ForumIAS LATEST
- 17 August | Navigating the Crest & Trough of Rankforgers by Mr. Ayush Sinha Click Here to Watch →
- 16 August | Are you the Average of the 5 People Around You by Mr. Ayush Sinha | Click Here to Watch →
- 16 August | First UPSC Mains Don't Chase AIR 1 by Mr Ayush Sinha | Click Here to Watch →
News: IISc researchers, with ARTPARK and Google, have released SraVaani, a multilingual speech recognition model for Indian languages.
About SraVaani
- SraVaani is a multilingual Indian speech recognition model designed to support regional and non-scheduled Indian languages.
- Developed by: It was developed by researchers at the Indian Institute of Science (IISc)’s SPIRE Lab, in collaboration with ARTPARK.
- Supported by: Google supported its development.
- Based on model: SraVaani is based on Project Vaani, one of IISc’s major programmes to understand the linguistic diversity of India.
- Key features:
- Speech recognition architecture: It uses a FastConformer-based architecture for automatic speech recognition.
- Wide language coverage: It covers 20 scheduled languages and 45 regional languages and dialects, including Garo, Angika, Chakma, Kokborok, Tulu, Bundeli and Bajjika.
- Pan-India coverage: It includes languages from the Northeast, eastern, western, northern, southern and central regions, along with English and Sanskrit.
- Strong performance: It was evaluated on eight public benchmark datasets and achieved the lowest average word error rate among the systems evaluated.
- For Garo, it recorded a 9.5% word error rate, compared with 69.4% for the next-best system.
- Open access: It is freely available on Hugging Face under an MIT licence, along with a demo and fine-tuning code.
- Wider applications: SraVaani can support further work in regional-language AI, low-resource speech-to-text, automatic language detection, dialect recognition and sovereign AI applications.
