Compensation of Differences of Sampling Frequency or Front-end for Speech Recognition