India’s push for home-grown, language-first AI has a new public building block. The Central Institute of Indian Languages (CIIL), Mysuru, which functions under the Department of Higher Education in the Ministry of Education, has launched Anuvani, an open-source ecosystem of speech AI models built for the country’s linguistic diversity.

The first release, announced on Tuesday, comprises two models. Anuvani-LID identifies which language is being spoken, while Anuvani-Diarization works out who is speaking and when. Both support all 22 constitutionally scheduled Indian languages as well as Indian English.

Open weights, ready-to-deploy toolkits

Rather than releasing models alone, CIIL is shipping each one as a complete package: peer-reviewed research papers, open-source model weights, on-device toolkits for mobile phones, and production-grade NVIDIA Triton Inference Server kits. The institute said the aim is to let developers, researchers, startups and public bodies deploy Indian-language speech AI without proprietary licensing hurdles, whether on cloud clusters or offline smartphones.

The models target problems that routinely trip up speech systems in India: conversations that switch between closely related languages, wide regional accent variation, people talking over one another, and noisy recording conditions.

What the models can do

• Anuvani-LID is built on a 115.6-million-parameter Conformer architecture and was trained on 54,643 hours of Indian speech spanning more than 31 million utterances, according to CIIL Academic Secretary Tariq Khan.
• Anuvani-Diarization has 76.1 million parameters and is designed to separate speakers in phone calls, meetings, large-room recordings and broadcast audio, including fast speaker changes and overlapping speech.

In a zero-shot evaluation on IndicDiarBench, covering 1,164 recordings and 107.7 hours of audio across all 22 scheduled languages, CIIL said the diarisation model posted the lowest Diarization Error Rate among the 18 systems it tested, using a strict scoring standard that counts overlapping speech.

From linguistic heritage to AI infrastructure

“For over five decades, CIIL has worked to preserve, document and advance India’s rich linguistic heritage. With the launch of the Anuvani speech AI ecosystem, we are bringing that deep linguistic foundation into the frontier AI era,” said Basavaraja Kodagunti, Director, CIIL. He added that releasing papers, weights and deployment kits openly would keep Indian languages a shared public resource and help citizens, developers and institutions take part in the digital economy in their mother tongues.

Why it matters

Language identification and speaker diarisation are foundational layers for voice assistants, call-centre analytics, transcription, captioning and citizen-service helplines. For enterprises and government departments building voice-first services for Bharat, openly licensed, India-trained components that can run on-device lower both cost and data-sovereignty concerns. Anuvani adds to a growing public stack of Indic AI resources alongside the IndiaAI Mission’s efforts, and could shorten the path from pilot to production for speech applications across the country’s many languages.

Image for representational purposes. Photo: Galak76 / Wikimedia Commons (CC BY-SA 3.0).

Be a part of Elets Collaborative Initiatives. Join Us for Upcoming Events and explore business opportunities. Like us on Facebook , connect with us on LinkedIn and follow us on Twitter.

"Exciting news! Elets technomedia is now on WhatsApp Channels Subscribe today by clicking the link and stay updated with the latest insights!" Click here!