Both, the had a cloud-based path for a long time, but they have shipped an on-device implementation for probably more than a year now.
In my own (limited) testing, the on-device implementation is generally preferred. I'm not entirely certain what technical solution is employed to perform the actual recognition, and it is bound to evolve, with useful feature we'll add to the spec. The standard allows for a wide variety of backends and some latitude in quality/features. Google certainly has nice in-house solutions for various performance envelopes. Paul. On Wed, Sep 16, 2026 at 5:26 PM Tom Ritter <[email protected]> wrote: > Is Blink's implementation also on-device or is it using Google's backend > (or maybe their JS AI API that goes to Gemini in the backend...?) > > On Wed, Sep 16, 2026 at 7:49 AM 'Paul Adenot' via [email protected] > <[email protected]> wrote: > >> Summary: >> This is a _strictly_ on device implementation of the SpeechRecognition >> interface. >> The speech synthesis counterpart defined in this specification is untouched. >> >> This can turn audio, directly from the microphone or a same-origin >> MediaStreamTrack into timed text, in real-time, with low-latency. >> >> Bug: >> Implementgation: >> https://bugzilla.mozilla.org/show_bug.cgi?id=webspeechondevice >> Activation in Nighly: https://bugzilla.mozilla.org/show_bug.cgi?id=2069803 >> Shipping: https://bugzilla.mozilla.org/show_bug.cgi?id=2072585 >> >> Specification:https://webaudio.github.io/web-speech-api/#speechreco-section >> >> Standards Body: >> Audio Community Group, to be adopted in the Audio Working Group next charter >> (currently being draft) >> >> Platform Coverage: >> All desktop platforms, Android tracked in >> https://bugzilla.mozilla.org/show_bug.cgi?id=1244242. >> It's mostly a question of performance, we want to make sure we can use >> hardware >> acceleration on mobile. >> >> Preference:media.webspeech.recognition.enable >> >> DevTools Bug: >> N/A >> >> Extensions Bug: >> N/A >> >> Use >> Counter:https://searchfox.org/firefox-main/search?q=&path=dom%2Fbase%2FUseCounters.conf&case=false®exp=false >> >> Standards-Positions >> Discussion:https://github.com/mozilla/standards-positions/issues/170 >> >> Other Browsers: >> Blink: Shipped >> WebKit: No information >> >> web-platform-tests: >> Complicated. There are some manual tests, but this needs a rather large model >> to run. I'm actively investigating. >> >> >> -- >> You received this message because you are subscribed to the Google Groups >> "[email protected]" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to [email protected]. >> To view this discussion visit >> https://groups.google.com/a/mozilla.org/d/msgid/dev-platform/CANWt0Wp5-8UEHVjycJiCmauvKU2ND4my%2B3dA6nuiT7wxvPsKZg%40mail.gmail.com >> <https://groups.google.com/a/mozilla.org/d/msgid/dev-platform/CANWt0Wp5-8UEHVjycJiCmauvKU2ND4my%2B3dA6nuiT7wxvPsKZg%40mail.gmail.com?utm_medium=email&utm_source=footer> >> . >> > -- You received this message because you are subscribed to the Google Groups "[email protected]" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion visit https://groups.google.com/a/mozilla.org/d/msgid/dev-platform/CANWt0Wp7hJM88p%2BEouqg8Q6rKgZRyN5p%2BoXsYSXTB0Ud3aeHfA%40mail.gmail.com.
