This paper proposes an emotion detection method using a combination of dimensional approach and categorical approach. Thayer's model is divided into discrete emotion sections based on the level of arousal and valence. The main objective of the method is to increase the number of detected emotions which is used for emotion visualization. To evaluate the suggested method, we conducted various experiments with supervised learning and feature selection strategies. We collected 300 music clips with emotions annotated by music experts. Two feature sets are employed to create two training models for arousal and valence dimensions of Thayer's model. Finally, 36 music emotions are detected by proposed method. The results showed that the suggested algorithm achieved the highest accuracy when using RandomForest classifier with 70% and 57.3% for arousal and valence, respectively. These rates are better than previous studies.

This thesis proposes the emotions acquired after listening to the music as an emotion model composed of 8 types of emotions, based on the emotion model studied previously. The 5 musical factors selected, that affect the emotion, are tempo, dynamics, amplitude change, brightness, and noise. According to the emotion model composed of 8 types of emotions, 160 songs categorized into the 8 types of emotions were selected, and the actual data was extracted and analyzed.Through the analysis of actual data, an emotion equation made of weighted value of 5 factors was derived, and an algorithm that can predict the emotion by mapping on the 2-dimensional emotion coordinate system through the emotion equation was designed. Also, a way of controlling emotion by moving the coordinates on the 2-dimensional emotion coordinate system was suggested.■ keyword :|Music Emotion|Control Emotion|Calculation Emotion|

No abstract

In this paper, we propose a visualization method of a music emotion on LED wall. Emotion in music is recognized by a suggested algorithm using a dimensional approach.