An approach to Arabic speech recognition based on improving the features extraction

Abstract EN

The aim of this paper is to perform an Arabic speech recognition system based on Arabic words.

For increasing the efficiency of the identification, this paper shows the advantage of using barkcepstrum over Fourier Transform-filter bank (FT-filter bank), which presently used parameters for Automatic Speech Recognition (ASR).

Thus, comparing the two methods has been achieved for independent Arabic speaker based on the linear decision function.

Experimental results obtained with many speakers have shown that the use of a bark-cepstrum, with an appropriate factor, improves the identification of Arabic speaker.