New approach to classification of Chinese folk music based on extension of HMM

Recently, class labels are commonly used to structure the increasing amounts of music available in digital form on the Web and are important for music information retrieval. An evaluation for automatic classification of Chinese folk music according to an audio taxonomy is presented. The audio taxonomy is organized as hierarchical, resulting in good coverage of Chinese folk music. Continuous Hidden Markov Model(CHMM) have been widely used to model the temporal evolution of dynamic sounds, especially music signal, whereas with an obvious drawback that the probability of time spends in a particular state, or state occupancy is geometrically distributed, which is not the case in real music signal. In this paper, we presented two extensions of standard HMM: Hidden semi-Markov Model(HSMM), and Segmentation Duration-Based HMM(SDBHMM), providing a comparison among them and Continuous Hidden Markov Model(CHMM). The former extension has been presented in speech recognition and we proposed the later one originally. Our result show that SDBHMM could achieve classification accuracy of 92.49% approximately and HSMM with 90.02%, both of which outperform standard CHMM.