Firefox 160 ships on-device speech-to-text, no cloud needed

Firefox 160 shipped on 10 November with on-device support for the Web Speech API’s SpeechRecognition interface on desktop. Call it and Firefox converts speech to text in real time, in 40 languages, without sending any audio to a server. It is on by default, no flag required.

The caveats are specific: desktop only, nothing on Android, and no cloud fallback either, so there is exactly one way this runs. Mozilla’s release notes point to the same change swapping in Apple’s Metal API for WebGL on macOS, and adding support for the legacy ::-webkit-input-placeholder selector as an alias for ::placeholder.

const recognition = new SpeechRecognition();
recognition.lang = 'en-US';
recognition.onresult = (event) => {
  console.log(event.results[0][0].transcript);
};
recognition.start();

Why it matters: a voice feature used to mean wiring up a paid transcription API or shipping your own model, just to get text out of a microphone. A standards-based API that does that locally, for free, in 40 languages and with no server round-trip, removes that step for anything that only needs to run in Firefox on desktop.

The caveat: this is Firefox-only behaviour for now. Other browsers support the same SpeechRecognition interface but with differing degrees of on-device support, so cross-browser code still needs testing per browser rather than assuming local-only processing everywhere.