AdiVaani: Technology for the margins

India is home to over 700 tribal communities and more than 100 distinct tribal languages and dialects, yet only a handful have a script, a dictionary or any presence online. For a Gondi-speaking farmer trying to read a Government health advisory or a Santhali student searching for a word in a textbook, the digital world has, until now, simply not spoken their language. AdiVaani is changing that through one sentence, one recording, one translated page at a time.

AdiVaani is an AI-based translation and language technology platform being developed under the Ministry of Tribal Affairs (MoTA) by IIT Delhi. Its mission is deceptively simple: build tools that let English and Hindi speak to tribal languages and let tribal languages speak back in text and in voice and vice versa.

AdiVaani has gone live with four languages - Santali, Mundari, Bhili and Gondi — with Kui and Garo close behind, and a further eight languages, including Katkari, Koya, Kokborok, Betta Kuruba, Chodri, Thadou Kuki and Tangkhul, lined up for the next phase.

A full toolkit, not just a translator

AdiVaani brings together a full toolkit under one roof, not merely a translator. It offers Text-to-Text Translation, converting English or Hindi text into a tribal language and back again, instantly; Speech-to-Text, which turns spoken tribal language into text and offers a lifeline for languages with few or no fluent readers; Text-to-Speech, which generates natural spoken tribal-language audio from written text so that non-literate speakers can access information aurally; and Speech-to-Speech, a cascaded pipeline that lets two people speaking different languages understand each other in near real time. The platform also carries an Optical Character Recognition feature that digitises old textbooks, folk manuscripts, primers and dictionaries existing only on paper, often the last surviving record of a dialect, alongside a public mobile App and Web Access portal carrying dictionaries, learning modules and feedback tools for continuous improvement.

Bringing the periphery to the centre

For decades, India’s tribal languages have existed at the edge of the country’s digital and administrative life, rich in oral history, ecological knowledge and cultural memory, yet almost invisible to the internet, to AI systems and often to policy itself. AdiVaani inverts that equation. A welfare scheme notification, a health advisory or an NCERT textbook chapter that once existed only in Hindi or English can now be rendered accurately and respectfully into Santali or Bhili and read aloud to someone who cannot read at all. Government officials, teachers, and health workers can, in turn, understand what a tribal citizen is trying to communicate in their own words. The traffic runs both ways and that is the point: AdiVaani is not simply “giving” tribal communities access to the mainstream; it is giving the mainstream access to tribal India.

The project treats tribal languages not as data sources for AI but as living knowledge systems connected to oral tradition, agriculture, ecology and community identity.

Not just for tribal India — A platform for everyone

AdiVaani’s promise extends well beyond tribal households. Researchers and linguists gain a reproducible, benchmarked dataset for some of the world’s most under-documented languages. Civil servants, doctors, teachers and NGO field workers posted in tribal districts, most of whom are not themselves tribal, get a real-time translation companion that turns awkward, error-prone communication into genuine dialogue. Students of linguistics and anthropology gain digitised manuscripts and oral archives that would otherwise decay unseen in local libraries. And every Indian, tribal or not, inherits a small but meaningful correction to the digital record: languages that were at risk of existing only in memory now have a searchable, speakable, teachable home online. This is what makes AdiVaani more than a translation app, it is public infrastructure for inclusion, built the way inclusion should be built: with the community at the centre, not at the receiving end.

Community first, technology second

AdiVaani’s most distinctive feature may not be its models at all but its method. Native speakers, translators, field linguists and community elders are treated as knowledge partners, not just annotators. Every dataset passes through a maker-checker-superchecker workflow, where community validators catch the subtle ways AI tends to “flatten” a minority language toward Hindi or English: a dropped suffix, an over-normalised idiom, a lost oral register. Machine-generated translations are never treated as final: they return to native speakers for correction before they re-enter the corpus. It is AI development grounded in consent, cultural respect and the recognition that a language is not a dataset it is a way a people remembers itself.

This is where the State Tribal Research Institutes (TRIs) become indispensable. Long before a sentence reaches an AI model, it is the TRIs in Jharkhand, Odisha, Chhattisgarh, Madhya Pradesh and elsewhere that anchor the entire corpus-building effort on the ground - mobilising native speakers, verifying dialectal variations across villages and blocks, and vetting every translated or transcribed sample before it is accepted into the training data. Their field knowledge decides which recordings are authentic and which idioms have been flattened in translation, making them the frontline check between raw community language and a usable digital corpus. Without this continuous, on-the-ground vetting by the TRIs, the models AdiVaani builds would carry the very distortions the project was set up to avoid.

Looking ahead

AdiVaani’s work is far from finished. Over the next nine months, it plans to add ten more languages, improve its speech recognition and voice technology, and build much larger collections of verified recordings reaching hundreds of thousands per language. But beyond the numbers, AdiVaani stands for something bigger: proof that technology can protect and support tribal languages instead of erasing them. In a country where languages are vanishing every few years, AdiVaani gives tribal communities a way to be heard in their own voices, on their own terms and gives the rest of India a reason finally to listen.

The author is Secretary, Union Ministry of Tribal Affairs; Views presented are personal.