voice assistants, phone menus, announcements in stations and airports.
• Content: narration for videos, presentations, e-learning and podcasts.
• Proofreading: hearing your own writing read back is one of the best ways to catch mistakes.

Text to speech vs speech to text

They are opposites. Text to speech turns writing into a voice. Speech to text (speech recognition or dictation) turns a voice into writing. They use related ideas about language and sound, but they are separate technologies.

Try it

The quickest way to understand text to speech is to hear it. Paste a paragraph into our free text to speech tool, pick a language and voice, and press Convert. Try one of the premium voices, then a standard one, and you will hear the difference between neural and formant synthesis right away.