Application of Knowledge Distillation to Multi-task Speech Representation Learning