Central Institute of Indian Languages launches Anuvani speech AI ecosystem
Big news for language tech: The Central Institute of Indian Languages (CIIL) just rolled out Anuvani, an open source speech AI ecosystem that covers all 22 constitutionally scheduled Indian languages plus Indian English.
Announced by CIIL, Anuvani comes with two main tools: one that can tell which language is being spoken (Anuvani-LID), and another that can figure out who's talking when voices overlap (Anuvani-Diarization).
Everything's free to use, so developers and organizations can build on it without any licensing headaches.
Anuvani LID uses 115.6-million-parameter conformer
Indian languages are super diverse, with tons of accents and speakers often talking over each other.
That's why CIIL director Basavaraja Kodagunti says it was important to blend India's rich linguistic heritage with cutting-edge AI.
The models are seriously powerful: Anuvani-LID was built on a 115.6-million-parameter Conformer architecture and trained on 54,643 hours of Indian speech comprising more than 31 million utterances, while the speaker tool beat out 17 other systems at telling voices apart, even in tricky situations.
Plus, both tools are easy to set up on different platforms, making them accessible for everyone looking to bring more Indian languages online.