Marine Predatory Algorithm for Feature Selection in Speech Emotion Recognition

In recent times, the recognition of human emotional states expressed through speech communication through speech communication has garnered significant interest among researchers in human-computer interaction. Various systems have been advanced to categorize states of speech emotion using features extracted from spoken utterances. Feature extraction plays a vital role in developing speech emotion recognition systems, as the performance of the learning model improves when the extracted features are reliable and capture the emotional characteristics of speech samples. However, some of the extracted features may be redundant, irrelevant, or noisy, which can diminish the classification performance of speech emotion recognizers. To address this issue and select efficient and precise speech emotion features, this paper introduces an effective feature selection method called MPA-KNN, which combines the marine predators algorithm with the KNN classifier. The proposed method’s performance is evaluated using three distinct speech databases: the Surrey Audio-Visual Expressed Emotion (SAVEE), the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS), and the Arabic Emirati-accented speech database. The results showed that MPA-KNN improved the accuracy of speech emotion recognition compared to classical machine learning and optimization feature selection methods. Moreover, it overcomes several competitors in the literature that utilize the same datasets.