A geometric evolutionary search for melody composition

In this paper, we present a genetic algorithm using geometric crossover to preserve musically important properties. The length of the chromosome is variable when we use a musical note as a gene. A geometric crossover based on the edit distance was used to combine chromosomes of variable length and succeeded in creating a better child by using a model in which melody pattern is learned. The model can evaluate the score of melody in a specified chord. As a result, we could successfully create a melody preserving parent's good patterns.