How text to speech works

See it for yourself

Your browser's own voices, live. Pick one, set the pace, press play.

Text to speech gives your eyes a rest without losing the thread. Change the speed and hear how the same sentence finds a different rhythm.

How it works

First, work out the words

Text is messier than it looks. Before a machine can say anything, it has to decide what the words actually are: whether "Dr." means doctor or drive, whether "St." is a saint or a street, how to read a year, a phone number, or a line of dashes. Get this stage wrong and the voice reads on confidently — and says nonsense.

Then, decide how it should sound

The same words can be read a hundred ways. This stage — prosody — chooses the pitch, the pace, and the phrasing: where to breathe, which word to lean on, how a question rises at the end. It is the difference between a voice you can listen to for an hour and one that wears you out in a minute. Move the speed slider in the demo above and you are adjusting this stage by hand.

Finally, make the sound

Last comes the audio itself. Older systems built it from hand-written rules, or by splicing together tiny fragments of recorded speech — which is why they sound stitched together. Newer neural voices are models trained on recorded speech that generate the waveform directly, and the joins smooth out. The demo on this page uses whatever voices your browser ships through the Web Speech API, so what you hear depends on your device — try a few, and you can often hear the generations side by side.

Where it came from

The first electronic speech synthesizer could not read anything. Homer Dudley's Voder, built at Bell Labs and demonstrated at the 1939 World's Fair, was played like an instrument: a trained operator shaped its electrical buzz and hiss into words in real time, and crowds lined up to hear a machine talk. The machine made the sound — a person still made the speech.

The step that matters most for this page came in 1976. The Kurzweil Reading Machine combined optical character recognition with speech synthesis, so a blind reader could place a printed page on it and hear the words. That pairing — recognize the text, then speak it — is still the basic shape of every read-aloud tool, including the demo above.

In the 1980s, DECtalk made synthesis widely available. Its voices were unmistakably robotic, but they were clear — and clear was what counted, because for the first time machine speech was something ordinary people could put to daily use.

The modern era began in 2016 with WaveNet, a neural network that generates the audio waveform itself instead of assembling it from rules or recordings. Voices stopped sounding like machines doing an impression of a person. From an operator playing hisses at a fair to a model that learned speech from speech — the whole arc is the three stages above, each getting better in turn.

Who it helps — honestly

The clearest case is the reader for whom decoding is expensive. If working out the words takes most of your effort — a hard page, a tired brain, a second language — there is not much left over for meaning. Listening while you read changes that: the voice carries the words, your eyes follow along, and your attention stays anchored to the sentence instead of draining away inside it.

It is also a quietly excellent proofreading tool. Your eye reads what you meant to write; a voice reads what you actually wrote. Hear your own draft aloud and the missing word, the doubled "the," and the sentence that never ends all announce themselves in a way silent rereading rarely manages.

Be honest about the limits, though. Text to speech is a support, not a shortcut past reading itself — listening to a page is not the same as learning to decode one, and a struggling reader still needs to be taught. What it does is keep someone inside real text, keeping up with school or work, while that learning happens.

One practical gap is worth knowing about: not every browser has a read-aloud button of its own on every page — Firefox keeps its Narrate voice inside Reader View. Helperbird adds one there, and highlights each word as it is spoken, so your ears and your eyes stay on the same word at the same time.

Try it in Helperbird

Helperbird brings these tools to any website, PDF, or Google Doc — no copying text anywhere, and nothing you read leaves your browser.

Text to speech Read any page aloud with word highlighting, in Chrome, Edge, Firefox, and Safari.

Step-by-step: How to use text to speech on any website

Questions people ask

Is there text to speech in Firefox?

Firefox's only built-in read-aloud is Narrate, which lives inside Reader View and works on article pages alone. Helperbird adds read-aloud on any page in Firefox, and in Chrome, Edge, and Safari.

Can you do text to speech on Google Slides?

Yes. Helperbird reads Google Docs on the page, and for Google Slides it prepares a readable view of the deck and speaks that — so a presentation can be listened to without copying anything out.

Why do some voices sound robotic and others human?

Older systems glue together tiny recorded fragments or generate sound from hand-written rules, which is fast but stiff. Newer neural voices are models trained on speech that generate the audio waveform itself — smoother, at the cost of more computing.