Speech emotion recognition: Application in distance learning education

Affective computing is a branch of artificial intelligence that tries to pass the innate human capabilities of emotional intelligence to machines to enhance a smooth interaction between humans and computer systems. Speech emotion recognition is an essential aspect of affective computing and plays a significant role in designing systems and machines that recognize, analyze, interpret, and simulate human emotional states. In this project, the concept of speech emotion recognition is being integrated into a distance learning system to investigate the lecture delivering performance via a distance learning platform. To archive this, the performance is classified into three categories: interesting, Neutral, and boring, depending on the lecturer's emotional state. The project is implemented using intelligent machine learning techniques to recognize and interpret a lecturer's emotional state. It is carried out in four major stages of data preprocessing, feature extraction, feature selection, and classification. Our model adopts a comprehensive feature generation approach that utilizes a shoelace graph pattern as a local feature generator alongside tunable Q wavelet transform (TQWT). The best four feature vectors from the feature generation stage are selected and merged to obtain the final feature vector. After that, we applied an NCA method at the feature selection stage to select 512 most discriminative features. We then perform the classification using the SVM classifier. Our proposed network performed well, giving us an accuracy of 96.41% when applied on the Turkish dataset and 94.97% when applied on the English dataset.