design-reviews#1267: WG Revision: Web Speech API: SpeechRecognitionResult Timestamps

#1267: WG Revision: Web Speech API: SpeechRecognitionResult Timestamps

Visit on Github

Opened

Aug 22, 2026

Specification

https://github.com/WebAudio/web-speech-api/pull/192

Explainer

https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md

Links

The WG's request for this TAG review: https://github.com/WebAudio/web-speech-api/issues/191
TAG review of the previous version of this specification, if any:
- https://github.com/w3ctag/design-reviews/issues/1093 (MediaStreamTrack input)
- https://github.com/w3ctag/design-reviews/issues/1146 (On-device speech recognition options)
A description of what has changed since our previous review: https://github.com/WebAudio/web-speech-api/pull/192

Feature 1:

Previous early design review, if any: N/A
An introduction to the feature, aimed at unfamiliar audiences: https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md
A description of the problems that end-users were facing before this proposal: https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md#problem
Alternatives considered: https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md#alternatives-considered
Examples of how to use the proposal to solve the end-users' problems: https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md#proposed-behavior--example-usage
What do the end-users experience with this proposal: End users experience synchronized, frame-accurate captions and translated subtitles during media playback and live conferencing, interactive "click-to-seek" transcript navigation, and seamless alignment between speech transcripts and recorded media tracks.
User research you did to validate the problem and/or design, if any: Live conferencing / media synchronization studies and interactive validation demos:
- Live Meeting Caption Sync & Recording Demo: https://alan33d.github.io/web-speech-demos/live_meeting_recording_demo.html
- WebRTC Caption Playout Delay Test: https://alan33d.github.io/web-speech-demos/webrtc_caption_sync_test.html
Web Platform Tests: https://github.com/web-platform-tests/wpt/tree/master/speech-api (specifically idlharness.https.window.js and SpeechRecognition-onresult-manual.https.html)

The specification

Follows the Web Platform Design Principles.
Includes Security and Privacy Considerations sections based on answers to the Security/Privacy Questionnaire.

Where and by whom is the work is being done?

GitHub repo: https://github.com/WebAudio/web-speech-api
Primary contacts:
- Alan Ding (@alan33d), Google, Spec Editor / Implementer
- Evan Liu (@evanbliu), Google, Spec Editor
- Paul Adenot (@padenot), Mozilla, WebAudio WG Member / Spec Editor
- Hongchan Choi (@hoch), Google, WebAudio WG Member / Spec Editor
Organization/project driving the specification: Google / Chromium
This work is being funded by: Google
Primary standards group developing this feature: Audio Working Group (WebAudio WG) / Audio Community Group (AudioCG)
Incubation and standards groups that have discussed the design:
- Audio Working Group: https://github.com/WebAudio/web-speech-api/issues/191, https://github.com/WebAudio/web-speech-api/pull/192

Feedback so far

Active horizontal reviews: https://github.com/WebAudio/web-speech-api/blob/main/explainers/speech-recognition-result-timestamps.md#security-and-privacy-considerations
Multi-stakeholder feedback:
- Chromium comments: Positive / In active development (https://chromestatus.com/feature/5811907077472256)
- Mozilla comments: Active discussion in https://github.com/WebAudio/web-speech-api/issues/191 and https://github.com/WebAudio/web-speech-api/pull/192#discussion_r3774155900 by @padenot (provided guidance on using double in seconds, non-nullable attributes, and naming speechStartTime/speechEndTime to mirror speechstart/speechend).
- WebKit comments: Standards position to be filed at https://github.com/WebKit/standards-positions/
Major unresolved issues with or opposition to this specification: None. Consensus reached on using non-nullable double representing seconds relative to audio stream start, named speechStartTime and speechEndTime.
Status/issue trackers for implementations:
- Chrome Status: https://chromestatus.com/feature/5811907077472256
- Chromium Issue Tracker: https://issues.chromium.org/issues/528037568

You should also know that...

W3C Audio WG/CG Meeting Discussion: This feature was presented and discussed during the W3C Audio WG/CG Teleconference on August 13, 2026 (see Meeting Minutes). The group aligned on developer needs for media timeline association (e.g. generating VTTCue captions, click-to-seek transcripts, and synchronizing captions with recorded video).

Mozilla's feedback in PRs was also discussed and addressed during the meeting.
Interactive demonstrations showcasing subtitle synchronization, recording timeline alignment, and the inadequacy of existing coarse DOM event timestamps were presented at the WG meeting and are available for testing:
- https://alan33d.github.io/web-speech-demos/live_meeting_recording_demo.html
- https://alan33d.github.io/web-speech-demos/webrtc_caption_sync_test.html

Track conversations at https://tag-github-bot.w3.org/gh/w3ctag/design-reviews/1267

Discussions

Comment by @alan33d Aug 24, 2026 (See Github)

There is some additional feedback we've received which requires updates to the Explainer. We will circle back here once those updates are made.

Comment by @alan33d Sep 10, 2026 (See Github)

The Explainer was updated, and TAG review can resume.

Comment by @lolaodelola Sep 23, 2026 (See Github)

Hi @alan33d,

Thank you for submitting this design review.

The only note I would make is that the Web IDL uses double for both the speech start and end times, we recommend DOMHighResTimeStamp as it allows for "comparison of timestamps, regardless of the user’s time settings" (Web Platform Design Principle 8.4), if there's a valid reason that the double is being used then we defer to the authors' judgment.

This issue is marked with "Resolution: decline" to reflect that we didn't do a thorough review as the architectural impact is minor. If you feel that there is a bigger architectural impact than what we've mentioned, please reopen the issue.

Comment by @alan33d Sep 23, 2026 (See Github)

Thanks for the feedback!

CC @padenot (Audio WG) since using double was their suggestion.

Using double aligns this API with attributes like BaseAudioContext.currentTime and HTMLMediaElement.currentTime which uses double and is measured in seconds. These are the timestamps we expect developers to compare these values with and avoid unit conversions when piping audio streams between Web Audio and VTTCue (https://github.com/WebAudio/web-speech-api/issues/191).