Arabic speech recognition using phase parameter

Abstract

Abstract In this paper, the main idea is to investigate the ability of the information presented in the speech signal phase in contributing in the speech recognition system. The database that we used is Arabic words (20 words) pronounced by 10 speakers in different times. The system adopts single frequency signals strategy by using filter bank in feature extraction stage. Gaussian model is used as a classifier. The results shows that will be absolutely error to ignore the phase parameter, especially in some system like MFCC and PLP models, which used only spectrum information. No comparisons are made; the paper is just examining the phase parameter under different circumstance, which include different Filterbank channels and different database styles (clean and noisy utterance).