Towards The Development of Computer Aided Speech Therapy Tool in Arabic Language Using Artificial Intelligence

Abstract

Speech disorders affecting children make their speech unclear, leading to ineffective communication and subsequent psychological problems. Articulation disorders are the most prevalent type; they are divided into four categories: substitution, omission, addition, and distortion. Interactive automatic speech monitoring tools offer a practical, adaptive and cost-effective alternative to face-to-face intervention sessions for speech therapy, however none do exist for Arabic language. Arabic is a complicated language with a variety of different dialects. It consists of 27 consonant phonemes with a wide range of places of articulation that span the whole vocal tract from lips to glottis, in addition to eight vowel phonemes. This dissertation is a step closer to an automatic Arabic speech therapy tool. It mainly investigates the possible features suitable for the diagnosis phase and the corresponding classification accuracy It describes the process of detecting Arabic speech articular disorder using the dataset found online of the letter (ر/r/). The feature extraction method used for is a widely known for its performance in different speech analysis application; Mel-frequency Cepstrum Coefficients using Matlab application. For the final decision, the classification phase, three models were examined: Support Vector Machine (SVM), Artificial Neural Network (ANN). and K-nearest neighbour (KNN) using Weka software. The system attained overall accuracies of 85.25%, 64.18% and 74.99% for KNN, SVM and ANN respectively.