Classical Music Composition Using Hidden Markov Models

Abstract

Hidden Markov Models are a widely used class of probabilistic models for sequential data that have found particular success in areas such as speech recognition. 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. To this end, we explore the utility of Hidden Markov Models in composing classical music. Specifically, we train various Hidden Markov Models on piano pieces from the Romantic era and consider the models' ability to generate new pieces that sound like they were composed by a human. We evaluate the compositions based on several quantitative metrics that measure the originality, harmonic qualities and temporal structure of the generated piece. We additionally conduct listening evaluations with listeners of varying levels of musical background to assess the generated musical pieces. We find that Hidden Markov 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 Hidden Markov Models to generate music that sounds like it was composed by a human is the lack of global structure and melodic progression in the composed pieces.