RECOGNITION OF ARABIC PHONETIC FEATURES USING NEURAL NETWORKS AND KNOWLEDGE-BASED SYSTEM: A COMPARATIVE STUDY

In this paper, we are concerned with the automatic recognition of Arabic phonetic macro-classes and complex phonemes by multi-layer sub-neural-networks (SNN) and knowledge-based system (SARPH). Our interest goes to the particularities of the Arabic language such as geminate and emphatic consonants and the vowel duration. These particularities are unanimously considered as the main root of failure of Automatic Speech Recognition (ASR) systems dedicated to standard Arabic. The purely automatic method constituted by the SNNs is confronted to an approach based on the user phonetic knowledge expressed by SARPH rules. For the acoustical analysis of speech as well as for the segmentation task, auditory models have been used. The ability of systems has been tested in experiments using stimuli uttered by 6 native Algerian speakers. The results show that SNNs achieved well in pure identification while in the case of semantically relevant duration the knowledge-based system performs better.