Sampling-Based Pruned Knowledge Distillation for Training Lightweight RNN-T | IEEE Journals & Magazine | IEEE Xplore