A difficulty classification method of musical score in piano teaching for new media

Abstract

The existing method of music difficulty classification depends on the manual classify by people which is inefficient. In piano teaching, we need to exercise a lot of music to pass the exams. For different ranking exams, we must exercise piano music with corresponding difficulty. To solve the problem mentioned, we propose a difficulty classification model for music on the basis of support vector machine (SVM) to assist piano teaching for new media. First, we regard the difficulty ranking of music as the classification with SVM. Secondly, we improve the Gaussian Radial Basis function with the features’ contribution which can obviously determine the difficulty. Finally, we propose a measuring learning SVM (ML-SVM) by the characteristics of piano music, such as strong subjectivity and universal correlation between features. We conduct the experiments on the music score datasets with various difficulties. The result can demonstrate that our ML-SVM outperforms to others based on SVM, such as logistic regression, linear kernel function, polynomial kernel function and Gaussian radial basis function. The performance of our method can reach 84.67% in the term of accuracy. Our method can effectively improve the piano teaching for new media.