Typsastra Khmer OCR & Document Reconstruction

The Khmer Document Reconstruction Project Public beta
⌂ Runs in your browser✓ Documents stay local unless you explicitly donate one displayed sample◫ Searchable, extractable PDF output
Loading OCR models…Change
OCR engineRecognition uses the compact INT8 model on CPU through SIMD WASM.

Checking WASM thread support…

OCR settings
Recognition crop paddingTop and bottom are percentages of line height. Left and right are source-image pixels.

Recover Khmer documents for long-term use

Reconstruct PDFs with preserved page images and a reviewed Unicode text layer for search, selection, extraction, and preservation-oriented workflows. Processing stays on this device. With appropriate rights and provenance controls, reconstructed text can also support RAG, OCR evaluation, and preparation of Khmer research or language-model corpora.

Public beta · OCR can contain recognition, spelling, and reading-order errors. Review important text before publication or archival use.
⇧
Drop a PDF or images herePDF, PNG, JPEG, or WebP
ⓘ Privacy, performance, and PDF output
Private processing

Your document and OCR results stay in this browser. Document content leaves only when you explicitly preview and donate one sample; typed feedback and content-free aggregate events use separate requests.

Current performance

First use downloads about 14 MB for INT8 OCR. OCR Settings can use 2–8 requested CPU threads where cross-origin isolation is available.

Searchable output

Exports preserve page appearance as an image with reviewed Unicode underneath. They support search, selection, copy, and extraction, but are not fully editable digital PDFs. Firefox's built-in PDF viewer may show broken search results; use a Chromium-based browser or compatible desktop PDF reader.

Read Privacy & Data Use
Detector: PP-OCRv6 tiny ↗Recognizer: colored-script AOU-CTC · INT8 WASM ↗Segmenter: Khmer Viterbi WASM ↗Dictionary: Seanghay Hay / RAC · noncommercial ↗Developed by Sovichea · GitHub ↗
0 / 0
Select a page to preview it.

Donate one reviewed sample

Review the exact sample below. Only this displayed crop and its displayed text and diagnostics will be sent—not the document, filename, page, or any other lines.

Generating the exact OCR crop…

Send feedback

Tell us what worked, what was confusing, or what should improve. Only the message you type is sent—never your document, OCR text, filename, or page data.

Use recognized text

Copy the reviewed document text to your clipboard or save it as a plain-text file.

Export searchable PDF

Review what the reconstructed PDF contains before choosing where to save it.

Preserves each page image and adds reviewed Unicode for search, selection, copy, and extraction. The result is not a fully editable digital PDF.

Firefox’s built-in PDF viewer may show broken search results. Use a Chromium-based browser or a compatible desktop PDF reader.