Technology
IISc Releases SraVaani Multilingual Speech Recognition Model

IISc Releases SraVaani Multilingual Speech Recognition Model

The Hindu1h
THE BRIEF
1

Researchers at IISc, ARTPARK, and Google launched SraVaani, an AI model supporting 65 Indian languages and dialects, including 45 previously underserved regional languages.

2

The model, available on Hugging Face under an MIT license, covers 20 scheduled and 45 non-scheduled languages, potentially benefiting 25 crore people based on 2011 Census data.

3

SraVaani demonstrated a 9.5% word error rate for the Garo language, significantly outperforming the next-best system which recorded a 69.4% error rate.