Music selection is difficult without efficient organization based on metadata or tags, and one effective tag scheme is based on the emotion expressed by the music. However, manual annotation is labor intensive and unstable because the perception of music emotion varies from person to person. This paper presents an emotion classification system for digital music with a resolution of eight emotional classes. Russell’s emotion model was adopted as common ground for emotional annotation. The music information retrieval (MIR) toolbox was employed to extract acoustic features from audio files. The classification system utilized a supervised machine learning technique to recognize acoustic features and create predictive models. Four predictive models were proposed and compared. The models were composed by crossmatching two types of neural networks, i.e., Levenberg-Marquardt (LM) and resilient backpropagation (Rprop), with two types of structures: a traditional multiclass model and the cascaded structure of a binary-class model. The performance of each model was evaluated via the MediaEval Database for Emotional Analysis (DEAM) benchmark. The best result was achieved by the model trained with the cascaded Rprop neural network (accuracy of 89.5%). In addition, correlation coefficient analysis showed that timbre features were the most impactful for prediction. Our work offers an opportunity for a competitive advantage in music classification because only a few music providers currently tag music with emotional terms.
Keywords
Arti cial Neural Networks, Classi cation Algorithms, Emotion Recognition, Music Information Retrieval
ECTI TRANSACTIONS ON COMPUTER INFORMATION TECHNOLOGY