Classical Music Specific Mood Automatic Recognition Model Proposal

Abstract

The purpose of this study was to propose an effective model for recognizing the detailed mood of classical music. First, in this study, the subject classical music was segmented via MFCC analysis by tone, which is one of the acoustic features. Short segments of 5 s or under, which are not easy to use in mood recognition or service, were merged with the preceding or rear segment using an algorithm. In addition, 18 adjective classes that can be used as representative moods of classical music were defined. Finally, after analyzing 19 kinds of acoustic features of classical music segments using XGBoost, a model was proposed that can automatically recognize the music mood through learning. The XGBoost algorithm that is proposed in this study, which uses the automatic music segmentation method according to the characteristics of tone and mood using acoustic features, was evaluated and shown to improve the performance of mood recognition. The result of this study will be used as a basis for the production of an affect convergence platform service where the mood is fused with similar visual media when listening to classical music by recognizing the mood of the detailed section.