Confucius R2T2 announced as an open-source streaming speech recognition model

A user shared NetEase Youdao's Confucius R2T2 as an open-source release, linking to GitHub and the project website. The author highlighted domain-aware recognition, one stack for every latency target, and a design that treats context as a core input. The website described R2T2 as an open-source streaming speech recognition model, but listed its model weights and code as “coming soon.”