Chinese folk instruments classification via statistical features and sparse-based representation

Compared with automatic classification according to western musical instrument, research on Chinese folk instruments music, which is an indispensable part of the world music, is very rare. As Chinese folk instruments have their own particular timbre, in this paper statistical features and a brand-new classifier-SRC in feature space were proposed to improve classification accuracy. With a large database consisting of 14 kinds of Chinese traditional instrument music, we extracted three sets of feature vectors and achieved sparse representation of categories via L¹ norm minimization on feature space to classify Chinese folk instruments automatically. The experiment results demonstrate that the statistical features can improve the accuracy with SVM or SRC significantly and SRC is a more effective classifier than SVM. With the combination of the three feature sets, SRC achieves the best accuracy of 99.8%. Compared with SVM, SRC achieves 4.5% improvement.