Comparative Performance Analysis of Metaheuristic Feature Selection Methods for Speech Emotion Recognition

Abstract

Emotion recognition systems from speech signals are realized with the help of acoustic or spectral features. Acoustic analysis is the extraction of digital features from speech files using digital signal processing methods. Another method used is the analysis of time-frequency images of speech with image processing. The size of the features obtained by acoustic analysis is in the thousands. Therefore, classification complexity increases and causes variation in classification accuracy. With feature selection, features unrelated to emotions are extracted from the feature space and it is aimed to contribute to the classifier performance. Traditional methods of feature selection are mostly based on statistical analysis. Another feature selection method is to use metaheuristic algorithms to detect and remove irrelevant features from the feature set. In this study, we compare the performance of metaheuristic algorithms for feature selection for speech emotion recognition. For this purpose, a comparative analysis was performed with four different datasets, eight metaheuristics, and three different classifiers. According to the results of the analysis, an increase in classification accuracy was obtained while reducing the feature size. In all datasets, the highest accuracy was obtained with the support vector machine. The highest accuracy for EMO-DB, EMOVA, eNTERFACE’05 and SAVEE datasets is 88.1%, 73.8%, 73.3%, and 75.7% respectively.