Sourashtra text-to-speech is now a reality for a language with very few digital tools. On October 16, 2026, the Press Information Bureau announced that the Digital India BHASHINI Division, under the Ministry of Electronics and Information Technology (MeitY), working with the Indian Institute of Science (IISc) in Bengaluru, has developed text-to-speech models for Sourashtra, a minority Indo-Aryan language spoken predominantly by the Sourashtra community in Tamil Nadu.
What the Sourashtra text-to-speech models offer
According to the government announcement, the models use the VITS architecture and generate Sourashtra speech directly from text. There are four models in all: a male and a female voice, each available for text written in Tamil script and in Sourashtra script. The models produce audio at 22,050 Hz and work directly on character-level text, so no separate phonemizer is needed.
The stated aim is to expand speech technology for languages with limited digital resources and to support use in digital applications, language learning, accessibility solutions, documentation and research. Applications named include reading Sourashtra text aloud.
The speech corpus behind the models
The models were trained on a Sourashtra speech corpus developed at the SPIRE Lab at IISc Bengaluru. The corpus contains more than 28 hours of gender-balanced recordings from one male and one female voice artist, including composed sentences and sentences drawn from literature, with transcriptions in both Sourashtra and Tamil scripts. The male voice model was trained on about 10,000 utterances and the female voice on about 11,400.
The SPICOR TTS 3.0 Sourashtra Corpus is now available on AIKOSH, BHASHINI’s platform for language datasets and resources. That means researchers, developers and institutions can use it to build further Sourashtra tools.
Why a small language matters
The announcement says Sourashtra is spoken by around 2.48 lakh people according to the 2011 Census, and has a limited digital presence. It says the work goes beyond preservation by bringing the language into the digital ecosystem, and could help community members access digital content, support learners with speech-enabled tools, and improve accessibility for people who rely on spoken content. BHASHINI says it currently supports 36 Indian text languages, 23 Indian voice languages and 35 international languages.
What the announcement does not say
The announcement is a government release rather than a peer-reviewed paper, and it does not publish listening-test scores or other quality measurements, so how natural the voices sound is not independently documented in what I could read. It also does not state licence terms for the models, or how the community was involved beyond the two voice artists. Those are fair questions, not criticisms, and the open availability of the corpus makes it possible for others to test the results.
Synthetic speech tools can also be misused to imitate real voices, a concern that applies to the whole field. The release does not discuss safeguards, and I will not guess at them. For another look at India’s approach to AI policy, see our piece on India AI regulation and the consultation paper due.
My view on Sourashtra text-to-speech
As a writer, I feel this is the kind of AI work that is easy to overlook but matters. Large models usually serve large languages first, and smaller communities wait. Seeing a language with about 2.48 lakh speakers get working voices, and a public corpus, is encouraging. At the same time, two voice artists is a small base, and I would like to see feedback from Sourashtra speakers on how the voices sound in everyday use. For another example of Indian research on AI, see a brain-inspired device from IIT Guwahati.
FAQ: Sourashtra text-to-speech
Who developed the Sourashtra text-to-speech models?
The Digital India BHASHINI Division under MeitY developed them with the Indian Institute of Science, Bengaluru, according to the Press Information Bureau.
How many models are there?
Four: a male and a female voice, each available for Tamil-script and Sourashtra-script text.
Where can the data be found?
The SPICOR TTS 3.0 Sourashtra Corpus, with more than 28 hours of speech, is available on BHASHINI’s AIKOSH platform.