Evaluation of Different Feature Extraction Techniques for Continuous Speech Recognition

摘要

Extracting human's voice feature is the most important process in any speech recognition system. There are many feature extraction techniques which are already used such as MFCC, LPC and ZCPA; but still have some problems especially in the continuous speech. It is important to evaluate different feature extraction techniques for continuous speech by making a comparison between these techniques as a trial to find the most suitable technique for speech recognition process, and trying to enhance the result by using PCA. Using PCA gives great better results especially for ZCPA technique as a comparison to other techniques.