New Trends in Speech Emotion Recognition

Abstract

In this study, sound energy and characteristics of sound were investigated. Then, emotion recognition models built upon sound data in the literature were reviewed. Speech emotion recognition studies which adopt the most suitable machine-learning algorithms making feature extraction using both acoustic analysis methods and spectrogram analysis methods were investigated. In light of these studies, implementation has been carried out using EMO-DB data. Speech emotion recognition is a difficult problem for machine learning. The analysis of a sound signal is difficult to make as it includes various frequencies and features. Speech is digitized using signal processing methods and then sound characteristics are obtained through acoustic analysis. However, the overall success rate changes as the changes in these characteristics differ according to the emotions (sadness, fear, anger, happiness, neutral, displeasure, etc.). Although different methods are utilized in both feature extraction and emotion recognition, the success rate varies according to emotions and databases.