The Study of the Classification of Chinese Folk Songs by Regional Style

This paper discusses a method of studying the region style classification of Chinese folk songs with support vector machine (SVM). According to geographical region of China, We have classified Chinese folk songs into 10 major categories, and used 500 Chinese folk songs in our experiment. 74 features have been extracted from audio files of the songs, and classified by an audio classifier on SVM. The experiment results show that sampling rate is not directly proportional to classification accuracy; SVM without feature selection is a very effective classification method for region style classification; the combination of 13-dimension MFCC and 10-dimension LPC features can achieve very similar results as that gained from SVM without feature selection. By using 30-second multi-clip classification and post-processing on classification result, the classification accuracy is improved from 47.4% to 75.2%, which is higher than that professional people got on music clip.