Confucius4-R2T2 is a stable-prefix streaming automatic speech recognition model for truly real-time transcription.
Confucius4-R2T2 streams speech into append-only text via stable-prefix supervision, with public code for real-time voice agents and translation.