Free tool

Tamil image to text, read on your device.

Patta and chitta copies, marksheets, government orders, certificates and book pages in தமிழ், read in your browser and given back as text you can paste, search or translate. English and numbers on the same page are read too.

2. The languages on the page, up to 3
Indian languages are read along with English, for the numbers and English words.

The first time, your browser fetches the reader: about 9 MB, and 1 to 6 MB for each other language. After that, a page takes a second or two, even a tilted phone photo.

Runs in your browser. Your files are never uploaded.

The text appears here, ready to copy or fix.

Several files, or a whole form? Tell Sochen “Read this scanned rent agreement and tell me the notice period and the deposit” and it does every step, checks each against the notification, and still keeps your files on your device.

Try Sochen free →

How Tamil reads

  • Tamil has no stacked conjuncts (only க்ஷ and ஸ்ரீ), so lines are simpler for a reader than Devanagari or Bengali. The catch is the pulli (புள்ளி), the dot above a consonant that removes its vowel: it's tiny, and dropping it turns க் into க.
  • Vowel signs are written before the consonant (ெ, ே, ை) or on both sides of it (ொ, ோ, ௌ). The reader returns them in correct Unicode order, so what you copy is spelt correctly.
  • The Grantha letters ஜ, ஷ, ஸ, ஹ and க்ஷ appear in loanwords and names. They're less common in the training data and are worth a glance in names and addresses.
  • Tamil documents print numbers as 0–9 almost everywhere, so the reader reads English alongside and the numbers come through cleanly. Tamil numerals (௧, ௨, ௩) are rare and are read as such.

Tamil documents people bring

  • Patta, chitta and A-register extracts from eServices Tamil Nadu, to read survey numbers and owners' names.
  • SSLC and HSC marksheets and transfer certificates from the Tamil Nadu board.
  • Government orders (GOs) and circulars from Tamil Nadu departments, to search a clause or translate.
  • Community, income and nativity certificates, and ration cards.
  • Pages of Tamil books, newspapers and notes.

Getting a better read

  • Sharpness matters more than anything: the pulli and the small loops of ள, ழ and ண are what get lost. Scan at 300 DPI, or zoom the phone camera until the text fills the frame.
  • Choose Tamil on its own; English is read with it automatically. Adding a second Indian language doesn't help a Tamil page.
  • An older Tamil PDF typed in a TAM, TAB or Bamini font copies out as gibberish. Choose only Tamil and the reader reads the picture instead, giving real Unicode Tamil.
  • Handwriting isn't read reliably. The reader is trained on printed Tamil type.

Worth switching on alongside Tamil: Nothing; English comes along by itself. The reader takes up to three languages, and English is read with every Indian language automatically.

Scanned PDFs are read page by page, up to 30 pages at a time; a PDF with its own text layer is copied exactly, without reading the picture.

Questions

Does the Tamil text paste into Word, Google Docs or a Tamil Nadu e-service form?+
Yes. The result is Unicode Tamil, the same as typing with a Tamil keyboard, so it pastes into any app that shows Tamil.
Why did a letter lose its dot (க instead of க்)?+
The pulli is the smallest mark in the script and the first thing a soft scan loses. Reread from a sharper image, or fix it in the text box before copying.
Can it read old Tamil books with Grantha-heavy text?+
It reads them, but Grantha letters and archaic ligatures are less common in the training data, so check those words.
Is the file uploaded?+
No. The Tamil model is fetched to your browser once and the reading runs on your device.