Speech Emotion Recognition using MFCC, GFCC, Chromagram and RMSE features

In recent years, increasing attention is given to the research of the emotions present in speech. Various systems are developed aiming to detect the emotions in the speaker’s statements. One of the biggest differences between a machine and a human is understanding the emotions of others and behaving accordingly. Researchers are working on bridging this gap by recognizing emotions in speech or voice. This paper proposes a deep learning-based technique for speech emotion recognition (SER). The SER system is based on various techniques that use distinguished modules for emotion recognition. The model differentiates emotions such as neutral state, happiness, sadness, anger, surprise, etc. The performance of the classification system is based on features extracted and generated models. The features utilized in this include energy, pitch, chromagram, mel-frequency spectrum coefficients (MFCC), and Gammatone frequency spectrum coefficients (GFCC). The emotions are classified using a two dimentional Convolutional Neural Network (CNN). The classification model achieved an overall accuracy of 92.59% on the test data which is comparatively better than the previous algorithm. In future, the intention is to increase the system performance and detect more emotions.