A Pop Music Style Recognition and Classification Approach Based on DBN

Abstract

In order to study the parameter selection and performance of deep belief network(DBN) in the classification of popular music, this paper proposes a music style recognition and classification algorithm based on feature selection and DBN. Firstly, the scheme preprocesses the music signal to obtain Mel's spectrum, and proposes the basic structure of music analysis by machine learning. Then, the conditional activation probability formula of feature fusion is redefined for the defect chapter of the restricted Boltzmann machine, and its learning algorithm is improved. Finally, the accuracy of classification is tested on FMA dataset. The simulation results show that the music characteristics of this paper have better classification effect than MFCC coefficient, and the training time is greatly reduced.