- The Indian Institute of Science (IISc) in Bengaluru released SraVaani on 13 August 2026 as an open-source multilingual speech-recognition model.
SraVaani, released by the Indian Institute of Science (IISc) on 13 August 2026, is an open-source multilingual speech-recognition model developed by the SPIRE Lab in collaboration with ARTPARK and Google.1
The model converts spoken words into text in 65 Indian languages and dialects, covering 20 scheduled Indian languages and 45 regional languages such as Garo, Angika, Chakma, Kokborok, Tulu, Bundeli, and Bajjika.
It supports 10 scripts and includes automatic language identification, making it a versatile tool for diverse linguistic needs.
SraVaani was trained using data from over 31,000 hours of natural conversational speech, collected from 156,000 people across 165 districts in 28 Indian states.
Performance data indicates a 9.5% word error rate for Garo, significantly outperforming the next-best tested system, which recorded a 69.4% word error rate for the same language.
The model is publicly available on the Hugging Face platform under an MIT licence, promoting accessibility and further development in the field of speech recognition.
“SraVaani is developed at the Indian Institute of Science in Bengaluru, targeting Indic languages. The model's open-source release aims to foster broader research and development in speech recognition for Indian languages.”
GK Today
