An Efficient Feature Selection Algorithm Based on Kullback-Leibler Divergence for Music Information Retrieval

In this paper, we propose an efficient feature selection algorithm for the music information retrieval based on the Kullback-Leibler divergence. In the proposed algorithm, we measure the separability of two classes using the Kullback-Leibler divergence between the corresponding Gaussian mixture models, and feature subset is chosen using the sequential forward selection method. We tested the proposed algorithm with ISMIR 2004 music database, which are categorized into six genres. 256 feature candidates were extracted for the experiments, which represent characteristics such as timbre, rhythm, pitch and so on. The classification was performed with the selected features using k-NN classifier. The experimental results are presented with regard to both conventional feature selection algorithms and the proposed algorithm.