v150 · Web APIs · Speech

On-device Web Speech API

Chrome 139 brings on-device speech recognition to the Web Speech API on Windows, Mac, and Linux — audio is transcribed entirely locally, never leaving the device, using the same familiar SpeechRecognition interface.

concepts

  1. Feature Detection & Setup

    Checks whether on-device speech recognition is available in this browser, shows the correct configuration to request local processing, and provides a live transcription widget with microphone access.

  2. Privacy Architecture Comparison

    Side-by-side diagram of what data flows to the cloud in classic Web Speech vs what stays entirely on the device with on-device recognition — highlighting the privacy implications for each approach.

  3. Offline Transcript

    A live transcription widget with a simulated "offline" toggle. On-device mode keeps transcribing even when the network is cut; cloud mode fails immediately. A data-flow diagram updates in real time to show which path your audio takes — local model vs. Google servers.

  4. Voice Memo Editor

    A dictation-powered text editor using continuous SpeechRecognition with the on-device path preferred. Interim results appear in grey as you speak; final results commit to the memo. Shows recognition state, language selection, and words-per-session stats.

  5. Confidence Visualizer

    Speak any phrase and see up to 10 alternative transcripts rendered as confidence bar charts. Illustrates how SpeechRecognitionResult[i].confidence varies across alternatives — useful for building spelling-correction UIs or routing ambiguous input to a disambiguation step.

  6. Local Triage Workflow

    A sensitive-note workflow that keeps speech transcription local, classifies urgency in-page, and turns dictated notes into support, clinic, or field-service follow-up fields with privacy status visible.

why it shipped

The Web Speech API has been in Chrome since version 25, but every recognition request was routed through Google's speech recognition servers — sending audio data to the cloud, requiring an internet connection, and creating privacy exposure for sensitive use cases (healthcare notes, legal dictation, personal journalling). Chrome 139 integrates the on-device speech model directly into the browser on desktop platforms. Developers use the exact same SpeechRecognition interface they already know; the on-device path is selected automatically when available and the application requests it. Transcription latency also improves for short utterances because the network round-trip is eliminated.

references