Abstract

We present data-driven methods for supporting musical creativity by capturing the statistics of a musical database. Specifically, we introduce a system that supports users in exploring the high-dimensional space of musical chord sequences by parameterizing the variation among chord sequences in popular music. We provide a novel user interface that exposes these learned parameters as control axes, and we propose two automatic approaches for defining these axes. One approach is based on a novel clustering procedure, the other on principal components analysis. A user study compares our approaches for defining control axes both to each other and to an approach based on manually-assigned genre labels. Results show that our automatic methods for defining control axes provide a subjectively better user experience than axes based on manual genre labeling.