Skip to content

[codex] Improve voice dictation hotwords#380

Merged
ErlichLiu merged 1 commit intomainfrom
codex/voice-dictation-hotwords
May 6, 2026
Merged

[codex] Improve voice dictation hotwords#380
ErlichLiu merged 1 commit intomainfrom
codex/voice-dictation-hotwords

Conversation

@ErlichLiu
Copy link
Copy Markdown
Owner

Summary

This PR improves the Doubao streaming voice dictation experience in two related areas:

  • Adds a local custom hotwords setting for voice input and passes those words to Doubao ASR through request.corpus.context.hotwords.
  • Makes transcript merging resilient when Doubao returns either full transcript updates or current-segment updates after VAD sentence boundaries.
  • Keeps the dictation pause window at 5 seconds and bumps @proma/electron to 0.9.17.

Root Cause

Users could experience natural pauses as recognition boundaries. The previous UI also replaced the whole visible transcript with each ASR event, which could lose earlier confirmed text when the provider returned only the current segment.

Impact

Users can now add product names, technical terms, and names directly in Voice Input settings without configuring a full Volcengine hotword table. Dictation should also better preserve text across pauses.

Validation

  • bun run typecheck
  • bun test
  • git diff --check

Notes

README/AGENTS documentation was not changed yet; project instructions say to update docs after explicit approval.

@ErlichLiu ErlichLiu marked this pull request as ready for review May 6, 2026 03:38
@ErlichLiu ErlichLiu merged commit 39af806 into main May 6, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant