DueproofGitHub

English and 22 Indian languages

Every language in the Eighth Schedule. Text is translated into English on your own machine by IndicTrans2, so one set of rules reads them all, and the outline always lands on the original words.

LanguageScriptTyped PDFsScans and photosTested end to end
हिन्दीHindiDevanagariYesYesYes
বাংলাBengaliBengaliYesYesYes
मराठीMarathiDevanagariYesYesNot yet
తెలుగుTeluguTeluguYesYesYes
தமிழ்TamilTamilYesYesYes
ગુજરાતીGujaratiGujaratiYesYesNot yet
اردوUrduArabicYesYesYes
ಕನ್ನಡKannadaKannadaYesYesNot yet
ଓଡ଼ିଆOdiaOdiaYesYesNot yet
മലയാളംMalayalamMalayalamYesYesYes
ਪੰਜਾਬੀPunjabiGurmukhiYesYesNot yet
অসমীয়াAssameseBengaliYesYesNot yet
मैथिलीMaithiliDevanagariYesNo OCR model yetNot yet
ᱥᱟᱱᱛᱟᱲᱤSantaliOl ChikiYesNo OCR model yetNot yet
کٲشُرKashmiriArabicYesNo OCR model yetNot yet
नेपालीNepaliDevanagariYesYesNot yet
سنڌيSindhiArabicYesYesNot yet
कोंकणीKonkaniDevanagariYesNo OCR model yetNot yet
डोगरीDogriDevanagariYesNo OCR model yetNot yet
ꯃꯤꯇꯩꯂꯣꯟManipuriMeetei MayekYesNo OCR model yetNot yet
बड़ोBodoDevanagariYesNo OCR model yetNot yet
संस्कृतम्SanskritDevanagariYesYesNot yet

Tested end to end with the real translation model: Hindi, Bengali, Telugu, Tamil, Urdu, Malayalam. The others use the same path; if you try one, please report what you find.

Two things to know

Languages that share a script

Dueproof tells languages apart by script. Hindi, Marathi, Nepali, Sanskrit, Konkani, Maithili, Dogri and Bodo all use Devanagari, so among the languages you switch on, the first one listed claims the script. DUEPROOF_LANGUAGES=mr,ta reads Devanagari as Marathi.

Without the translation model

The model's weights are a separate, licence-gated download. Without them, Indian-language text is still read, searched and shown, but duties are found only through words your profile lists in that language.

# Switch languages on when you start it
DUEPROOF_LANGUAGES=hi,ta,bn dueproof serve

# or all 22
DUEPROOF_LANGUAGES=all dueproof serve

Languages in the documentation