Compiling Hierarchical Dependency Graph for Large- Span Musical Expressive Feature Analysis Using Multi- Scaling Probabilistic Graphical Models

Ren Gang, Xuchen Yang, Zhe Wen, Dave Headlam, Mark F. Bocko Dept. of Electrical and Computer Engineering, Edmund A. Hajim School of Engineering and Applied Sciences, Univ. of Rochester Dept. of Music Theory, Eastman School of Music, Univ. of Rochester Rochester, NY 14627, USA [email protected], [email protected],[email protected],[email protected], [email protected] Abstract—Music performance conveys profound music understanding and artistic expression in musical sound. These performance-related dimensions can be extracted from audio and encoded as musical expressive features, which is based on a highdimensional sequential data structure. In this paper we propose a structure learning based method using probabilistic graphical models that obtains a hierarchical dependency graph from musical expressive features. The hierarchical dependency graph we proposed serves as an intuitive visualization interface of the internal dependency patterns within feature data series and helps music scholars identify in-depthconceptual structures.