Skip to content

Avoid aborting ASR word merge on out-of-span timestamps - #252

Merged
0xShug0 merged 1 commit into
mainfrom
fix/249-qwen3-asr-word-merge-drop-outside
Aug 15, 2026
Merged

Avoid aborting ASR word merge on out-of-span timestamps#252
0xShug0 merged 1 commit into
mainfrom
fix/249-qwen3-asr-word-merge-drop-outside

Conversation

@0xShug0

@0xShug0 0xShug0 commented Aug 15, 2026

Copy link
Copy Markdown
Owner

Summary

  • drop fully out-of-source word timestamps during chunk word merge instead of aborting the whole request
  • log a warning with the dropped word and source/keep/span details when logging is enabled
  • keep structural validation errors, including reversed word spans and invalid source/keep spans
  • update the audio chunking unit test for the new non-fatal behavior

Fixes #249

Validation

  • cmake --build build/debug --target audiocpp_cli -j$(nproc)
  • cmake --build build/debug --target audio_chunking_test -j$(nproc) && build/debug/bin/audio_chunking_test
  • reproduced the issue with build/logs/issue_249_recreate_english/tuned/full_audio/compressed_56_as_chunk10_150s.wav; after the fix the same command exits 0 and logs [warning][audio.chunking] dropping word timestamp outside chunk span ...

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

1 participant