Repository navigation
feat(voice): on-device voice input (Chrome local recognition, never the cloud) - #100
Merged
Merged
Conversation
…tion 'device' is accepted wherever 'browser' is: validation, drafts, hosted rules and follow-up copies. /api/transcribe still accepts only 'installation'. Changing the mode advances the study revision (shared contract scenario). Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
… the cloud
Device mode gates the mic on SpeechRecognition.available({processLocally}),
installs the language pack on first press while typing stays available, and
starts only recognizers that have the processLocally attribute and accept
true. language-not-supported and service-not-allowed hide the mic for the
session. There is no fallback to cloud dictation or server transcription.
Consent presents locality as the browser's promise: the browser turns speech
into text on this computer and says the recording stays there; the study
receives only the text the participant sends. zh-CN verified as accepted by
Chrome 155 (available() for all six tags).
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
…ut; Spanish tú register The device availability effect re-ran for every mode on a language change, resetting state and stopping an active browser dictation. Device detection is now its own effect that returns early outside device mode (regression test). The Spanish device lines used usted; the file uses tú. Both from the Kimi review. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Gemini, Kimi and Grok reviewed the five device-voice translations as the 5.1 strings were. No promise was strengthened or weakened; fixes are term and register consistency in ja, zh and ko. Decisions recorded in docs/translations/REVIEW-2026-10.md. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
menotbobbybrown
pushed a commit
to menotbobbybrown/openinterviewer
that referenced
this pull request
Oct 11, 2026
On-device voice input (linxule#100), analysis files in the study export (linxule#101), OpenNext 1.20.10 / wrangler 4.148.0 with Next held at 16.3 (linxule#98), provider SDKs (linxule#95). Code-only release; rollback to 5.3 refuses device-mode studies. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Adds a fourth per-study voice mode,
device: desktop Chrome turns speech into text on the participant's computer (SpeechRecognition.processLocally). Decision record: the on-device voice spike (GO for Chrome local recognition; NO-GO for self-hosted Whisper and Moonshine).Privacy invariant. A device recognizer starts only if
processLocallyexists on it and reads backtrueafter being set. There is no fallback to cloud dictation or server transcription.language-not-supported,service-not-allowed, a failed install, or a missing API hide the mic for the session; participants type instead./api/transcribestill accepts onlyinstallation.Consent. Locality is worded as the browser's claim, not ours: the browser "turns your speech into text on this computer and says the recording stays there; this study receives only the text you choose to send." This is in all six languages. The five translations are drafts pending review.
Verified in real Chrome 155 (macOS):
available()returns downloadable for all six tags (en-US, zh-CN, fr-FR, ja-JP, ko-KR, es-ES;zh-CNconfirmed), andinstall()succeeds for en-US (~5–30 s). Not verified: recognition of real speech. A fake-microphone clip producedno-speech, so a manual check by speaking in desktop Chrome is required before participants use it.Rollback. The mode is a study-config field bound to the revision. No schema change, but a 5.3 build refuses studies saved with
device(save 400, links 409, canonical load 503). Before rolling back, set those studies to Off on 5.4 and reissue links.Reviews: Kimi approved with no blockers. Its two minor findings were fixed after checking them against the code: a language change reset installation/browser voice state, and the Spanish register was inconsistent.
npm run check:cloudflare: 20/20 lanes on a clean tree at the head commit.Inspired by Yinghua's voice-input work in her fork, credited in ACKNOWLEDGMENTS as in 5.1.0.
🤖 Generated with Claude Code