Multivariate Analysis and Kernel Methods for Music Data Analysis

There is an increasing interest in customizable methods for organizing music collections. Relevant music characterization can be obtained from short-time features, but it is not obvious how to combine them to get useful i nformation. First, the relevant information might not be evident at the short-t ime level, and these features have to be combined at a larger temporal level into a ew feature vector in order to capture the relevant information. Second, we nee d to learn a model for the new features that generalizes well to new data. In thi s contribution, we will study how multivariate analysis (MVA) and kernel metho ds can be of great help in this task. More precisely, we will present two modifie d versions of a MVA method known as Orthonormalized Partial Least Squares (OPL S), one of them being a kernel extension, that are well-suited for discover ing relevant dynamics in large music collections. The performance of both schemes wi ll be llustrated in a music genre classification task.