At the India AI Impact Summit, Shunya Labs introduced Vaķ, an open-weight voice AI and real-time translation system supporting 55 Indian languages and 2,970 language-pair translations. Vaķ includes a top-ranked open-weight speech recognition model and a neural text-to-speech engine, all available for free deployment.
The system, Vaķ, offers real-time “Any-to-Any Translation” across India’s 55 languages, ensuring seamless conversion between them with less than 1.5 seconds of latency. Vaķ also presents zero-shot voice cloning, enabling natural-sounding output in all 55 languages with features like custom voice creation and emotion control.
With the Pingala ASR ranking first on the Hugging Face OpenASR leaderboard, Vaķ boasts a CPU-first architecture providing sub-250ms latency for edge and offline use. The model weights are accessible for local deployment, empowering organizations to utilize the models on-premises without relying on foreign servers.
Vaķ by Shunya Labs facilitates cross-linguistic citizen services, healthcare outreach, judicial access, and educational delivery on a national scale, making these services technically and economically feasible. The platform covers a wide range of languages, including Indo-Aryan, Dravidian, Sino-Tibetan, Austroasiatic, and Indian English, reaching over 1.17 billion native speakers collectively.
