Emotion recognition through voice by MLP classifier and Deep Sequential Neural Network

This thesis aspires to provide a thorough study on Speech Emotion Recognition in the field of Machine Learning. The main objective is to simplify the path towards emotion recognition by voice without sacrificing efficiency. In other words, the presented thesis studies the limits of least computationally complex algorithms in SER. By the end of this study, a traditional method which is MLP classifier and a Deep Learning approach with Deep Sequential Neural Network are compared. The algorithms use RAVDESS for training. Different combinations of three features, MFCC, Mel-spectrogram, and Chroma are utilized in both algorithms to determine the most efficient combination.