Gnani.AI to launch its voice foundation model at AI Impact Summit | Start Ups
Ananth Nagaraj, co-founder and chief technology officer of Gnani.AI, one of the AI startups selected under the IndiaAI Mission, said that it will launch the voice foundation model at the India AI Impact Summit next week.
The model is built on a 14-billion-parameter voice AI base model that provides multilingual, real-time speech processing with advanced reasoning capabilities. The model is designed for low-latency, speech-to-speech communication and is intended for applications in customer support, training, accessibility, and public-facing systems.
“We are launching our core voice-to-voice model in six languages and aim to expand it to all 22 languages in the next 18 months,” Nagaraj said. The six languages are English, Hindi, Kannada, Telugu, Tamil and Gujarati.
Gnani will also launch a multilingual text-to-speech model with the ability to clone voices in a hyper-realistic way. This system, called Vachana STT, is trained on more than one million hours of real-world audio data, covering more than 1,056 areas.
“It’s going to be a voice plus avatar. So if you also look at our avatar that we’re going to use the customer service representative, you’ll see that it’s a digital twin. So in our voice-to-voice model or speech-to-text and our LLM understanding layer, they’re all going to use that avatar along with text-to-speech.”
According to Infosys chairman Nandan Nilekani, Voice AI is the only practical interface and is expected to be the next big thing that will bring true digital equality in India. “Just as UPI made digital payments effortless for everyone, voice-driven interfaces can remove barriers to opportunity for every citizen in agriculture, education and other sectors. Literacy will no longer be a barrier,” he said in a statement last month.
When asked about this, Nagaraj also agreed.
“Human-machine interaction will be through voice. Since India is so diverse, it will have voice AI in its native language. I think voice AI will introduce technology to almost 100 crore users once it unlocks human-machine interaction through native languages.”
Vivek Raghavan, co-founder of Sarvam AI, said in November that he was also hopeful of coming up with the independent AI model.
Alongside Sarvam and Gnani, Soket will develop India’s first open-source 120 billion-parameter base model optimized for the country’s linguistic diversity, targeting sectors such as defence, healthcare and education. And Gan AI will build a 70 billion-parameter multilingual base model targeting text-to-speech capabilities.
Gnani, which counts Samsung Venture Investment among its investors, expects to end this financial year with revenues of $20 million, up from just ₹56 crore in the previous year, as it works with more financial services firms such as HDFC Bank, Bank of Baroda and IDFC First Bank.
