Sarvam AI's Saaras V4 Brings Multilingual Speech-to-Text to India with 22 Languages + English

Sarvam AI Launches Saaras V4: A Game-Changer for Multilingual Speech Recognition

Sarvam AI has just released Saaras V4, a groundbreaking speech-to-text model that covers all 22 Indian languages alongside global English. Available immediately through Sarvam's API at ₹30 per hour, this release marks a significant milestone in democratizing speech recognition technology for India's diverse linguistic landscape.

What Makes Saaras V4 Stand Out?

Saaras V4 combines sophisticated architecture with practical functionality. The model pairs an audio encoder with a 3-billion parameter hybrid state-space decoder, creating a compact yet powerful solution. What sets this release apart is its attention to real-world use cases and developer needs.

Key technical features include:

• Comprehensive Language Coverage: All 22 Indian languages plus English for truly inclusive speech recognition
• Keyterm Prompting: Support for up to 50 custom terms to improve accuracy for domain-specific vocabulary
• 5 Output Modes: Multiple output formats from a single model, reducing integration complexity
• Ultra-Low Latency: First-token latency under 150 milliseconds for real-time streaming applications
• Streaming Capability: Live transcription support for conversational AI and live events

Why This Matters for AI Tool Users

For developers and businesses operating in India or serving Indian-speaking populations, Saaras V4 addresses a critical gap. Most mainstream speech-to-text tools focus primarily on English and a handful of widely-spoken languages. This creates friction for Indian startups, enterprises, and regional applications that need to serve their native-speaking user bases.

The affordable pricing at ₹30 per hour makes enterprise-grade speech recognition accessible to Indian businesses of all sizes. Compared to global alternatives, this positions Saaras V4 as an attractive option for cost-conscious organizations looking to build voice-enabled applications.

The keyterm prompting feature is particularly valuable for specialized domains. Healthcare providers need medical terminology recognized accurately. Legal firms require proper names and case references. E-commerce platforms benefit from product names being transcribed correctly. Saaras V4's ability to handle up to 50 custom terms directly addresses these professional needs.

Implications for the AI Landscape

This release highlights a broader trend: specialized AI models targeting underserved markets are becoming increasingly viable. While major AI labs focus on English-centric, global solutions, companies like Sarvam AI are building tools optimized for specific regions and languages.

The sub-150ms first-token latency is particularly noteworthy in a competitive landscape where real-time performance has become table stakes. Users expect voice applications to respond with minimal delay—Saaras V4 meets that expectation.

Additionally, the ability to get five different output modes from a single model suggests thoughtful API design. Whether developers need raw transcription, structured JSON, or formatted text, they don't need to maintain separate models or endpoints.

The Road Ahead

As voice interfaces become central to how people interact with technology—particularly in markets where text input may be challenging—speech-to-text capabilities become increasingly critical infrastructure. Saaras V4's launch suggests that India-focused AI tools are maturing rapidly and competing effectively on both technical merit and practical value.

This news was originally reported by MarkTechPost.

The Bottom Line

Saaras V4 represents more than just a new speech-to-text model—it's evidence that AI tooling is becoming more inclusive and locally optimized. For anyone building applications in India or serving Hindi, Tamil, Telugu, Kannada, and other Indian language speakers, this tool deserves serious consideration. The combination of comprehensive language support, real-time performance, flexible output formats, and affordable pricing makes it a compelling choice for developers looking to build truly accessible voice applications.