Song structure identification of Javanese gamelan music based on analysis of periodicity distribution

In a song played by multiple instruments, there is distribution of periodicities that comes from different playing patterns among groups of instruments. We propose a visualization of this distribution for analyzing song structure of Javanese gamelan music. A predefined number of periodicities along with their confidence levels are obtained using comb filter resonator. The filter is applied to the auto-correlation function of overlapping analysis frames of the musical track. We cluster the distribution based on the proximity of two parameters, which are periodicity and confidence level. In this way, we assume that each cluster center represents the periodicity of a group of instruments. We observe four features of the visualization, namely the width and the average height of periodicity distribution, the pattern of dominant periodicities, and the fluctuation of the most dominant periodicity. Those features implicitly give us information regarding the strength applied to the notes, the estimated number of instruments, and the accent of song according to those features, from which we make an inference about the structure. We provide the experiment with a database of thirty Javanese gamelan songs and compare the analysis of lancaran, ladrang, and ketawang song structures. The results show that using this method, lancaran received the highest performance, which is 0.94 F-measure, followed by ketawang and ladrang with F-measure of 0.90 and 0.75 respectively.