Robust Knowledge Distillation from RNN-T Models With Noisy Training Labels Using Full-Sum Loss

Open in new window