MLBytes Workshop: Classical Music Composition Using State Space Models

Algorithmic composition of music has a long history and with the development of powerful deep learning methods, there has recently been increased interest in exploring algorithms and models to create art. We explore the utility of state space models, in particular hidden Markov models, in composing classical piano pieces. We find that the state space models are fairly successful at generating new pieces that have largely consonant harmonies, especially when trained on original pieces with simple harmonic structure. However, we conclude that the major limitation in using these models to generate realistic sounding music is the lack of melodic progression in the composed pieces. Anna Yanchenko is a PhD student in the Department of Statistical Science at Duke. She is interested in machine learning for time series applications, especially with applications to modeling music. She currently works with Professors Sayan Mukherjee and Peter Hoff on algorithmic composition and hierarchical audio modeling. Anna holds an M.S. in Statistical Science from Duke and a B.S. in Physics from the University of Virginia. Prior to starting the PhD, Anna worked at MIT Lincoln Laboratory.