Recognition language packs
Short answer: Simplified Chinese, English, Thai, Traditional Chinese and Japanese are built in; other scripts need a language pack. Download it under "Scanned text → Recognition language packs…". On a computer without internet, download on a connected computer and use "Import from file". Downloads and imports are both checked against SHA-256.
Download
In “Scanned text → Recognition language packs…”, select the language packs you need and click download. Downloads happen only when you click. Each download is checked against SHA-256, and a pack that fails the check is not installed. Language packs are downloaded from third-party model hosts (ModelScope, GitHub, PyPI), not through our servers.
Language pack sizes: PaddleOCR packs 8–14 MB, Tesseract packs 1–6 MB.
Offline import
- On a computer with internet access, open the same window and download the language packs you need.
- Copy the files to the target computer.
- In “Recognition language packs…”, click “Import from file”.
Imports are checked against SHA-256 in the same way.
Read-back packs (optional)
For Devanagari and Kannada you can install an additional Tesseract pack that acts as a second recognizer during read-back verification. Key fields pass only when both recognizers read them correctly. Ordinary text passes when either one reads it exactly right, which reduces “needs review” results caused by misread vowel marks. Once installed for Kannada, Tesseract handles searching and the original model is used for checking. An Arabic read-back pack also exists, but the current version does not yet support searching or replacing Arabic and other right-to-left scripts, so it is not used for now.
Japanese reading dictionary
The “Japanese reading dictionary” language pack (SudachiDict, about 42 MB to download) is used when replacing words that carry furigana. With the dictionary installed, new readings are generated from it. Without it, the old furigana is left unchanged and you see the notice “Furigana not updated”.
Scripts that cannot be recognized
Syriac and Thaana (Maldives) cannot be recognized in scans. Both are written right to left, and the current version does not yet support searching or replacing them in text-layer PDFs either. For measured results on scans in each language, see the language support matrix.
FAQ
Where are language packs installed?
%LOCALAPPDATA%\PDFTextReplacer\models.
What happens if a language pack is missing when I search?
You are prompted to install it, and the search runs again automatically after installation.