Automatic Han Chinese folk song classification using the musical feature density map
Suisin Khoo, Zhihong Man, Zhenwei Cao · 2012
Automatic music classification has received increased attention during the past decade. A system employing artificial neural network (ANN) techniques for the classification of Han Chinese folk songs is presented in this paper. Melodies of Han Chinese folk songs are machine-classified according to the different geographical region of the folk song's origin. Both audio and symbolic representations of music are studied. A novel encoding method called musical feature density map (MFDMap) is proposed to encode the symbolic musical features extracted from each folk song for machine classification. Our simulations demonstrate that the regularized extreme learning machine (R-ELM) classifier can achieve 72% classification accuracy using the MFDMap with three of the four suggested symbolic features.