Skip to content

CaptiOCR v0.17.4

Choose a tag to compare

View all tags
@carlosacchi carlosacchi released this 24 Apr 20:36
v0.17.4
b29c154

v0.17.4: Post-processing fidelity fixes

  • Cap speaker-label qualifier regex to <=30 chars and reject candidates
    with sentence punctuation; prevents OCR-mangled frames (e.g. missing
    closing paren) from being treated as a giant "full name" and replacing
    every occurrence of the real speaker label with corrupted text.
  • Anchor _find_novel_words to a match block ending near the end of the
    previous text instead of the last match anywhere; preserves middle
    utterances bracketed by repeating speaker labels (e.g. "Pause", "OK",
    "enabled Workspace") that were silently dropped before.
  • Add scripts/reprocess_capture.py utility to re-run the post-processing
    pipeline against any raw capture file.
Assets 5
Loading