Evolutionary music composition system with statistically modeled criteria

摘要

The paper concerns an original evolutionary music composition system. On the basis of available solutions, we have selected a finite set of music features which appear to have a key impact on the quality of composed musical phrases. Evaluation criteria have been divided into rule-based and statistical sub-sets. Elements of the cost function are modeled using a Gaussian distribution defined by the expected value and variance obtained from an analysis of recognized music pieces. An evolutionary algorithm, considering a reference sequence of chords as an input, is created, implemented and tested. The results of a sampling survey (poll) proves that the melodies generated by the system arouse the interest of a listener.