Bengaluru-Based Gnani.ai Launches Indic Speech-to-Text Model Under IndiaAI Mission

Bengaluru-based conversational AI company Gnani.ai has launched Vachana, a new speech-to-text (STT) model trained on more than one million hours of real-world voice data, following its selection under the government-backed IndiaAI Mission.

Designed as enterprise-grade speech recognition infrastructure, Vachana STT supports multiple Indian languages and is built for deployment across sectors such as banking, telecom, and customer support.

The Bengaluru-based startup said the model is engineered to address India’s linguistic diversity and complex speech patterns at a foundational systems level, rather than as a simple localisation layer.

“Speech recognition in India is not a localisation problem. It is a foundational systems problem,” said Ganesh Gopalan, co-founder and chief executive officer of Gnani.ai. “Vachana STT is built as core infrastructure, trained on how India actually speaks, and designed to operate across channels.”

Vachana STT is the first release in Gnani.ai’s upcoming VoiceOS stack and is available immediately via API for enterprise customers. Early adopters will receive up to one lakh free minutes of usage, the company said.

According to Gnani.ai, the model has been trained on proprietary multilingual datasets spanning more than 1,056 domains. It supports both real-time and batch transcription and is already deployed at scale, processing around 10 million calls per day with a P95 latency of 200 milliseconds.

In evaluations using internal and public datasets, Vachana STT achieved 30–40% lower word error rates for low-resource Indian languages and 10–20% lower error rates for the country’s eight most widely used languages, including Hindi, Tamil, Telugu, Kannada, Bengali and Marathi.

Gnani.ai said the model is built to handle compressed audio, fluctuating network conditions and high concurrency, making it suitable for compliance monitoring, analytics and voice-driven enterprise workflows.