LEXO
မြန်မာ Download

LEXO 03 Use cases

v0.4.1 AGPL-3.0

What people digitize with LEXO

Five common jobs, the workflow that fits each one, and where LEXO's limits are.

Researchers and archives

You have scanned books, periodicals, or manuscripts in Burmese and need text you can search, quote, and cite. The scans are often two-up spreads with running heads and page numbers.

How it goes

  1. Open the scan and split two-up spreads in the PDF tab.
  2. Crop away running heads and page numbers for all pages at once.
  3. Run Google Docs OCR, then proofread each page against its image.
  4. Export Markdown: one heading per page keeps citations traceable to the original page.

Keep in mind

OCR output needs proofreading before you quote it. LEXO doesn't detect columns, tables, or footnotes, so complex layouts may need manual cleanup.

Publishers and EPUB production

You're bringing an out-of-print Burmese title back as an ebook, and the only source is a printed copy or an old PDF.

How it goes

  1. Scan or open the book, split spreads, and crop to the text block.
  2. OCR it, then correct text in the side-by-side view, where the page image is always next to the line you're fixing.
  3. Export Markdown and feed it to your EPUB toolchain.

Keep in mind

LEXO produces text, not typeset ebooks: headings, emphasis, and images are yours to restore. Only digitize work you have the rights to.

Burmese NLP and LLM datasets

You need large amounts of clean Myanmar Unicode text for training or evaluation, and much of what exists is in scanned PDFs.

How it goes

  1. Batch-OCR folders of PDFs from the CLI, which resumes if interrupted.
  2. Or export JSONL per document: one record per page with the source, page number, and how the text was obtained.
  3. Filter out failed pages and deduplicate by the document's SHA-256 doc_id.
lexo ocr-batch ./pdfs --out-dir ./txt
lexo ocr book.pdf --format jsonl -o book.jsonl

Keep in mind

Text is NFC-normalized with zero-width spaces kept, but OCR errors remain. Sample and review before training on it, and respect the licences of the source documents.

Legacy-font documents

Office files and PDFs from the Win Innwa era look Burmese on screen, but copy-paste gives you Latin letters. Search, sorting, and screen readers all fail on them.

How it goes

  1. Open the PDF. If text-layer extraction shows ASCII where Burmese should be, it's a legacy font.
  2. Switch to Google Docs OCR, which reads the rendered page instead of the hidden ASCII.
  3. Export Myanmar Unicode text.

Keep in mind

OCR re-reads the page, so check the output as you would any OCR. A Win-to-Unicode font converter on the raw text is the alternative when you can't upload the document.

Students and teachers

Course readers, past exam papers, and handouts circulate as photocopies and phone photos. You want text you can search, annotate, or rework.

How it goes

  1. Open phone photos as an image set: one document, one page per photo.
  2. Rotate and crop each page so only the text is left.
  3. OCR, fix the few mistakes, and export plain text.

Keep in mind

Clear, flat, well-lit photos give the best results. Handwriting is outside what LEXO is built and tested for.