A Private Document Scanner That Reads Bangla — Offline

Scanning a paper with your phone is routine now. But a scan is just a picture until the app can read the words in it. Once it can, the scan becomes useful: type a dag number, a patient ID or a policy number and the right document turns up.

For English documents that has worked for years. For Bangla it mostly hasn't. Many apps read nothing from a Bangla page, or produce a stream of meaningless Latin letters. And several of the apps that do read Bangla well send the photo to a company's server to do it. For an NID card or a land deed, that is worth a second thought.

Why choosing a scanner for Bangla papers is different

• Bangla is hard to read by machine. Conjunct letters, vowel signs above, below and beside the consonant, the reph: the script is far harder to recognise than English. Even an app that says it "supports Bangla" is worth testing on one of your own papers.
• Numbers come in two scripts. Bangladeshi papers use both ১২৩ and 123. If a dag number or a receipt number is printed in Bangla digits, it matters whether you can still find it by typing English digits.
• These are the most personal papers you own. NID, passport, land records, medical reports. "Processed in the cloud" means the image leaves your phone, and you should know where it goes.

What reading Bangla on the phone changes

• Search by what the paper says. Type "creatinine" and the lab report comes up; type a mouza name and the khatian comes up. You don't need to remember when you took the photo.
• It works without the internet. Reading and searching both happen on the phone, so they work with no connection, and nothing is uploaded.
• You can copy the text. A long reference or policy number can be copied straight into a form instead of retyped from the paper.

Five tips for better results

1. Good light, flat paper

Scan in daylight or under a bright lamp, with the paper as flat as you can get it. Folds and shadows hide Bangla vowel signs and conjuncts, which is where misreadings start.

2. Switch on only the languages you need

Keep on just the languages your papers are in, for example Bangla and English. After switching a new language on, use Read text again on documents you saved earlier, one at a time or the whole vault at once. Titles, categories and expiry dates stay as they were.

3. Pick page quality to suit the paper

Small print, such as a deed or a lab report, is best kept at High or Standard. Compact saves space, around 0.2 MB a page, but faint print can become harder to read.

4. Check what was read

Each document has a Recognised text section; tap Show text to see what Docuary read. Give it a quick look to make sure the words you will search for later came through correctly.

5. Don't rely on handwriting

Printed text reads best. Handwritten prescriptions or the handwritten parts of a deed can't be read reliably. For those, add a word or two you'll search for to the document's title yourself.

What to expect, and what not to

• Most printed words, not all. In one of our tests, 69 of the 76 Bangla words on a printed leaflet were read correctly. That is plenty for search, but for anything legal, check the original paper.
• Dates in Bangla digits are found; dates written as words aren't yet. "১৫/০৮/২০২৬" is recognised, but a date like "১৫ আগস্ট" won't fill in an expiry date by itself, so enter that one by hand.
• A scan is a copy, not a replacement. Government offices will still want the original or a certified copy. The vault is for keeping papers safe and finding them fast.

The short version

Two questions decide whether a scanner suits Bangla papers: does it actually read Bangla, and does the reading happen on your phone? When both answers are yes, your NID, deeds and medical reports can be found by the words inside them, and no photo ever leaves your phone.

Related reading

Frequently asked questions