Summary
International Workshop on Smart Info-Media Systems in Asia
2023
Session Number:RS2
Session:
Number:RS2-1
Automatic Generation of A Cappella Scores by Deep Neural Networks Considering Music Structure
Yijun Long, Kaoru Arakawa,
pp.34-38
Publication Date:2023/8/31
Online ISSN:2188-5079
DOI:10.34385/proc.77.RS2-1
PDF download
Summary:
A method to generate a cappella-style chorus scores for a given main melody is proposed using deep neural networks. A cappella is a style of choral music which is composed only of human voices. People like to enjoy a cappella, but it is difficult for them without special knowledge on music to arrange an arbitrary song into an a cappella style. On the other hand, a method of generating chorus scores automatically for the input of main melody was proposed for Bach-style chorus using deep neural networks, such as DeepBach. DeepBach is suitable for Bach-style chorus, but not for a cappella, because a cappella has a unique structure that is composed of synchronous and asynchronous parts to the main melody. In this paper, a new method to generate chorus scores of a cappella is proposed by considering the structure of a cappella in the frameworks of DeepBach. Experimental results show that the generated chorus score is a cappella-like and preferable by subjective evaluation.