A 5B-parameter voice-to-voice AI model designed to transform conversational intelligence across India.
Gnani.ai has unveiled Inya VoiceOS, India’s first 5-billion-parameter voice-to-voice foundational AI model, developed under the India AI Mission. The model was showcased at the India AI Impact Summit 2026 by Co-Founder and CEO Ganesh Gopalan and inaugurated by Prime Minister Narendra Modi. Positioned as a research preview, Inya VoiceOS precedes the company’s upcoming 14-billion-parameter flagship model.
Unlike conventional speech AI, Inya VoiceOS bypasses the traditional speech-to-text and text-to-speech stages, operating directly in acoustic and semantic space. This end-to-end spoken intelligence allows the AI to encode phonetics, prosody, semantics, and intent simultaneously, preserving nuances such as tone, pacing, and emotion.
Training at Scale
The model has been trained on one of the largest sovereign voice datasets for Indian languages: 5 billion parameters, over 14 million hours of multilingual speech pretraining data, 1.2 million hours of task-specific fine-tuning, and more than 8 trillion text tokens for linguistic reasoning.
Capabilities and Performance
Inya VoiceOS supports streaming, interruption-aware inference, overlapping speech, and mid-utterance corrections without conversation resets. Technical highlights include sub-second end-to-end latency, 24 kHz audio output with natural prosody, native support for 15+ Indian languages, and robust handling of code-mixed speech.
Applications Across Sectors
The model is expected to power government services such as multilingual citizen helplines, grievance redressal, and emergency response, enabling emotion-aware and natural interactions. In enterprises, it can drive voice-operated workflows in BFSI, healthcare, insurance, and logistics, offering context-aware, hands-free conversational automation. Built entirely in India, Inya VoiceOS ensures full data sovereignty while placing Gnani.ai among a handful of global firms capable of deploying advanced voice-to-voice AI architectures.Gopalan emphasized the model’s unique approach: “Voice-to-voice AI is not just about speed; it’s about preserving what makes human conversation effective. Inya VoiceOS adds emotion and context to every interaction, delivering significantly higher accuracy than traditional cascaded systems.”
