amongglue/topaz-shorthand-data

Working copy of the training data for the Topaz shorthand recognition models. Not curated.

• gregg-anniversary/processed/: data behind the 61M/117M Gregg Anniversary models.
  • books/*.pkl: line crops with labels (<book>_v4.pkl raw alignment, _clean filtered, _strict stricter filter; *_before_split are backups).
  • keys/: OCR'd keys (<book>.json) and hand-reviewed corrections (<book>_fixes.tsv).
  • ocr/<book>/: page images and ocr.json per book. src/: source PDFs. gutenberg/: texts used as keys for the literature editions.
• sources/<system>/: raw shorthand material collected per system, with manifest.jsonl (url, title, year, key available, notes).

Scans of printed books from many years; copyright status varies by book.