BHASHINI, IISc release AI voice models for minority Sourashtra language

The Digital India BHASHINI Division and the Indian Institute of Science (IISc), Bengaluru have released text-to-speech (TTS) models for Sourashtra, a minority language spoken mainly by the Sourashtra community in Tamil Nadu. The release was announced on 16 October 2026. The models can read Sourashtra text aloud in male and female voices, from text written in either the Tamil script or the Sourashtra script, and the speech data used to build them, more than 28 hours of recordings, has been made available on the government's AIKosh platform.

Most AI voice technology is built for languages with hundreds of millions of speakers and huge amounts of recorded and written data. Smaller languages are often left out, which means their speakers cannot use voice assistants, screen readers or audio content in their own language. Over time, that can push younger speakers towards dominant languages and weaken the smaller ones.

A text-to-speech model is a step towards closing that gap. With it, any text written in Sourashtra, such as a story, a public notice or a lesson, can be turned into audio automatically. That helps people who cannot read the script, people with visual impairments who rely on screen readers, and children learning the language. Supporting both Tamil script and the language's own script matters too: many speakers write Sourashtra in Tamil letters, while the traditional script carries cultural value.

Publishing the speech dataset on AIKosh may matter as much as the models themselves. Training data is usually the hardest part of building AI for a small language. By making 28-plus hours of balanced male and female recordings openly available, the developers let researchers, start-ups and students build further tools, such as speech recognition or translation, without starting from zero. This fits the aim of BHASHINI, India's national language technology mission, to make digital services available in Indian languages.

For India's technology ecosystem, the release is also a template. The steps are clear: record a balanced set of native speakers, train a model on that data, support the scripts the community actually uses, and publish the data so others can build on it. Universities, language institutes and community groups working on other small languages can follow the same path, and start-ups can use such open datasets to offer services in markets that larger companies often ignore.

Short-term: Developers and community organisations can now use the models and the dataset to create audio content, reading aids and educational material in Sourashtra.

Long-term: If the approach is repeated for other minority languages, it could help preserve languages that have few digital resources, and make government and private digital services accessible to more Indians in the language they speak at home.

Who is affected: Sourashtra speakers in Tamil Nadu and elsewhere, teachers and cultural organisations working to preserve the language, people with visual impairments, and AI researchers and start-ups building Indian-language tools.

Questions and Answers