Gnani.ai unveils Inya VoiceOS under India AI Mission

Gnani.ai has announced the deployment of Inya VoiceOS, described as India’s first five-billion-parameter voice-to-voice foundational AI model developed under the India AI Mission. The model was inaugurated by Narendra Modi at the India AI Impact Summit 2026, where Co-Founder and CEO Ganesh Gopalan demonstrated the technology. The current version is being released as a research preview ahead of a planned 14-billion-parameter voice-to-voice model.

Inya VoiceOS is designed to deliver end-to-end spoken intelligence by operating directly in acoustic and semantic space, removing the need for intermediate speech-to-text and text-to-speech stages. The company states that this architecture allows the system to process and generate speech natively rather than relying on cascaded pipelines.

Model Scale and Training

According to Gnani.ai, Inya VoiceOS has been trained on a large sovereign voice dataset focused on Indian languages. The model comprises five billion parameters and has been pre-trained on more than 14 million hours of multilingual speech data. It has also undergone over 1.2 million hours of task-specific speech fine-tuning and incorporates more than eight trillion text tokens to support linguistic grounding and reasoning.

Native End-To-End Speech Processing

Unlike traditional cascaded systems, the model consumes and generates speech tokens directly. It jointly encodes phonetics, prosody, semantics and intent, and is designed to preserve paralinguistic cues such as tone, emotion, pacing and pauses. The system supports streaming and interruption-aware inference, and can handle overlapping speech and mid-utterance corrections without resetting the conversation.

Gnani.ai reports sub-second end-to-end latency and 24 kHz audio output with natural prosody. The model provides native support for more than 15 Indian languages and is designed to manage code-mixed speech.

Potential Applications and Availability

The company indicates that the system can be deployed in government services, including helplines, grievance redressal platforms and emergency response systems, to enable multilingual conversational interfaces. Enterprise applications include voice-driven workflows across banking and financial services, healthcare, insurance and logistics.

Gnani.ai states that Inya VoiceOS has been built, trained and deployed entirely within India, with an emphasis on maintaining data sovereignty.

Speaking at the event, Ganesh Gopalan said that voice-to-voice AI represents a shift in architecture intended to preserve the qualities of human conversation, including emotion and context, while improving accuracy compared with conventional cascaded systems.