Abstract

Algorithmic music composition is a popular area of research in computer aided music; it is the application of computer algorithms to create music. One subtopic of this area is the automated creation of pleasing melodies. This paper explores two machine learning approaches to algorithmically creating melodies: Markov chains and genetic algorithms. Markov chains can generate a rough imitation of the corpus of music upon which they are based. Though Markov chains can generate a rough approximation of a musical style, we want to refine the music to bring it closer to the desired target style. To accomplish this we use genetic algorithms to take an initial population of melodies and tweak and remix them to create better melodies. Genetic algorithms have been used extensively in previous research, but this paper proposes the use of a long short-term memory artificial neural network to act as a surrogate fitness function, rather than defining a set of rules for good music and penalties for breaking those rules

Advisor

Sommer, Nathan

Second Advisor

Fox, Nathan

Department

Computer Science; Mathematics

Disciplines

Artificial Intelligence and Robotics | Composition | Software Engineering

Keywords

algorithmic composition, music and computing

Publication Date

2018

Degree Granted

Bachelor of Arts

Document Type

Senior Independent Study Thesis

Download

Share

COinS
 

© Copyright 2018 Thomas Matlak