Paste text and hear it read using the voices already on your device. Nothing is installed and nothing is downloaded, because your browser and operating system already do this — which is also why the voice list looks different on Windows, macOS and Android. Voices that run locally are separated from those marked online, and that distinction is worth reading: online voices are synthesised on your browser vendor's servers, so the text is sent to them. Local ones never leave your machine. One thing no browser tool can do is give you an audio file — the sound goes straight to your speakers and the page is not allowed to capture it.
| Local | Online | |
|---|---|---|
| Text leaves your device | No | Yes |
| Works offline | Yes | No |
| Quality | Varies by system | Usually better |
| Availability | Always | Needs a connection |
No, and no browser-based tool can. Speech synthesis plays through the system audio path and the page is given no access to the samples, so there is nothing to save. Recording your own machine's output is the only route.
It depends on the voice. Voices that run locally on your device never transmit anything. Voices marked online are synthesised on your browser vendor's servers, so the text goes to them — which is why the two are labelled separately here rather than mixed into one list.
Because the voices come from your operating system, not from this page. Each platform ships its own set, and some let you install more through system settings.
Voices load asynchronously in most browsers. The list fills in a moment after the page loads; if it stays empty, your browser may have no voices installed.
System voices use general pronunciation rules and get names, acronyms and loanwords wrong. Respelling a word phonetically in the text usually fixes it for that reading.
Some browsers stop after a few minutes of continuous speech. If a long passage cuts off, read it in sections.
This uses the browser's own SpeechSynthesis API, which means no model is downloaded, nothing is installed and playback starts instantly. The voices come from your operating system, so the list differs on Windows, macOS, Android and Linux — and it loads asynchronously, which is why a naive implementation shows an empty dropdown: the first synchronous call to getVoices almost always returns nothing, and the voiceschanged event has to be waited for. Voices are grouped by whether they run locally, and that distinction is surfaced rather than hidden, because the non-local ones are synthesised on the browser vendor's servers, which means the text is sent to them. For a tool on a site whose premise is that nothing leaves your device, quietly routing text to Google or Microsoft would be a misrepresentation. The other honest limitation is that there is no download. Browsers speak through the system audio path and give the page no access to the resulting samples, so no client-side tool can produce an audio file — anything offering one is either recording server-side or using a different engine entirely.