Sarvam AI to Unveil Its 1st Commercial Voice-to-Voice Endpoint Tool with 10 Indian Languages in Next 6-12 Months
In an exciting development for startups and businesses aiming to integrate voice experiences into their services, Sarvam AI, a forefront Artificial Intelligence (AI) startup dedicated to crafting foundational models for Indian languages, has announced plans to release its first commercial voice-to-voice endpoint tool. This innovative tool is set to revolutionize customer support and various voice-related functions across industries.
During a session at the SaaSBoomi annual event held in Chennai on March 7, Sarvam AI’s co-founder, Vivek Raghavan, shared insights about the upcoming tool. “You can expect voice-to-voice endpoints supporting at least 10 Indian languages, providing users with exemplary experiences built on this technology, alongside example experiences for developers to utilize and enhance,” Raghavan stated.
The announcement comes in the wake of Sarvam AI’s recent release of OpenHathi-Hi-v0.1, the inaugural Hindi large language model (LLM) in the OpenHathi series. Developed atop Meta AI’s Llama2-7B architecture, this model is designed to perform at par with GPT-3.5 specifically for Indic languages, potentially positioning it as a significant advancement in AI for the Indian demographic.
Raghavan highlighted OpenHathi’s enhanced English to Hindi translation capabilities, indicating a superior performance when compared against GPT-4 and GPT-3 models. Nevertheless, the journey to perfecting a foundational model for Indian languages is met with considerable challenges, particularly in data collection, the high costs of tokenization, and evaluating innovative projects.
The firm is navigating these challenges head-on, building on the momentum within the Indian AI ecosystem. A notable mention during Raghavan’s address was the launch of IndicVoices by IIT Madras’ AI4Bharat research lab. Released on March 6, IndicVoices is an open-source natural speech dataset encompassing 22 Indian languages. This project, aimed at collecting spontaneous speech data, is expected to significantly benefit AI innovations within the subcontinent.
“This is something very good for the ecosystem. The data from AI4 Bharat will be high-quality and beneficial for startups. However, there’s still much work to be done towards building a foundational model,” added Raghavan, acknowledging the progress and the road ahead for Indian language AI models.
Sarvam AI was inaugurated in July 2023 by Vivek Raghavan and Pratyush Kumar, both of whom previously worked at AI4Bharat, an initiative supported by Infosys co-founder Nandan Nilekani. The startup quickly gained momentum, becoming the first Indian AI startup to secure $41 million in its Series A funding round in December 2023. The round was led by Lightspeed Ventures with contributions from Peak XV Partners and Khosla Ventures.
With a mission to develop a full-stack offering for Generative AI, Sarvam AI is set to embark on a journey from pioneering research-led innovations in custom AI model training to providing an enterprise-grade platform for the authoring and deployment of AI applications. The startup aims to collaborate with Indian enterprises, co-developing domain-specific AI models to leverage their proprietary data.
As the landscape of AI continues to evolve, Sarvam AI’s upcoming tool signifies a crucial step forward in making advanced voice technologies accessible and relevant across various sectors, bolstering customer experience and operational efficiency through the power of AI. The focus on Indian languages not only underlines the increasing importance of localized solutions in the global tech sphere but also marks an exciting chapter in India’s AI innovation journey.
