Speech Audio Analytics based Classification of Human Emotions using Machine Learning and Deep Learning Models

Speech emotion recognition is a significant area of research with applications in human-computer interaction, affective computing, and psychological studies. This paper presents a comprehensive investigation into speech emotion recognition using the RAVDESS dataset and various machine learning and deep learning models. Our study focuses on data preprocessing, feature extraction, model implementation, and performance evaluation. The Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDEES) dataset is used for performing training and testing for the custom model and in order to improve the generalization of the model, noise and audio stretching are explicitly added. Specifically, the effectiveness of decision trees, K-nearest neighbors (KNN), multilayer perceptron (MLP) classifiers, recurrent neural networks (RNNs) including long short-term memory (LSTM) and gated recurrent units (GRU), and convolutional neural networks (CNNs) is measured for classifying emotions from audio data. Through a series of experiments and analysis, the most effective model is identified based on the insights of its performance.