Acoustical-similarity-based Musical Instrument Hierarchy and Its Application to Musical Instrument Identification

This paper describes a method of constructing a musical instrument hierarchy reflecting the similarity of acoustical features. Although this acoustical hierarchy is useful for various purposes as well as investigating the timbres of musical instruments, it has not been reported in the literature. The main issues in constructing such a hierarchy are what feature space is used and how to obtain the representative position of each instrument in the feature space. To solve the first issue, we use a feature space that facilitates identification of 6247 solo tones of 19 instruments with an accuracy of 80%. To solve the second issue, we approximate the distribution of each instrument using a large number of sounds. Experimental results using 6247 solo tones of 19 instruments show that the hierarchy obtained by our method is partially different from a conventional one. In addition, this paper reports category-level identification of non-registered instruments using our hierarchy.