Repository navigation
fix(asr): add dolly/dougie/dory → Dotty corrections (#130) - #132
Merged
Merged
Conversation
Whisper mis-transcribes the wake-name "Dotty" as "Dolly", "Dougie", or "Dory" — none of which were in _ASR_CORRECTIONS, so name-addressed triggers like "goodnight Dotty" never matched. Add the three mishearings (observed in the #39 bench transcripts) to the existing correction table. Verified end-to-end through the asr-correct → phrase-correct → state-detect pipeline: "Good night, Dolly." now resolves to "good night Dotty." and fires the sleep trigger. Low risk in this single-device kid-robot context where those words are overwhelmingly mishearings of the name. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Pull request overview
Adds additional ASR post-processing corrections so Whisper mis-transcriptions of the wake-name “Dotty” (as “Dolly”, “Dougie”, or “Dory”) are normalized back to “Dotty”, allowing existing name-addressed state triggers (e.g., sleep phrases) to match reliably.
Changes:
- Extend
_ASR_CORRECTIONSinreceiveAudioHandle.pywithdolly/dougie/dory → "Dotty"mappings.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Closes #130.
Whisper mis-transcribes the wake-name "Dotty" as Dolly, Dougie, or Dory (observed in the #39 bench transcripts), none of which were in
_ASR_CORRECTIONS— so name-addressed triggers like "goodnight Dotty" never matched and sleep never fired.Change
Adds three entries to the existing correction table in
receiveAudioHandle.py:Verification
Simulated the full
_apply_asr_corrections→_apply_phrase_corrections→_detect_state_phrasepipeline on the issue's real transcripts:Good night, Dolly.good night Dotty.Good one, Dougie.Good one, Dotty.Hey, Dory, wake up.Hey, Dotty, wake up.The comma in "Good night, Dolly." is cleaned up by the existing
("good night", 0.7)fuzzy phrase-correction stage, so thegood night dottysubstring trigger matches.Low risk in this single-device kid-robot context where those words are overwhelmingly mishearings of the name. The optional secondary item from the issue (whether bare "good night" should also trigger sleep) is deliberately not included here.
Note: no regression test added —
receiveAudioHandle.pyhas no existing test harness (it pulls in 7core.*modules only present in the xiaozhi container), so this is a data-only change verified by isolated simulation.🤖 Generated with Claude Code