How to convert a scanned Burmese PDF to Unicode text

A scanned PDF is a stack of photographs. Each page looks like text, but to a computer it’s an image: you can’t search it, copy from it, or feed it to a spell-checker. To get Burmese text out, something has to read the pictures. That’s optical character recognition, OCR.

This guide walks through the whole job with LEXO, a free desktop app built for Burmese: from a raw scan to Myanmar Unicode text you can edit and publish.

Before you start

You need:

• LEXO, installed with uv tool install lexo (or pipx install lexo). See installation.
• A Google account. LEXO uses Google Docs OCR, which is free and reads printed Burmese well.
• The one-time Google Cloud setup, which gives LEXO permission to use Google Docs OCR through your account. It’s six steps; follow the setup guide once and you’re done.

Then sign in:

lexo login

Step 1: Check whether you really have a scan

Some PDFs that look scanned actually carry a text layer. If yours does, you don’t need OCR at all, and extraction is instant and exact. Try it first:

lexo extract book.pdf -o book.txt

If LEXO reports that the file “looks scanned”, or the output is empty, carry on. If you get English letters where the page shows Burmese, the PDF uses a legacy font: see Win Innwa PDF to Unicode.

Step 2: Open the PDF in the desktop app

lexo gui

Click Import PDF or Image and pick your file. LEXO makes a working copy, so nothing you do changes the original until you save.

Step 3: Split two-page spreads

If each scanned image holds two book pages side by side, split them before OCR. Otherwise lines from the left and right pages can end up interleaved.

Open the PDF tab in the Tune panel, click Show split line, drag the line onto the gutter, and click Split two-up at line.

Step 4: Crop running heads and page numbers

Book titles at the top of every page and page numbers at the bottom will appear in your text unless you remove them. Switch to the Edit tab: a crop box appears on the page. Drag its edges to the text block, set Apply to to All pages, and click Apply crop.

Step 5: Run Google Docs OCR

Go back to the Run tab, choose Google Docs OCR, and click Run. The language hint is Burmese by default.

Each page is rendered to an image, uploaded to your own Google Drive, read by Google Docs OCR, and deleted again. Pages appear in the text pane as they finish. A long book takes a while; the time depends on your connection and on Google.

If a page fails because of a network hiccup, LEXO retries it automatically. Anything still failing at the end can be re-run on its own with Retry Failed Pages.

Step 6: Proofread side by side

OCR is good, not perfect. Go through the pages with the image beside the text. Zoom into the page with Ctrl and the scroll wheel when a line is hard to read. Typical things to check:

• stacked consonants and medials in small or worn type;
• punctuation (၊ and ။) next to numbers;
• lines broken across a page boundary.

Step 7: Export

Pick a format under Export as and click Export:

• Plain text for reading and editing;
• Markdown if you want a heading per page, which is useful for citations and EPUB work;
• JSONL for datasets, with one record per page.

The text is normalized to Unicode NFC, and Burmese zero-width spaces are preserved. More on formats.

Doing it from the command line

Everything above has a CLI equivalent. Prepare the PDF, then OCR it:

lexo pdf split-spread scan.pdf -o pages.pdf
lexo pdf crop pages.pdf --top 7 --bottom 9 -o cropped.pdf
lexo ocr cropped.pdf --lang my -o book.txt

For a folder of PDFs, use batch OCR:

lexo ocr-batch ./scans --out-dir ./txt

Watch the whole workflow

A note on privacy

Pages you OCR are uploaded to your Google Drive for recognition, then deleted. Everything else (opening, cropping, splitting, proofreading, exporting) happens on your computer. Only OCR documents you’re allowed to send to Google. Details are in Privacy and authentication.