IISc releases voice AI model for 65 Indian languages

The Indian Institute of Science (IISc) has released an open-source voice AI model, SraVaani, that recognises speech in 65 Indian languages and dialects, the Times of India reports. Separately, Bengaluru startup Sarvam AI launched Epoch Builder Edition, a platform for building and deploying large language models for Indian languages, according to The Hindu.

SraVaani, developed by IISc's SPIRE Lab with ARTPARK and Google support, performs on par with leading systems on common languages and far better on underserved ones like Garo. It is freely available on Hugging Face. Sarvam AI's platform offers 7B and 70B parameter multilingual LLMs trained on 2 trillion tokens, with private preview from August 2026, the company said.

A third effort, by AI4Bharat lab at IIT Madras, involves collecting speech data from over 500 districts to train AI for Indian languages. The team records natural conversations to capture accents and dialects, said researcher Kaushal Bhogale to The Hindu. Together, these initiatives aim to make voice AI accessible to all Indians.