NVIDIA Launches Tools to Break Language Barriers in AI Speech Technology

NVIDIA has introduced new tools that could make it easier for people across Europe to use artificial intelligence (AI) in their own languages — even in those that rarely get digital support.

Out of about 7,000 languages spoken worldwide, only a small fraction are well supported by AI. To close this gap, NVIDIA worked with researchers from Carnegie Mellon University and Fondazione Bruno Kessler to create Granary, one of the world’s largest open speech datasets. It contains about a million hours of audio in 25 European languages, including smaller ones like Maltese, Estonian and Croatian.

AI Speech Technology

This data powers two new AI models: Canary and Parakeet. Canary focuses on accuracy, offering high-quality transcription and translation between English and 24 European languages. Parakeet is designed for speed, able to handle long recordings quickly and efficiently. Both models can automatically detect languages, add punctuation and even provide word-by-word timestamps.

Why does this matter? Until now, many languages lacked enough training data to build accurate AI systems. That meant weaker speech recognition for users in those regions. Granary changes this by offering “clean,” ready-to-use data without the need for time-consuming manual annotation.

The result: more inclusive AI that can power practical applications such as multilingual chatbots, customer service voice agents, or near-real-time translation tools.

With these releases, NVIDIA is opening the door to AI speech technology that doesn’t just work for major languages, but also reflects the linguistic diversity of Europe.

Business News