A review on objective music structure analysis
This paper summarizes the applications and the state of the art of objective music structure analysis. Two principal types of methods, namely "state" and "sequence" approaches are reviewed after applications are presented. Two kinds of objective features, static and dynamic, and their related techniques are compared. The static features include Mel-frequency cepstral coefficients (MFCCs), chroma, Discrete Cosine Transform (DCT)-based chroma, Constant Q Transform (CQT), Constant Q Logarithmic Power Spectrum etc. Similarity measures, including Euclidean, cosine, Kullback-Leibler (KL) and Structure-based Distance are summarized. Evaluation measures, including F-measure, cluster purity value, value based on conditional entropies, edit distance are described. Concluding remarks and future directions are pointed out on the basis of this review. © 2009 IEEE.
