Sarvam AI Powers Multilingual Video Dubbing And Translation In 11 Indian Languages

New Delhi: Sarvam AI launched Sarvam Studio, an innovation designed to assist content producers in making their work multilingual throughout India. The release is a part of the company's larger series of announcements regarding its push for autonomous AI. 

Sarvam Studio: One Content, Many Languages

The goal of Sarvam Studio is to give content producers the ability to translate a single piece of work into several Indian languages.

With AI-powered video dubbing, Studio can generate high-fidelity dubs in 11 Indian languages. In an expert study cited by the organization, participants preferred Sarvam Studio for overall quality and production readiness. 

Additionally, the technology provides agentic document translation for long-form, genre-neutral content. Evaluations revealed that readers favored Studio's work across a variety of categories, according to Sarvam. Studio received the highest translation quality rating in a head-to-head comparison of translations of actual documents. The business claimed that its products were acknowledged for their constant excellence in a variety of challenging fields, such as academic, fictional, and legal topics.

Another feature highlighted is structure preservation. Without the need for manual redesign, Studio preserves the original document layout, including tables, headings, diagrams, and page hierarchy.

Sarvam Studio aims to enable the production of multilingual content at scale, from textbooks and novels to national addresses and lecture videos.

Expansion Across Speech, Vision and Voice

Alongside Studio, Sarvam revealed updates to several of its AI models. All 22 of the scheduled Indian languages are now supported by Saaras V3, the most recent version of its voice recognition model. The model now incorporates real-time streaming and is built to manage noisy speech and mixed languages. Word-level timestamps, automatic language identification, and multi-speaker audio diarization are further features.

Additionally, the business unveiled Bulbul V3, its most recent text-to-speech model. Bulbul V3 produced the highest listener preference and the lowest error rates across use cases and languages in a third-party, independent human listening survey.

Sarvam Vision, a 3-billion parameter vision-language model, was also announced. According to the corporation, it sets a better standard for Indian languages and is comparable with the best outcomes in English digitalization.

Conversational AI at Scale

Every day, more than a million minutes of conversations are powered by Sarvam's conversational agent platform, Samvaad. According to the company, use cases including 24/7 sales assistants, hybrid onboarding experiences via phone and WhatsApp, and population-scale outreach are growing.

Sarvam claims that because of the close coordination between its AI research and product teams, 80% of its calls are identical to those made by human callers. According to reports, the agents virtually double the capture of sales interest and increase interaction across customer service use cases by more than five percentage points.

The business also emphasized findings from large-scale conversational log analysis. Working with a credit card firm, for instance, helps pinpoint the reasons why consumers decide not to purchase a card.

Push for Sovereign AI

Sarvam AI said it is developing what it calls a full-stack sovereign AI platform grounded in Indian languages and datasets, with services deployed at population scale.

In order to develop at-scale computing, sovereign models, and institutional capacity for AI adoption, the company established strategic agreements with the governments of Tamil Nadu and Odisha. According to the statement, the strategy will integrate AI as a public utility across agencies and be implemented state-wide rather than through limited trial projects.

Additionally, Sarvam presented their multi-agent orchestration platform, Arya. On a typical ETL process, the company claimed that Arya, when combined with GPT 4.1 mini, delivered about five times higher accuracy at roughly ten times lower cost than Claude Code with agent swarm. There are plans to make Arya open-source with a debugging interface and containerized runtime.

ALSO READ: Sarvam AI Launches Saaras V3, Boosting Real-Time Speech Recognition in 22 Languages