Rule mining for local boundary detection in melodies

The task of melodic segmentation is a long-standing MIR task that has not yet been solved. In this paper, a rule mining algorithm is employed to find rule sets that classify notes within their local context as phrase boundaries. Both the discovered rule set and a Random Forest Classifier trained on the same data set outperform previous methods on the task of melodic segmentation of melodies from the Essen Folk Song Collection, the Meertens Tune Collections, and the set of Bach Chorales. By inspecting the rules, some important clues are revealed about what constitutes a melodic phrase boundary, notably a prevalence of rhythm features over pitch features.