Structured Pruning of Self-Supervised Pre-trained Models for Speech Recognition and Understanding