1D CNN based approach for speech emotion recognition using MFCC features

Abstract

Identifying the emotions from speech is a challenging work for a machine learning (ML) algorithm, that show a decisive appearance in the area of speech emotion recognition (SER). SER gives an important appearance in many real-life applications such as human-robot interaction, human behavior assessment, in a call center, and many more. In this field, researchers focused on either hand-crafted classifiers or deep learning approaches used to increase the recognition rate. In this work, our target to contribution i) reduce the computation cost of the SER model and ii) improve the average accuracy of the SER model than the state-of-the-art. To achieve the above target, we proposed a novel approach for SER. In which Melfrequency cepstral coefficients (MFCC) are used to extract the features from audio file and apply 1D convolutional neural network (1D CNN) to recognize the emotions. The proposed approach is evaluated on popular public speech corpora Ryerson audio-visual database of emotion (RAVDESS). And average accuracy (82.93%) was reported better as compared to the existing SER model with reduced computation cost.