Would it be possible to make the project work with PDFs?

[victorgit4hub]

I'm trying to convert an encyclopedia to docx and epub. However, when I convert to epub the pages are blank, and when I convert to docx it loses the OCR. I believe it's due to the number of images. I tried using PDF Wondershare and PDF24 Tools. Does anyone have any suggestions?

[hydropix]

Hi,

The real issue here is PDF extraction, not the translation itself. Your blank pages in EPUB and the lost OCR in DOCX both come from that step. If your encyclopedia is scanned, each page is an image of text, not actual text, so a normal converter finds no text layer to extract and gives you blank pages. That also explains the lost OCR in DOCX: the conversion dropped the text layer that sat on top of the images.

PDF is a rendering format, not a content format. It describes where to place glyphs on the page, not the logical structure of the text (reading order, paragraphs, headings, captions vs body), so all of that has to be rebuilt with heuristics. With an image-heavy encyclopedia, the result quickly becomes unusable. This is why we do not support PDF directly.

The key is to fix the extraction before translating. For a scanned book, I would suggest pdf-craft (https://github.com/oomol-lab/pdf-craft): it is open source and built specifically to turn scanned PDFs into a clean EPUB, re-running OCR on the images to rebuild real text. Get a clean EPUB out of it first, then run that file through the translator. The translation quality will be exactly as good as your extraction.