Deep Learning and Audio Based Emotion Recognition

While tightening and expansion of our facial muscles cause some changes called facial expressions as a reaction to the different kinds of emotional situations of our brain, similarly there are some physiological changes like tone, loudness, rhythm and intonation in our voice, too. These visual and auditory changes have a great importance for human-human interaction human-machine interaction and human-computer interaction as they include critical information about humans' emotional situations. Automatic emotion recognition systems are defined as systems that can analyze individual's emotional situation by using this distinctive information. In this study, an automatic emotion recognition system in which auditory information is analyzed and classified in order to recognize human emotions is proposed. In the study spectral features and MFCC coefficients which are commonly used for feature extraction from voice signals are firstly used, and then deep learning-based LSTM algorithm is used for classification. Suggested algorithm is evaluated by using three different audio data sets (SAVEE, RAVADES and RML).