NVIDIA Unleashes Speech AI Revolution for Europe’s Overlooked Languages

Artificial intelligence may seem everywhere, but it speaks only a fraction of humanity’s 7,000 languages—leaving vast communities digitally sidelined. NVIDIA aims to change that with the release of a sweeping open-source toolkit empowering developers to create advanced speech AI for 25 European languages. While the lineup covers major tongues, its real breakthrough lies in supporting those routinely ignored by big tech—languages like Croatian, Estonian, and Maltese—unlocking opportunities for multilingual chatbots, real-time translation services, and voice-driven applications that truly understand their users.

At the heart of this initiative is Granary, a colossal speech library amassing roughly a million hours of human audio, meticulously curated to train AI in the subtleties of recognition and translation. Complementing this treasure trove are two new models: Canary-1b-v2, a powerhouse for high-precision transcription and translation, and Parakeet-tdt-0.6b-v3, engineered for lightning-fast real-time tasks. Both are already available on Hugging Face, with Granary’s research debuting at the Interspeech conference in the Netherlands. Remarkably, NVIDIA’s collaboration with Carnegie Mellon University and Fondazione Bruno Kessler sidestepped the traditional, slow grind of manual annotation by building an automated pipeline using its NeMo toolkit, transforming raw, unlabelled audio into clean, high-quality training data.

The results are striking. Granary’s efficiency means developers need only half the data volume to hit accuracy benchmarks compared to other popular datasets. Canary delivers translation and transcription quality that challenges models triple its size—while running up to ten times faster—while Parakeet can digest a 24-minute multilingual meeting in one pass, intelligently detecting languages and preserving professional formatting with word-level timestamps. By making these resources open and accessible, NVIDIA is sparking a new wave of AI innovation—one where technology finally speaks to people in their own language, no matter how small the community.