heb_rivok and heb_rivok_lite, Hebrew models for Tesseract OCR
Copyright 2026 Rivok Labs

heb_rivok is a fine-tuned derivative of the tessdata_best "script/Hebrew" model, and
heb_rivok_lite of the tessdata_best "heb" model (https://github.com/tesseract-ocr/tessdata_best),
Copyright Google, licensed under the Apache License, Version 2.0.

Their training text includes the Hebrew and English training text and word lists of
tesseract-ocr/langdata_lstm (https://github.com/tesseract-ocr/langdata_lstm), licensed under
the Apache License, Version 2.0.

Training images were rendered with fonts under the SIL Open Font License 1.1 (Google Fonts
Hebrew families, SIL Ezra, Liberation), the GNU FreeFont licence, the DejaVu fonts licence and
GPL-2.0 with font exception (Culmus). No font is distributed with these models.

Version 2.0 was also trained on:
- printed lines from public-domain books scanned on Wikimedia Commons, labelled with their
  validated Hebrew Wikisource transcriptions (CC BY-SA 4.0, Hebrew Wikisource contributors);
- Miqra according to the Masorah (Hebrew Wikisource, CC BY-SA 4.0,
  https://en.wikisource.org/wiki/User:Dovi/Miqra_according_to_the_Masorah);
- the Mishnah (Torat Emet, public domain) and Project Ben-Yehuda texts (public domain);
- modern Hebrew text vowelled with dicta-il/dictabert-large-char-menaked (CC BY 4.0, Dicta).

The example pages in examples/ use public-domain scans from Wikimedia Commons (National Library
of Israel) with reference text from Hebrew Wikisource (CC BY-SA 4.0). Examples are generated
locally by the scripts and not distributed in this repository.
