India AI Impact Summit 2026: Gnani.ai launches 5-billion-parameter voice AI model under IndiaAI Mission

At the India AI Impact Summit 2026, held on February 17, homegrown AI company Gnani.ai introduced an innovative voice-to-voice foundational model featuring an impressive 5 billion parameters. This cutting-edge technology is designed to facilitate seamless spoken interactions across a variety of Indian languages, as part of the nation’s initiative to develop its own AI capabilities. The launch was graced by Prime Minister Narendra Modi, who inaugurated the Inya VoiceOS model, which is currently available as a research preview. This model represents a departure from traditional speech systems that typically convert voice to text and back again. Instead, Inya VoiceOS directly processes audio, allowing for more fluid conversations and quicker responses. Ganesh Gopalan, the co-founder and CEO of Gnani.ai, emphasized that voice-to-voice AI transcends mere speed improvements. He explained, "It is about a fundamentally different architecture that preserves the qualities that make human conversation effective." With Inya VoiceOS, the company aims to infuse emotion and contextual understanding into each interaction, achieving significantly higher accuracy compared to conventional systems. The Bengaluru-based firm disclosed that the model was trained on over 14 million hours of multilingual speech data and utilized more than 8 trillion text tokens. It supports more than 15 Indian languages, operating directly within acoustic and semantic spaces. This enables the model to recognize tone, emotion, and pacing, while also managing interruptions and overlapping speech without needing to restart conversations. Gnani.ai envisions the technology being applied in various sectors, including government helplines, emergency response systems, and enterprise applications in banking, healthcare, and logistics. Notably, the entire development, training, and deployment of this model occurred within India, reinforcing the country’s commitment to data sovereignty in AI. Looking ahead, Gnani.ai plans to release a more extensive 14-billion-parameter model following this initial research preview.