Speech Emotion Recognition using Machine Learning Algorithms

SER is gaining a lot of popularity in affective computing, Due to its advanced techniques and having numerous applications. It is important to note that while having a conversation, people express additional feelings and information that adds value to their speech. With this in mind, three steps of SER are identified: data capture, feature selection process, and classification of data based on emotional parameters. Each of these steps can be performed through the use of ML techniques with ease. There has been an abundance of effort to capture the ML methods deployed for SER, and almost none attempt to explain the methods and techniques of these procedures, or the state-of-the-art solutions that these problems. One of the primary challenges are the low rates of accuracy in classification in the speaker independent tests. This work aims to analyze the literature on the subject of SER from the viewpoint of machine learning during the past ten years in terms of methods, challenges, and progress in the three steps of its implementations.