How to OCR a PDF in Bengali, Hindi, Arabic and more
Updated
Contents
To OCR a PDF in a language other than English, open OCR PDF, select every language printed on the page under Languages, add your scan, and press Recognize text. 34 languages are offered, including Bengali, Hindi, Arabic, and Tamil, and one job can combine up to 4.
Choosing the language is the step that decides the result. English is selected when the page opens, and a page read in the wrong language does not fail: the engine returns fluent-looking nonsense in the alphabet it was given, with no error.
OCR PDF runs on a secure server. Your scan, up to 100 MB, travels over an encrypted connection and is deleted when the job finishes, and the download link stops working after 60 minutes.
Step by step
- 1Open the OCR PDF tool.
- 2Under Languages, select every language the document uses, up to 4. English starts selected; deselect it if the page has none.
- 3Under Output, choose Searchable PDF, Text sidecar for a .txt file, or Searchable PDF + text.
- 4Drop one PDF, JPG or PNG into the Add a scan to OCR area and press Recognize text. The file uploads over an encrypted connection only when you press it.
- 5Download the result. The link stops working after 60 minutes, so save it right away.
- 6Search the PDF or read the text to check it. If it looks like gibberish, change the language selection and run the job again.
The languages on offer
The menu offers 34 languages: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Ukrainian, Polish, Swedish, Turkish, Chinese (Simplified), Japanese, Korean, Vietnamese, Indonesian, Thai, Arabic, Hindi, Bengali, Chinese (Traditional), Greek, Hebrew, Persian, Urdu, Marathi, Nepali, Gujarati, Tamil, Telugu, Malay, Filipino, and Swahili. The job API accepts only languages on this list and refuses any other rather than guessing.
If our server ever lacks the pack for a language you selected, the job stops before recognition starts and the error names that language, rather than blaming your file.
Documents in more than one language
A form printed in Bengali with English headings, or a Hindi letter with an English address block, needs every language on the page selected. Up to 4 can be combined in one job; to swap one out at the limit, deselect a language first.
Select only the languages that are really there: each one you add is more recognition work for the job.
Straight to an editable Word document
PDF to Word can run the OCR in the same job: tick “Recognise text in scans (OCR)”, pick the languages, and our server recognises the scan first and converts the result into a DOCX. Its picker offers 30 of the 34 languages.
Arabic, Hebrew, Persian, and Urdu are not offered for PDF to Word's OCR: written right to left, their recognised words come out of the conversion out of order. OCR PDF reads them in order, so for those languages use OCR PDF and take its searchable PDF or its text.
Common questions
Which languages can the OCR read?
34: English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Ukrainian, Polish, Swedish, Turkish, Chinese (Simplified), Japanese, Korean, Vietnamese, Indonesian, Thai, Arabic, Hindi, Bengali, Chinese (Traditional), Greek, Hebrew, Persian, Urdu, Marathi, Nepali, Gujarati, Tamil, Telugu, Malay, Filipino, and Swahili. Up to 4 can be combined in one job.
Why does my OCR result look like gibberish?
Usually the selected language does not match the page. Bengali read as English, for example, comes back as Latin letters that look plausible and mean nothing, with no error. Select the language the page is printed in and run the job again; a blurry, skewed or sideways scan causes the same symptom.
Can I OCR Arabic, Hebrew, Persian, or Urdu?
Yes, with OCR PDF, which reads them in order in its searchable PDF and its text. Arabic, Hebrew, Persian, and Urdu are not offered for PDF to Word's OCR, because their recognised words come out of that conversion out of order.
Can I turn a scan straight into an editable Word document?
Yes, for 30 of the 34 languages. In PDF to Word, tick “Recognise text in scans (OCR)” and pick the languages; our server recognises the text first and converts the result in the same job.
Is my scan uploaded?
Yes. Recognition needs a native OCR engine, so your file is uploaded over an encrypted connection to a secure server, which deletes the input when the job finishes. The download link stops working after 60 minutes, and a daily storage rule erases the stored result.