A robust feature selection method based on meta-heuristic optimization for speech emotion recognition

Most of the traditional feature selection methods do not show effective performance on speech emotion recognition systems. One of the recent advances in the feature selection is using meta-heuristic optimization algorithms. Individually, each algorithm plays a key role in many speech processing based applications. However, hybrid meta-heuristic are one most impressive technique in the field of feature selection and optimization problems. Hence, in this paper proposed a new robust hybrid-meta-heuristic feature selection model named as CSEO FS model as to improve the accuracy of SER task and also reduce the burden of computing capability. In this study, we investigated the performance of the proposed approach using speech-based emotional data-sets such as EMoDB and RAVDESS, which are primarily used in the development of human-computer interaction systems. The experimental results confirm the superiority of the proposed feature selection in terms of classification accuracy, precision, f1-score and number of selected features. Compared to state-of-the-art feature selection methods for SER systems, our experimental results show us achieving 94.35% and 96.78% high-level emotion recognition rates, respectively.