IISc Unveils SraVaani: India's First Multilingual Speech Model for 65 Languages

Bengaluru: Researchers at the Indian Institute of Science (IISc) have released SraVaani, a new speech recognition model that supports 65 Indian languages and dialects. Developed by the SPIRE Lab in collaboration with ARTPARK and Google, SraVaani is being described as the first multilingual Indian automatic speech recognition (ASR) model to cover such a wide range of languages, including more than 40 that are not officially supported by existing speech technology.

Speech recognition systems have traditionally worked well only for a few dominant languages, leaving many regional and non-scheduled languages without reliable speech-to-text capabilities. SraVaani aims to bridge this gap by covering 20 scheduled languages and 45 regional languages and dialects. This could potentially extend speech AI services to about 25 crore people whose languages are not properly handled by current systems, based on the 2011 Census figures.

The model's coverage is designed to be pan-India, spanning 19 languages from the Northeast, 16 from eastern India, nine from the west, eight from the north, six from the south, and five from central India, along with English and Sanskrit. It includes languages such as Garo, Angika, Chakma, Kokborok, Tulu, Bundeli, and Bajjika, which are often left out of mainstream speech technology.

Early results on several of these languages are promising. For instance, SraVaani shows a word error rate of just 9.5% on Garo, compared with 69.4% for the next-best system evaluated. This suggests that the model can achieve high accuracy even for languages with limited prior resources.

“As India builds its own capabilities in artificial intelligence, inclusive language technology must be part of that ambition. At IISc, research has always been in service to the nation, and SraVaani, serving more than 60 Indian languages, is a contribution toward that,” said Govindan Rangarajan, Director of IISc.

The model is freely and publicly available on the Hugging Face platform under an MIT licence, allowing researchers, developers, and startups to use and build upon it without restrictions. This open approach is expected to accelerate the development of speech-based applications for Indian languages, from voice assistants to transcription services, and help bring the benefits of AI to a broader population.

SraVaani's release comes at a time when India is focusing on building self-reliant AI capabilities. By including languages that have been historically underserved, the model represents a step toward more equitable and inclusive AI development. It also highlights the potential of collaborative research between academic institutions, industry, and global technology partners.

While further work is needed to refine the model and expand its coverage, SraVaani offers a solid foundation for future improvements. The research team hopes that making the model publicly available will encourage contributions from the wider community, leading to better support for even more languages and dialects in the future.