Text to Speech

Where the voices come from

There is no voice library on this site. The Web Speech API asks your browser which voices it has, and they come from two places: the operating system's own speech engine, and voices the browser adds itself. That is why the list differs on every device — Windows offers voices such as Microsoft David and Zira, Macs and iPhones voices such as Samantha and Daniel, Android whatever its speech engine provides, and Linux browsers whatever speech-dispatcher exposes, often espeak-ng.

Chrome adds a set of voices named Google, and Edge adds Microsoft voices labelled Online (Natural). They sound better than many system voices — and they do not run on your device.

On this device or online: the privacy difference

Every voice reports whether it runs locally, and the list shows it. An on-device voice is spoken by your operating system: the text goes from this page to the system's speech engine and no further. An online voice sends the text to a speech service — Google's for Chrome's Google voices, Microsoft's for Edge's Online (Natural) voices — which turns it into audio and sends it back. This page itself sends nothing either way; it is the browser that makes the request.

For a draft email, a medical letter, a contract or anything else private, choose a voice marked on this device. The tool defaults to one whenever one exists, and remembers your voice and speed in this browser only. The on-device or online label is the browser's own report; a web page has no way to check it independently.

Firefox is the one wrinkle. It marks voices that reach it through Linux's speech-dispatcher as not local, yet Mozilla runs no speech service and those voices are normally synthesised on the machine by espeak-ng — so the list calls them system speech service rather than online.

Why it reads sentence by sentence

Text is handed to the speech engine one sentence at a time. That drives the highlighting and the progress bar, and it avoids a long-standing Chrome bug: with its online voices, a single long utterance silently stops after roughly fifteen seconds. Short pieces never reach that limit, so long documents are read to the end. Very long sentences are split at a comma or a space, and abbreviations such as Dr. and e.g. are not mistaken for the end of a sentence.

Pause stops at the current word, and Resume picks up from there. Changes to the voice, speed or pitch apply from the next sentence. If a voice does not report word positions — Chrome's online Google voices do not — the current sentence is highlighted instead of each word.

Speed, pitch and a good reading

Speed runs from half to twice the voice's normal pace, and 1.0 is roughly conversational. For text you know well, many listeners are comfortable at 1.3 to 1.5×; for proofreading, slower works better. Pitch moves the voice up or down, though some voices ignore it.

Punctuation shapes the reading: commas and full stops make the pauses, and a missing full stop runs two sentences together. Numbers, dates, abbreviations and symbols are read differently by different voices — 1/2 may come out as "one slash two" — so write out anything that must be read a particular way. A voice reads its own language well and others badly, which is why the Language list groups them: match the voice to the text.

Hearing a text read aloud is also one of the most reliable ways to proofread it. The ear catches missing words, repeated words and clumsy sentences that the eye skims straight over.

No audio download, and why

Browsers only let web pages play speech through your speakers — the Web Speech API cannot save it to a file — so there is no MP3 button here. Sites that offer one with these voices either record your speakers or send your text to their own server to be synthesised, which is exactly the privacy trade this tool avoids.

If you need a file, your operating system can make one. On a Mac, the Terminal command say -o speech.aiff -f text.txt reads a text file into an audio file with the system voice. On Windows, PowerShell's System.Speech synthesizer writes a WAV with SetOutputToWaveFile. Both run entirely on your computer.