Intent to Prototype: SpeechRecognitionResult Timestamps

11 views
Skip to first unread message

Alan Ding

unread,
12:30 PM (6 hours ago) 12:30 PM
to blink-dev, m...@chromium.org, ev...@chromium.org, rei...@chromium.org, etien...@google.com
Contact emails
alan...@chromium.org, ev...@chromium.org

Explainer
github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md
Specification
github.com/WebAudio/web-speech-api/pull/192
Summary
The Web Speech API currently does not allow developers to associate the transcribed text with specific segments of the audio source, making it difficult to map the generated text to the timeline or determine on-device transcription latency. We propose adding audioStartTime and audioEndTime (DOMHighResTimeStamp?) attributes on the SpeechRecognitionResult interface.

Blink component
Blink>Speech

Web Feature ID
speech-recognition

Motivation

Web applications are increasingly adopting on-device Automatic Speech Recognition (ASR) via the Web Speech API (SpeechRecognition) to enhance user privacy and lower backend processing costs. However, the WebSpeech API operates as a "black box" regarding local processing delays:


  1. Lack of Latency Tracking & Backend Recovery: When client devices with constrained CPU/GPU resources struggle to process audio in real time, significant transcription lag accumulates between when audio was spoken and when recognition results are emitted. Web applications have no web-exposed timing metrics to programmatically detect this lag or trigger an optional failover to a high-quality cloud ASR backend provider, resulting in degraded user experiences (e.g., stale or delayed captions in live video conferencing tools).

  2. Timeline Association: Developers cannot readily map recognized transcript text to specific segments of the source audio timeline, creating challenges for automated subtitle cue alignment and media editing tools.


Initial public proposal
https://github.com/WebAudio/web-speech-api/issues/191
Requires code in //chrome?
True (initial integration with Speech Recognition service and SODA backend; later with ODML/LLM backend).

Tracking bug
crbug.com/528037568

Estimated milestones
Target DevTrial: M153

Link to entry on the Chrome Platform Status
https://chromestatus.com/feature/5811907077472256

This intent message was generated by Chrome Platform Status.
Reply all
Reply to author
Forward
0 new messages