Knowledge Distillation for Neural Transducers from Large Self-Supervised Pre-Trained Models