In-depth motivic analysis based on multiparametric closed pattern and cyclic sequence mining

The paper describes a computational system for exhaustive but compact description of repeated motivic patterns in symbolic representations of music. The approach follows a method based on closed heterogeneous pattern mining in multiparametrical space with control of pattern cyclicity. This paper presents a much simpler description and justification of this general strategy, as well as significant simplifications of the model, in particular concerning the management of pattern cyclicity.
A new method for automated bundling of patterns belonging to same motivic or thematic classes is also presented.

The good performance of the method is shown through the analysis of a piece from the JKUPDD database. Ground-truth motives are detected, while additional relevant information completes the ground-truth musicological analysis.

The system, implemented in Matlab, is made publicly available as part of MiningSuite, a new open-source framework for audio and music analysis.