Sourashtra Text-to-Speech Models Released by IISc and BHASHINI By Elets News Network - 20 November 2026

Main Building of the Indian Institute of Science in Bengaluru with its clock tower under a blue sky

The Indian Institute of Science (IISc), Bengaluru, and the Digital India BHASHINI Division of the Ministry of Electronics and Information Technology (MeitY) have released openly available Sourashtra text-to-speech (TTS) models and a 28-hour speech corpus. IISc said on 20 November 2026 that this is the first such speech resource built for Sourashtra, a minority Indo-Aryan language spoken mainly in Tamil Nadu.

The release adds a low-resource language to India’s growing stack of public language AI, alongside efforts such as CIIL’s Anuvani speech AI ecosystem for the 22 scheduled languages.


What the Sourashtra text-to-speech models include

The resource was built at IISc’s Speech and Image Processing and Embedded Research (SPIRE) Lab. It has four models: a male and a female voice, each available for text written in Tamil script and in Sourashtra script.

  • Architecture: VITS (Variational Inference with adversarial learning for end-to-end Text-to-Speech), which turns text straight into speech.
  • Training data: about 10,000 utterances for the male voice and about 11,400 for the female voice.
  • Output: audio at 22,050 Hz, working on character-level text with no separate phonemiser needed.
  • Access: the models are published on Hugging Face by SPIRE Lab for developers and researchers.

SPICOR TTS 3.0 corpus opens on AIKosh

The models are trained on the SPICOR TTS 3.0 Sourashtra Corpus, which has more than 28 hours of gender-balanced recordings from one male and one female voice artist. The text covers finance, politics, food, arts, literature, weather and science, with transcriptions in both scripts.


The corpus is open on AIKosh, India’s national platform for AI datasets and models, which already hosts Indian models such as CoRover’s BharatGPT models. MeitY first announced the models in a Press Information Bureau release in October 2026.

Why it matters for Indian language AI

Sourashtra is spoken by about 2.48 lakh people, according to the 2011 Census, and has very little digital presence. Prasanta Kumar Ghosh, professor at IISc, said AI can greatly improve digital access for communities whose languages are underrepresented in current technology, and called the models an important step towards turning written Sourashtra into natural-sounding audio in two scripts.

The models can read Sourashtra text aloud, support language learning and accessibility tools, and give researchers a base for further work. The move fits a wider BHASHINI push into smaller languages, seen this week in the Goa AI Policy’s planned Konkani LLM, and follows commercial launches such as Chariot’s Indic text-to-speech model.

Key takeaways

  • IISc and BHASHINI have released four open Sourashtra text-to-speech models, a first for the language.
  • The SPICOR TTS 3.0 corpus has 28+ hours of speech and is open on AIKosh.
  • Sourashtra has about 2.48 lakh speakers, mostly in Tamil Nadu.

FAQ

What is Sourashtra?

Sourashtra is a minority Indo-Aryan language spoken mainly by the Sourashtra community in Tamil Nadu. It is written in both Tamil script and its own Sourashtra script.

Where can developers get the Sourashtra TTS models?

The four models are on SPIRE Lab’s Hugging Face page, and the SPICOR TTS 3.0 speech corpus is available through AIKosh.

Featured image: Main Building, Indian Institute of Science, Bengaluru. Photo: Sayantan0510 via Wikimedia Commons (CC BY-SA 4.0).

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!

Related artificial intelligence