Automatic Music Composition with Evolutionary Method

Mr. Lo Man Yat (University of Essex)

myloh@essex.ac.uk

Introduction

The overall research goal is to develop a system that is capable of automatically generating music using an evolutionary algorithm. Conventionally, this has been done by evolving music sequences whose fitness is then evaluated using a human listener. However, this human-in-the-loop approach is very slow. The idea behind the work is to replace the human by a music evaluation algorithm, which can be trained on existing music samples- for example, as created by classical composers such as Beethoven. Such a system could provide a fast, cheap music composition tool.

The aim of the research is to provide a solid foundation for future experimentation. One of the key ideas is to develop a trainable music evaluation system with appropriate discriminative power. There are several objective tests that will be carried out, which will measure the degree to which the system is able to distinguish:

E.g.

• Classical music from random music
• Classical music from pop music
• Mozart from Beethoven

Research questions

• How should the music be represented as input to the system (polyphony, note duration)?
• What form of model should the music evaluation algorithm be based on?
• What representation should the evolutionary algorithm operate in?
• What evolutionary variation operators should be used?
• How should the success of the system be evaluated?

Success criteria

• Demonstration of how the music evaluation discrimination power varies with the major setup parameters.
• Demonstration of how the configuration of the EA affects the ability of the system to converge to highly fit music.
• Demonstration of the importance of music representations in affecting the N-gram model.

Papers

Man Yat Lo and Simon M. Lucas , Evolving Musical Sequences with N-Gram Based Trainable Fitness Functions, IEEE Congress on Evolutionary Computation (2006). (pdf)

Listening tests

A small set of listening tests carried out in order to get some idea of how the evolved music sounds to human listeners.Basically, I need your help !! Please rate the follow music melody. Listen each of the melody and rate it. ( 1 is the worst, 10 is the best).