IISc launches SraVaani speech recognition model for 65 Indian languages
IISc, teaming up with ARTPARK and Google, has launched SraVaani, a speech-to-text model trained on 65 Indian languages and dialects.
What's cool is that it finally brings tech support to over 40 languages that today's speech recognition systems do not officially support, making things way more inclusive for millions of speakers across the country.
SraVaani reaches about 250 million people
SraVaani covers all regions of India and supports languages like Garo, Tulu, Chakma, and Angika, reaching about 250 million people.
It's free to use on Hugging Face under an MIT license.
The model also nails accuracy: it scored a word error rate of just 9.5% on Garo (compared to the next-best system evaluated's whopping 69.4%).
IISc Director Govindan Rangarajan calls it a big step toward truly inclusive AI in India.