Evolutionary Composition Using Music Theory and Charts

With the development of human science and technology, applications of computer are more and more comprehensive. Using artificial intelligence (AI) to drawing, thinking, and problem solving becomes a significant topic. Recently, research on automatic composition using AI technology and especially evolutionary algorithms is blooming and has received promising results. A common issue at the current evolutionary composition systems is their requirement for subjective feedback of human sensation as evaluation criterion, which is vulnerable to the fatigue and decreased sensitivity after long-time listening. This paper proposes using music theory with the information from music charts in the evaluation criterion to address this issue. Specifically, we generate the weighted rules based on music theory for the fitness function. The weights are determined according to the download numbers from music charts. These weights obtained can interpret the music style and render an objective measure of compositions. Experimental results show that the proposed method can effectively achieve satisfactory compositions.