Music emotion recognition based on a modified brain emotional learning model

Listening to music can evoke different emotions in humans. Music emotion recognition (MER) can predict a person’s emotions before listening to a song. However, there are three problems with MER studies. First, the brain is the seat of music perception, but the simulation of MER based on the brain’s limbic system has not been examined so far. Secondly, although the effect of individual differences is recognized on the perception and induction of music emotion in the literature, less attention has been paid to the personalization of the model. Finally, most previous studies have emphasized the classification of music pieces into emotional groups, while often a piece of music creates several emotions with different values. The purpose of the present study is to introduce an optimized model of brain emotional learning (BEL) which is combined with Thayer’s psychological model to predict the quantitative value of all emotions that hat would reach a specific person by listening to a new piece of music. The proposed model consists of 12 emotional parts that work in parallel where each part is responsible for evaluating one Thayer’s specific emotion. Four neural areas of the emotional brain are simulated for each part. The input signal is adjusted using Thayer’s dimensions and a fuzzy system. The average of the results obtained with the proposed model were: R2 = 0.69 for arousal, R2 = 0.36 for valence, and MSE = 0.051, which was better and faster than the multilayer network models and even the original BEL model for all emotions.