Robust Analysis and Weighting on MFCC Components for Speech Recognition and Speaker Identification