Musical Style Replication Using Apprenticeship Learning

Computer-based analysis of tonal music has been an active area of research for over a decade. In particular, machine learning based methods have been applied to composing music and solving various problems in musicology such as classification, visualization, search and stylistic analysis. Building a model of a composer’s style is a challenging problem of great interest in the Music Information Retrieval and Musicology research community. Interesting applications include style characterization tools for the musicologist, generation of stylistic metadata for intelligent retrieval in musical databases, music generation for Web and game applications, machine improvisation with or without interaction with human performers, and computer-assisted composition [4]. Simply generating music by computer is a field of active research. However, in our research, we have been unable find a specific instance of the AL algorithm. Markov chains are used for stochastic composition, but the probabilities are usually given as inputs to the algorithms, rather than learned. Learning techniques often focus on artificial neural networks or genetic programming techniques, although the MUSE system by Schwanauer utilizes an algorithm to learn voice leading rules. David Cope’s work presented in [3] combines grammatical generation system with a rule-based approach to compose music in a composer’s style. The novelty of our approach lies in the Machine Learning method chosen. This method models the composer’s inclinations and learns from musical pieces without incorporating rules or making assumptions about musical structure.