Shunya Labs, a startup from the Nasscom GenAI Cohort 1, has announced the launch of Vāķ, India’s largest open-weight voice AI model. The release was unveiled in partnership with Nasscom at the India AI Impact Summit 2026 held at Bharat Mandapam in New Delhi.
Vāķ is a real-time speech translation system supporting all 2,970 language combinations across India’s 55 mother tongues. It is launched alongside the world’s top-ranked open-weight automatic speech recognition (ASR) model and a neural text-to-speech (TTS) engine. All three components are released together and are available free for deployment.
India, home to 1.4 billion people and over a thousand language varieties — including 61 languages spoken by more than one million people — has long faced limitations in speech AI coverage. Most existing systems support only 5 to 10 Indian languages, forcing developers, startups, and government institutions to depend on foreign APIs and overseas servers. This places citizen voice data under external control while leaving large linguistic populations underserved. Many Indian languages are also not mutually intelligible, even within the same language family, making true linguistic inclusion far more complex than surface-level multilingual support. As a result, widely spoken languages such as Bhojpuri, Rajasthani, Chhattisgarhi, and Magahi — each used by tens of millions — have largely remained outside mainstream AI systems. Vāķ is designed to bridge this gap by offering real-time, comprehensive, and sovereign language coverage that reflects India’s full linguistic diversity.
At its core, Vāķ functions as a real-time Any-to-Any translation system that converts speech from any of India’s 55 languages into any of the remaining 54, with end-to-end latency of under 1.5 seconds. Beyond translation, it preserves the speaker’s voice, tone, and emotional expression across languages. The system also supports zero-shot voice cloning, meaning no prior voice data is required.
The name Vāķ draws from ancient Vedic philosophy, where it represents the power of speech — the force that transforms thought into expression and gives voice to knowledge. One of the oldest concepts in Sanskrit tradition, the name reflects how the technology is rooted in India’s own intellectual heritage. With full model weights publicly available, organisations can deploy Vāķ locally on their own infrastructure, ensuring that no data leaves their environment.
The release includes three model families launched simultaneously. Pingala ASR ranks first on the Hugging Face OpenASR Leaderboard, achieving a record-low Word Error Rate of 3.10%. It features a CPU-first architecture with sub-250 millisecond latency, making it suitable for edge and offline use. Vāķ’s neutral TTS produces natural-sounding speech across all 55 languages, with support for streaming output, custom voice creation, zero-shot voice cloning, and control over prosody and emotion. The real-time Any-to-Any translation engine enables conversion across all 2,970 language pairs while preserving speaker identity and emotional nuance within 1.5 seconds.
These capabilities allow organisations to deploy the system entirely on their own infrastructure, without relying on foreign servers, recurring API fees, or external dependencies. This is particularly critical for government agencies, healthcare providers serving rural populations, and courts delivering justice in local languages, where data sovereignty is essential. With all 2,970 translation combinations available out of the box, Vāķ makes large-scale multilingual citizen services, healthcare outreach, judicial access, and education delivery technically and economically viable for the first time nationwide.
Commenting on the launch, Sourav Bandyopadhyay, Founder and Chief Scientist of Shunya Labs, said, “India should not have to rely on foreign APIs to understand its own people. With Vāķ, we are releasing the world’s best-performing open-weight voice AI across 55 Indian languages — covering every language spoken by more than a million people. This is sovereign, self-funded innovation developed within the Nasscom ecosystem. Our vision is simple: every developer can build, every government can deploy, and every Indian can be heard in their own language.”
Ankit Bose, Head of AI at Nasscom, added, “Shunya Labs, as a Nasscom GenAI Cohort 1 startup, represents the globally competitive innovation emerging from India’s AI ecosystem. Developing high-performance open-weight models across Indian languages demonstrates both technical excellence and a strong commitment to inclusive and sovereign AI. The launch of Vāķ strengthens the role of the Nasscom GenAI ecosystem in helping startups innovate at global standards while contributing to India’s digital future.”
Vāķ covers India’s full linguistic spectrum, including 43 Indo-Aryan languages such as Hindi, Bengali, Marathi, Gujarati, Urdu, Bhojpuri, and Rajasthani; seven Dravidian languages including Telugu, Tamil, Kannada, and Malayalam; three Sino-Tibetan languages including Meitei (Manipuri), Bodo, and Garo; one Austroasiatic language, Santali; and Indian English. Together, this represents more than 1.17 billion native speakers.
The launch also aligns with national initiatives such as the IndiaAI Mission, Digital India, BHASHINI, and Atmanirbhar Bharat.