BHASHINI Unveils AI Powered Speech Technology for Sourashtra Language

In a major advancement for minority language preservation and digital accessibility, the Digital India BHASHINI Division under the Ministry of Electronics and Information Technology has partnered with the Indian Institute of Science, Bengaluru, to build advanced Text-to-Speech models for Sourashtra. Spoken primarily by the Sourashtra community in Tamil Nadu, this low-resource Indo-Aryan language now gains a crucial digital footprint through cutting-edge artificial intelligence.

Developed using the VITS deep-learning architecture, the new speech engine generates natural-sounding synthetic audio directly from character-level text without requiring a separate phonemizer. The models were trained on the high-quality SPICOR TTS 3.0 Sourashtra Corpus created at the SPIRE Lab at IISc Bengaluru. Featuring over 28 hours of gender-balanced audio recordings across more than 21,400 clean utterances, the dataset incorporates literary works and everyday conversational sentences in both the native Sourashtra script and the Tamil script.

The rollout includes four distinct speech models, providing male and female voices capable of processing text inputs written in either Sourashtra or Tamil script. By hosting the underlying speech dataset on BHASHINI’s AIKOSH platform, the initiative provides open resources for researchers, developers, and institutions to build next-generation educational tools, screen readers, and public service applications. This milestone bridges the digital gap for over 2.48 lakh Sourashtra speakers, ensuring that heritage languages stay vibrant in India’s evolving digital landscape.