What happened
IISc’s SPIRE Lab, working with ARTPARK and Google, released SraVaani, described as the first multilingual Indian speech recognition model trained on 65 Indian languages and dialects. The release focuses on expanding speech-to-text capabilities beyond the set of languages that most existing speech recognition systems support officially.
Key details reported in the release include:
Coverage: SraVaani is reported to include 20 scheduled languages and 45 regional languages and dialects, for a total of 65.Accessibility: SraVaani is openly available on Hugging Face.Licence: The model is shared under an MIT licence.Design goal: The model is designed to provide speech-to-text for underserved languages—languages that often get weaker support from dominant-language speech AI.
The release also reports evaluation performance against a next-best system for at least one language pair:
Garo word error rate: 9.5% for SraVaani versus 69.4% for the next-best system (as reported in the release’s evaluation highlights).
Related current affairs
- Jitin Prasada holds bilateral meetings in Morocco on trade, investment and digital transformation
- What changes when AI moves from reading viral genomes to designing them?
- Google’s AI advantage faces its biggest test yet
- Parliament Question: Government schemes and missions for Artificial Intelligence, Research and Innovation
- TDB-DST supports Griva Healthcare for commercialization of indigenous AI-powered cervical cancer screening technology (GrivaVision)
- Prime Minister inaugurates “Param Pragya” AI-based high-performance computing and project cluster facility at IIT Delhi