DefinePK

Abstract

Speech Emotion Recognition (SER) is gaining significant attention in the field of human-computer interaction (HCI) over past decade. Specially in the fields like health, security, communication, and entertainment. But due to the lack of research on how to boost the speech processing efficiency, the current emotion recognition systems need improvement and more accuracy. To enhance the accuracy, we proposed an Effective Speech Emotion Recognition System (ESERS) which is a hybrid approach that uses Autoencoders (AEs) for denoising and robust feature extraction with a Self-Attentional Convolutional Neural Network–Bidirectional Long Short-Term Memory (CNN-BLSTM) architecture for effective temporal and contextual modeling. Using CREMA Dataset, we achieved Weighted Accuracy (WA) improved from 73.9% to 81.6% and Unweighted Accuracy (UA) increased from 68.5% to 82.8%. which shows absolute improvement of 7.7% and 14.3%, and relative improvements of 10.4% and 20.9% respectively. Hence, to enhance system efficiency, the hybrid approach outperforms traditional approaches currently in use.