Emotion Recognition System for Specially Needed People with Optimized Deep Learning Algorithm

In the past few years, emotion recognition has emerged as one of the attractive areas in the field of signal processing. This paper presents the emotion recognition system for specially needed people. Initially, the input speech signal is read and subjected to the preprocessing to further improve the signal. In the second step, the features are extracted using frequencybased features, such as spectral flux, spectral centroid, spectral crest, and spectral roll-off. Finally, the emotions are classified using deep belief networks (DBN), which is trained by moth Search Optimization Algorithm (MSA) along with the standard gradient descent algorithm (SGD).The performance of the proposed method of emotion recognition is analyzed using evaluation measures, such as False Rejection Rate (FRR),accuracy, and False Acceptance Rate (FAR). The effectiveness of the proposed method is revealed by comparing the performance with the existing methods. From the analysis, it is depicted that the proposed method outperforms the existing models with a maximal accuracy of 98.5%, minimum FAR and FRR values of 0.63 % and 0.77 %, respectively.