Development of a computer–based composition method by analyzing and modifying scale-free network of music

Abstract

본 논문에서는 척도없는 네트워크의 성질을 기반으로 한 알고리즘 작곡방법을 제안하였다. 일반적으로 알고리즘 작곡은 확률론적인 방법이나 결정론적인 방법을 이용한다. 기존의 방법은 주어진 음표들을 변경하거나 추가하는 방법을 이용한다. 그러나 이러한 방법은 여러 번의 반복을 통해 적합한 선율을 찾아야만 한다. 전체 곡의 선율에 대해 고려하지 않은 채 변경되거나 추가되는 선율이 전체 곡에 부합하는지의 여부를 사람이 직접 판단해야 하기 때문이다. 이 논문에서 제안하는 새로운 방법은 기존 방법의 한계를 극복한다. 새롭게 작곡된 곡은 원래의 곡과 같은 네트워크 누적분포도의 성질을 보이며 유사한 멜로디 경향성을 보인다. 최적화 방법으로는 유전알고리즘을 사용하였다. 선율의 경향성을 모델화하기 위하여 선율 경향성 곡선법과 노드 연결 알고리즘을 개발하였다. 이 방법을 한국 동요인 ‘고향의 봄’에 적용하였으며 원곡의 음악적 분위기와 유사한 결과물을 얻을 수 있었다. 새 곡은 원곡의 네트워크 성질과 멜로디 경향성을 유지하면서도 원곡과는 다른 멜로디를 가진다. |This dissertation proposes a new approach to composition based on a scale-free network theory. In general, the algorithm uses deterministic and probabilistic methods of composition. Existing methods have been proposed to perform a composition by adding or changing the process defined by given melody notes. However, this method has the disadvantage that it must be designed to track through several trial and errors because it was made without taking into account the whole melody in the music. In this paper, we propose a new method for composing that will overcome the shortcomings of traditional methods. A Newly composed song has a distribution network similar to the original and the tendency of melody is also retained. The method for the optimization was performed using a Genetic Algorithm. We developed a method to track the tendency of the melody curve and node connection algorithm to overcome the limitations of the existing method. We applied the proposed method to the Korean song “Hometown Spring” and succeeded in producing the original song and musical atmosphere. New songs are designed differently than the original song, while maintaining the network properties and melodic tendencies of the original song.; This dissertation proposes a new approach to composition based on a scale-free network theory. In general, the algorithm uses deterministic and probabilistic methods of composition. Existing methods have been proposed to perform a composition by adding or changing the process defined by given melody notes. However, this method has the disadvantage that it must be designed to track through several trial and errors because it was made without taking into account the whole melody in the music. In this paper, we propose a new method for composing that will overcome the shortcomings of traditional methods. A Newly composed song has a distribution network similar to the original and the tendency of melody is also retained. The method for the optimization was performed using a Genetic Algorithm. We developed a method to track the tendency of the melody curve and node connection algorithm to overcome the limitations of the existing method. We applied the proposed method to the Korean song “Hometown Spring” and succeeded in producing the original song and musical atmosphere. New songs are designed differently than the original song, while maintaining the network properties and melodic tendencies of the original song.