Speech Emotion Recognition Using MFCC and Wide Residual Network

Emotion recognition from speech has been a topic of research from many years due to its importance in human-computer interaction. While a lot of work has been done upon recognizing emotions through facial expressions, recognition of emotions through speech is still a challenging task in Machine Learning due to the obscure knowledge about the effectiveness of different speech features. In this work, Mel-frequency cepstral coefficients (MFCCs) has been used as a feature extractor for speech files. Further, classification of speech signals has been done using Convolution Neural Network (CNN) in the form of Wide Residual Network (WRN) followed by a Dense Neural Network (DNN). To train and test this approach we used Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) and Toronto Emotional Speech Set (TESS) databases together. Results show that the proposed approach is gives an accuracy of 90.09% in recognizing emotions from speech into 8 categories.