Gnani.ai Launches Indic Speech-To-Text Model Under IndiaAI Mission

Bengaluru-based generative AI startup Gnani.ai has launched Vachana STT, a speech-to-text model built for Indian languages, under the IndiaAI Mission. The startup said the model has been trained on more than 1 Mn hours of real-world voice data and is part of its upcoming voice technology stack, VoiceOS.

Gnani.ai claims the model delivers lower error rates across several Indian languages compared to existing speech recognition systems. According to the startup, Vachana STT performs better in noisy, real-world conditions such as call centres and customer support environments, where audio quality is often poor.

The model supports multiple Indian languages, including Hindi, Bengali, Tamil, Telugu, Kannada, Malayalam, Marathi, Gujarati, Punjabi, Odia and Assamese. Gnani.ai said the system has been trained across over 1,000 domains and does not need additional fine-tuning for different use cases.

The startup claims that Vachana STT is already being used by enterprises in sectors such as banking, telecom and customer support. It said the system currently processes around 10 Mn calls a day, with near real-time transcription.

The launch is part of the government-backed IndiaAI Mission, under which a small group of startups has been selected to build core AI models within the country. Gnani.ai is among the companies chosen to develop what the government calls “sovereign” AI infrastructure.

The model is available through APIs for enterprise customers, with limited free usage offered to early users.

Gnani.ai was selected under the IndiaAI Mission earlier this year as one of the startups tasked with building core AI infrastructure from India. The decision was taken after the IT ministry evaluated over 500 proposals received since the call for applications opened in January.

Gnani.ai was mandated to develop a large-scale voice AI foundational model focused on Indian languages and real-world speech use cases. 

As per the government’s announcement in May, Gnani.ai was building a 14 Bn-parameter voice AI model designed for multilingual, real-time speech processing.

The project falls under the foundational model pillar of the IndiaAI Mission, which aims to create indigenous AI models trained on India-specific data instead of relying on global APIs or foreign-built systems. The ministry had said the chosen startups would receive access to national compute infrastructure and other incentives to build these models at scale.