[3rdparty]: Wrong results in German despite excellent input quality

[DReffects]

Simple sanity checks

• This is an issue with an app that uses OCRmyPDF for OCR
• I am using a recent version of the third party app
• I will include a file that reproduces the issuse

Third party app name and version

Paperless-ngx v2.19.5

Describe the bug

Hello there,

I am using OCRmyPDF from within paperless-ngx and am having serious issues with German documents despite excellent input quality.

Config is set to German+English language ("deu+eng").
When using only "deu" the issue with this specific document does not occur. Seems like the language detection fails in this case.

Here's the affected page from the PDF.

OCR-Problem.pdf

"Hängeschränke" and "Unterschränke" is both representing ä as a resulting in "Hangeschranke" and "Unterschranke".
It works on other parts of the document, so it's not broken in general. For excample "Geschirrspüler" is being recognized with the ü intact.

Steps to reproduce

1. Import attached file into Paperless-ngx
2. Trigger OCR
3. Check plain content / ocr result

Files

OCR-Problem.pdf

OCRmyPDF version

16.11.0

Relevant log output

There's nothing obvious besides tesseract complaining about diacritics but when does it not...



[2025-11-08 12:10:37,797] [INFO] [celery.worker.strategy] Task documents.tasks.consume_file[84b68a89-4453-45c6-b090-33d6c15a5aae] received
[2025-11-08 12:10:37,838] [INFO] [paperless.tasks] ConsumerPreflightPlugin completed with no message
[2025-11-08 12:10:37,842] [INFO] [paperless.tasks] WorkflowTriggerPlugin completed with: 
[2025-11-08 12:10:37,842] [INFO] [paperless.consumer] Consuming OCR-Problem.pdf
[2025-11-08 12:10:37,869] [INFO] [paperless.parsing.tesseract] pdftotext exited 0
[2025-11-08 12:10:39,615] [INFO] [ocrmypdf._pipeline] page is facing ⇧, confidence 13.99 - rotation appears correct
[2025-11-08 12:10:42,740] [WARNING] [ocrmypdf._exec.tesseract] [tesseract] lots of diacritics - possibly poor OCR
[2025-11-08 12:10:42,797] [INFO] [ocrmypdf._pipelines.ocr] Postprocessing...
[2025-11-08 12:10:42,826] [INFO] [ocrmypdf._pipeline] Image optimization ratio: 1.00 savings: 0.0%
[2025-11-08 12:10:42,826] [INFO] [ocrmypdf._pipeline] Total file size ratio: 0.91 savings: -9.7%
[2025-11-08 12:10:43,793] [INFO] [paperless.parsing] convert exited 0

[jbarlow83]

Using latest ocrmypdf directly, Tesseract 5.5.1 and latest language packs, the accents were correctly recognized. Perhaps paperless is using an older Tesseract or some other processing step?