IIT Madras' CoE Rolls Out Four AI Models for Indian Languages

CHENNAI: Bodhan AI, a Centre of Excellence in AI for Education incubated at the Indian Institute of Technology-Madras, has launched four foundational AI models for Indian languages, making them available to the wider technology and education ecosystem as Digital Public Goods.

Developed in partnership with AI4Bharat, the models cover speech recognition, speech generation, machine translation and optical character recognition (OCR). They are being released with open weights and hosted APIs on sovereign digital infrastructure, enabling developers and institutions to adapt and build applications using the models.

The speech recognition model supports 27 Indian languages while the OCR model covers 23 languages. Bodhan-Translate supports 22 languages and the text-to-speech model supports 23 languages.

The models have been trained and optimised using NVIDIA Nemotron open models and technologies, including the NVIDIA NeMo framework. Bodhan AI has post-trained NVIDIA Nemotron 3.5 ASR to support Indian languages, including regional dialects and accents.

The initiative forms part of the broader Bharat EduAI Stack, envisioned as sovereign digital public infrastructure for education and aimed at developing AI capabilities for India's multilingual education ecosystem.

IIT-M Director V. Kamakoti said India's AI journey needed technology that “understands India”, adding that the new voice and vision models marked an important step towards building sovereign digital public infrastructure for AI in education.

Bodhan AI and NVIDIA will also collaborate on datasets, training methodologies and evaluations for future foundational models for Indian languages. AI4Bharat will contribute its expertise in open-source datasets, tools and models for Indian-language AI.