Sarvam Launches Voice-Enabled AI Bot for Indian Languages
Indian AI Startup Sarvam Launches Multilingual Voice-Enabled AI Bot Supporting Over 10 Indian Languages
Indian AI startup Sarvam has introduced a voice-enabled AI bot supporting more than 10 Indian languages, aimed at addressing the linguistic diversity within the country. The Bengaluru-based company, which primarily serves businesses and enterprises, recognizes the preference for native language speech over typing. Co-founder Vivek Raghavan emphasized the challenges of typing in Indian languages and highlighted the convenience of voice interaction.
The AI agents offered by Sarvam can be integrated into platforms like WhatsApp and can handle traditional voice calls, with pricing starting at ₹1 per minute. The company has developed its voice-enabled agents using a foundational small language model called Sarvam 2B, trained on a dataset of 4 trillion tokens, all synthetically generated. Raghavan explained the necessity of using synthetic data due to the limited availability of Indian language content on the open web.
Furthermore, Sarvam is launching an audio-language model named Shuka, built on its Saaras v1 audio decoder and Meta’s Llama3-8B Instruct, which will be open-sourced. Additionally, the company is introducing a generative AI workbench, A1, designed to aid lawyers in regulatory lookups, document drafting, redaction, and data extraction.
Sarvam stands as part of a select group of Indian startups promoting AI use cases aligned with the country's interests and supporting the government's efforts to develop bespoke AI infrastructure. The government's IndiaAI program incorporates initiatives like IndiaAI Compute Capacity, aiming to establish a supercomputer powered by at least 10,000 GPUs, and Bhashini, which aims to democratize access to digital services across various Indian languages. Raghavan expressed the company's readiness to collaborate with the government on these initiatives.
Key Takeaways
- Sarvam AI launches voice-enabled bot supporting over 10 Indian languages.
- AI agents priced at ₹1 per minute, deployable on WhatsApp and traditional calls.
- Sarvam 2B model, trained on synthetic data, costs a tenth of industry standards.
- Shuka audio-language model and A1 legal AI workbench also introduced.
- Sarvam supports India's sovereign AI efforts, contributing to the IndiaAI program.
Analysis
Sarvam's introduction of a multilingual voice-enabled AI bot is poised to address India's linguistic diversity, offering increased accessibility for businesses. This move could potentially disrupt the market by providing cost-effective AI solutions at ₹1 per minute, impacting competitors and telecom providers. The utilization of synthetic data for training models reduces reliance on scarce web content, establishing a precedent for future AI development in data-sparse regions. In the long term, Sarvam's alignment with IndiaAI initiatives positions the company to benefit from government support in building national AI infrastructure, potentially leading to broader adoption of AI technologies tailored to Indian needs.
Did You Know?
- Sarvam 2B Model:
- The Sarvam 2B, a foundational small language model developed by Sarvam AI, has been specifically trained on a dataset of 4 trillion tokens. These tokens are synthetically generated to address the scarcity of Indian language content available on the open web. This model serves as the backbone of Sarvam's voice-enabled AI agents, enabling them to effectively understand and process multiple Indian languages.
- Shuka Audio-Language Model:
- The Shuka model, introduced by Sarvam AI, is designed to process and understand spoken language inputs. It is built on the Saaras v1 audio decoder and Meta’s Llama3-8B Instruct, enhancing the capabilities of Sarvam's AI agents in handling voice interactions. The integration of Shuka with Meta’s Llama3-8B Instruct, which will be open-sourced, demonstrates a collaborative approach to advancing AI technology in language processing.
- A1 Generative AI Workbench:
- The A1 workbench, a specialized generative AI tool introduced by Sarvam AI, is tailored for legal professionals. It assists with various tasks such as regulatory lookups, document drafting, redaction, and data extraction, leveraging AI to streamline legal workflows. This tool makes it easier for lawyers to manage and process large volumes of legal documents and data efficiently.