Improvement of Arab Digits Recognition Rate Based in the Parameters Choice

Automatic speech recognition (ASR) is the process of automatically recognizing the speech on the basis of information obtained by acoustic features extracted from the speech signal. Because features extraction is the first component in ASR systems, the quality of the later component depends from the quality of feature extractor. The goal of this work is to study and implement features (representations) extraction, which are robust to the differences between the acoustic conditions of training and evolution. These features will be evaluated in an Automatic Arab digits recognition system. A particular attention will be taken to the robust features extraction methods (CMS, CGN, RASTAPLP, MBLPCC, and LPC MFCC).