Automatic Analysis and Musicological Interpretation of Human Free Sorting of Musical Excerpts

Most content-based music recommendation systems are relying on audio features which do not always match with musicological criteria. This paper describes the experimental protocol and the results of a sorting experiment, which leads to an 'average categorization' by volunteers. An automatic analysis afterward aims at identifying relevant acoustic parameters based on the obtained categories and sub-categories. A musicological analysis was also done in parallel.