Moving From Traditional Machine Learning to Deep Learning for Speech Based Emotion Detection on Different Datasets

Speech Emotion Recognition is a way to predict emotions of humans from their speech. SER technology has several applications in different fields such as in medical field to find patient’s emotional state for better diagnosis, in education, entertainment as well as better human computer interaction (HCI). For a better SER several researches are going on since last two decades and different approaches are presented by researchers to solve the task. Traditional machine learning models were used for SER by different researchers but now with the advancement of neural networks, the deep learning approaches are to be proven best for predicting emotions from speech signals. Therefore, we compare existing approaches and databases in Speech Emotion Recognition (SER) in order to arrive at workable solutions with more solid understanding of this open-ended problem. This is due to the development of neural networks, the constant requirement for precise and almost real-time SER in human computer interactions. This research gives the complete platform for an SER framework from machine learning models to deep learning models and the available datasets for SER.