Accurate and Structured Pruning for Efficient Automatic Speech Recognition