A Computer-Aided Speech Disorders Correction System for Arabic Language

In this paper, we propose a hybrid technique that integrates both the audio visual analysis techniques to automate speech disorders treatment for Arabic language that can be used in many developing countries. We proposed a technique that is based on audio visual analysis of the patient's speech. For patient's audio analysis, we used the mel frequency cepstrum coefficients (MFCC's) and linear predictive cepstrum coefficients (LPCC's) as the key features to classify the audio. In addition, we used visual features for the analysis of the patient's video based on the patient's appearance. Audio visual features techniques are combined for increasing the efficiency of our recognition system. We present a comparative evaluation for both audio and video features. We perform the features evaluation using Dynamic time warping (DTW) for speech features and histogram based approach analysis for visual feature. Finally, we used a neural network based classifier to differentiate between normal and abnormal speech. This research presents an expert system with an interactive computerized environment that has the ability to treat patients with speech disorders problems.