Arabic corpus implementation: Application to speech recognition

This paper presents an Arabic speech recognition system using different features such as MFCC, PLP, LPC We propose to handle noisy data in order to evaluate the robustness of our system under noise conditions. The paper explores, also, the mechanism of building a corpus of the Arabic Tunisian language. Thus, we present the implementation of an automatic speech recognition system under the HTK platform. Experimental tests are performed to prove the effectiveness of different features for Arabic speech samples recognition. The performances on noisy samples from Arabic Tunisian and TIMIT databases show the robustness of the proposed system especially with noisy environments.