Text to Speech

Where the voices come from

The page asks the browser for its voice list and pre-selects the default English voice, or the first English voice if none is marked default, or the first voice of any language. The list arrives asynchronously, so for a moment the menu may read “Loading voices…”.

Does the text leave your device?

This site never receives the text; speech is requested from the browser's SpeechSynthesis API on the page. Whether the audio is produced locally depends on the voice. Voices built into Windows, macOS, iOS and Android render on the device. Chrome's “Google …” voices and Edge's “Online (Natural)” voices send the text to the vendor's servers and stream the audio back, which is also why they go silent offline. Pick a system voice for anything sensitive.

Speed and pitch in real numbers

At 1.0× most English voices read at roughly 150–175 words per minute, so a 300-word passage takes about two minutes. At 2.0× the same passage takes around a minute, and at 0.5× about four. The slider steps by 0.1, and the current value is printed in the label, so 1.3× is a reproducible setting rather than a guess.

Pitch 1.0 is the voice's natural pitch; 0 is as low as the engine allows and 2 as high. Network voices and newer neural voices often ignore pitch or clamp it to a narrow band, so the slider does more on the classic system voices. Both settings are fixed when playback starts: moving a slider mid-sentence changes nothing until you press Restart.

Known quirks

• Desktop Chrome with a Google network voice can stop after roughly 15 seconds on a long passage. This is a long-standing Chrome bug; choose a local voice such as a Microsoft or Apple one, or read a paragraph at a time.
• The whole box is spoken as a single utterance. There's no way to skip to a sentence; Stop and trim the text instead.
• On iPhone, the ring/silent switch and a muted tab both silence playback even though the page shows it speaking.
• Android Chrome treats Pause as a cancel on some versions, so Resume may not continue where it stopped.
• Nothing is recorded. If you need a file, record the system audio with another app while this page plays.

Matching the voice to the language

A voice reads every text in its own language's phonetics. An en-US voice given a French paragraph produces an American accent mangling French vowels, and an Arabic voice given English does the reverse. For a pronunciation check, choose a voice whose code matches the text — fr-FR, ar-SA, de-DE — and the list only contains codes your device has installed.

Listening is also a strong proofreading pass: a voice reads exactly what is on the page, so a missing word or a doubled “the the” that your eye skips becomes obvious. Setting the speed to about 1.2× keeps the pace brisk without losing the mistakes.