Automatic composition system using genetic algorithm and N-gram model considering melody blocks

Abstract

In this paper, we propose an automatic composition system using genetic algorithm and N-gram model considering melody blocks such as verse, bridge and chorus. In the proposed system, the melody is divided into some blocks such as verse, bridge and chorus, some new melodies are generated using melody feature in each block. The feature on tone and length of sounds in each block of the sample melodies is trained in some N-gram models, and they are used in the calculation of the fitness in the melody generation by the genetic algorithm. The tone and length of sounds are expressed as genes, and some melodies similar to the trained sample melodies are generated by the genetic algorithm. In the proposed system, the fitness of the gene for each melody block is calculated based on (1) transition of rhythm, (2) transition of tone, (3) transition of tone and length, (4) transition of the number of sounds per bar, (5) similarity of rhythms between phrases, (6) rate of unique sounds of scales, (7) difference between two sounds and (8) length of rests. We carried out a series of computer experiments and confirmed that the some melodies reflecting the feature in each melody block of the trained sample melodies can be generated.