Classifying Arabic phonemes and extracting their attributes using deep learning in view of mispronunciation detection

Abstract

Mispronunciation Detecting is an important issue in computer-aided language learning (CALL) systems, where locating errors in pronunciation helps the language learner to obtain an accurate assessment of pronunciation correctness, these systems have received great attention because they give language learners the possibility to improve their language proficiency without the need to direct communication with language specialists, by making use of modern learning methods and advanced technologies. This research aims to find appropriate methodologies in building a system that helps the language learner to detect and correct pronunciation errors, By studying the possibility of categorizing phonemes and distinguishing them automatically, especially the similar and common ones in the sound output and some character traits, in addition to studying the possibility of distinguishing speech descriptors in order to use them in detecting pronunciation error and determining its type. The focus of the research was on the Arabic language, with the aim of reducing the research gap that exists between the linguistic technologies that support the Arabic language and its counterparts in international languages, which have achieved great progress in many fields.