Hybrid Differential Evolution and Particle Swarm Optimization for Speech Emotion Classification

This paper presents a novel evolutionary-based hybridization approach for speech emotion classification. This approach is based on two optimization algorithms: differential evolution (DE) and particle swarm optimization (PSO), using three types of acoustics features to represent and explore a ‘tri-population’ environment. Mel-frequency cepstral coefficients (MFCCs), prosodic features, and auditory-based parameters are considered to compose each population, respectively. The hybridization of PSO and DE methods, as well as the relevance of acoustic representation techniques, are investigated to provide the most effective configuration for speech emotion classification. Evaluation experiments are carried out using the Emotional Prosody Speech and Transcripts, and the support vector machine (SVM) as an objective function. The results showed that the PSO-DE-PSO hybrid method achieves the best performance.