基因演算法於音樂創作之應用: 藝術系統的建模及電腦模擬

Abstract

對於演算式作曲所使用的各種演算法(包括基因演算法)，其中某些雖然可以創作出品質不錯的樂曲，但是如何根據各種演算法所需，挑選出合適的函式及參數依然是一個難以解決的問題，這也是影響樂曲品質的重大因素之一。本篇論文將試著整合基因演算法和社會學的方法，提出一個模擬真實世界中藝術系統的模型，透過作曲方法及鑑賞方法的共同演化，自動調適其函式及參數。 本系統具備了下列的特色: 1. 不同演算法的作曲方法及鑑賞方法可以共存於本系統。 2. 本系統會產生類似真實世界中流行時尚演變的現象，方便使用者依據不同的時期及地理位置，挑選不同風格的樂曲。 3. 這個方法具有一般性，可以推廣到其他使用電腦進行創作或設計的問題上。 在本研究中將先使用視覺化的效果來分析此模型的性質，再將其應用到演算式作曲上。

Some algorithmic composition mechanisms (including using the Genetic Algorithm) can compose music with good qualities. However, it is in general difficult to choose appropriate functions and parameters based on the criteria of these algorithms. This is the major factor affecting the quality of the resulted music. In this thesis, we will combine the Genetic Algorithm method and some principles from Sociology to present a model that simulates art systems in real world. Through the co-evolution of the composing and evaluating algorithms, the functions and parameters used in these two algorithms can adapt automatically. The proposed system has the following characteristics: 1. Different composing and evaluating algorithms can coexist in this system. 2. The fashion phenomena in the real world can emerge from this system. It will help users select music with different styles based on different time periods and different geographical positions. 3. This method provides a general framework and can be applied to other problems that employ computers in creation or design. In this study, we will use visualization tools to analyze the properties of this model, and then apply the system to algorithmic composition.