Music composition using recurrent neural networks and evolutionary algorithms

Pelletier, Calvin

Permalink

https://hdl.handle.net/2142/99544 Copy

Description

Title

Music composition using recurrent neural networks and evolutionary algorithms

Author(s)

Pelletier, Calvin

Issue Date

2017-12

Keyword(s)

• artificial intelligence
• music composition
• machine learning
• artificial neural network
• evolutionary algorithm

Abstract

The ability to generate original music is a significant milestone for the application of artificial intelligence in creative fields. This paper explores two techniques for autonomous music composition: recurrent neural networks and evolutionary algorithms. Both methods utilize data in the Nottingham Music Database of folk songs in ABC music notation. The neural network inputted and outputted individual ASCII characters and quickly learned to generate valid ABC notation from training on the dataset. The fitness function for the evolutionary algorithm was evaluated by comparing various characteristics of the generated song to target characteristics calculated using the dataset. Both methods are capable of composing homophonic music consisting of a monophonic melody and a chord progression

Type of Resource

text

Language

en

Permalink

http://hdl.handle.net/2142/99544

Owning Collections