v142 · miscellaneous

Web Speech API contextual biasing

This feature enables websites to support contextual biasing for speech recognition by adding a recognition phrase list to the Web Speech API.

concepts

  1. Context Phrases

    Type some biased phrases, hit the mic — two recognisers run in parallel (with and without the hint) so you can watch how biasing changes the transcript.

  2. Voice Commands as Bias Grammar

    The other obvious use case: voice control. A media-player command set (play, pause, prev, queue, …) fed in as biased phrases — say a command, see the recognised match resolve and trigger the action.

  3. Domain Dictation Lab

    Pick a domain (medical, legal, software, cooking) — the bias phrase list flips automatically. Two parallel recognisers (biased vs baseline) run side-by-side so you hear the difference on technical vocabulary. Tune the boost level, add your own phrases, watch hits highlight live.

  4. Bias Phrase Lab

    Side-by-side recognition panels — one plain, one with a SpeechGrammarList. A phrase tag editor with domain presets (Medical/Legal/Tech) builds a live JSGF grammar string. "Start both" launches two SpeechRecognition instances simultaneously; bias-hit words are highlighted in the transcript and tracked with per-phrase hit counts.

why it shipped

The Web Speech API is a web standard API that allows developers to incorporate speech recognition and synthesis into their web pages. Currently the Web Speech API uses generalized speech recognition models and cannot prioritize certain words or phrases, or adapt to user-specific or domain-specific vocabularies. Adding recognition phrase list provides the Web Speech API with an explicit mechanism to support contextual biasing.

references