Shipped batch-convert-scanned-pdf-to-word-with-ocr — pdf.js raster plus serial OCR into one sectioned Word file, not a JPG-batch accept swap.

Date: 2026-09-08 22:15 Summary: Shipped batch-convert-scanned-pdf-to-word-with-ocr — pdf.js raster plus serial OCR into one sectioned Word file, not a JPG-batch accept swap. Visibility: people

[question] 实现

Tool links (English)

batch-convert-scanned-pdf-to-word-with-ocr

[try to solve]

Context

N4 was briefed as a scanned-PDF queue that must rasterize pages, OCR them, and download one Word file. Neighbors already cover digital text-layer PDF→Word (A2) and photo OCR→Word (N4-jpg-batch). Extra Information Gain had to be PDF-only intake, page skip, and a headed DOCX — not changing the photo tool’s accept.

Process

Catalog shard (page.style: opts, pdf / pdf-docs), ASCII icon, and batchConvertScannedPdfToWordWithOcrPage.ts.
Client pipeline: PDF dropzone (reject images), peek page counts with pdf.js, caps 5 files / 20 pages / 12 MB, Convert all loads Tesseract once, serial page render → OCR → one docx Blob. Gold HUD (Load / Render / Read / Pack) with CSS spin so a frozen main thread still shows motion. No auto-loadSample (Tier 2 WASM). Load sample builds a two-page Helvetica PDF via pdf-lib and runs the same Convert all path.
English master copy, then independent ten-locale shards from 03-locale-briefs.md. How labels match Convert all / Download Word (and locale button strings).
Neighbor FAQ on A2, N4-jpg-batch, and extract-text-from-pdf now name this H1 (CROSS_TOOL_UPDATE=1). README and ocr-scan planner marked implemented.
coverage:gate --phase=all, lint:tool-page --require-html, lint:seo, lint:vendor, isolation, and verify:tool all green.

Root cause / analysis

A scanned PDF has no reliable selectable text layer. Copying A2 would fail; copying N4-jpg-batch would be a doorway. The product difference is page rasterization plus a file/page Word outline.

Solution

Ship this slug only. One PDF still uses Convert all / Download Word. Failed pages skip. Default includes the page picture above OCR text. Digital text-layer files and JPG stacks stay on neighbor URLs.

Notes / boundaries

Inbound related-graph warn: this slug is not yet listed in other tools’ related arrays (FAQ names it). Optional follow-up.
Live click-through of Convert all was not run in a browser this session; prerendered HTML was checked for buttons, accept=application/pdf, HUD, loadSample, and /{(\w+)}/ templates.
Caps: 5 PDFs, 20 pages, 12 MB. Handwriting and table layout are not promised.

[actions]

Added src/site/tool-catalog.d/batch-convert-scanned-pdf-to-word-with-ocr.json, src/pages/batchConvertScannedPdfToWordWithOcrPage.ts, src/site/i18n/tools/batch-convert-scanned-pdf-to-word-with-ocr/*, public/icons/tools/batch-convert-scanned-pdf-to-word-with-ocr.svg
Neighbor FAQ: batch-convert-jpg-to-editable-word-with-ocr, turn-pdf-into-word-document, extract-text-from-pdf
README + ocr-scan planner/README
CROSS_TOOL_UPDATE=1 npm run verify:tool -- --slug=batch-convert-scanned-pdf-to-word-with-ocr