Detection, Classification, and Recombination of Musical Style Motifs
Daniel Muldrew

Background: A motif is a recurring structure found in systems of music, visual arts, DNA sequences, brain

networks. They are thought to be the fundamental building blocks of the system. In particular, a musical motif is a repeated sequence of notes, rhythms, or harmonies. For example, here is a classic example of a rhythmic motif:

Beethoven's 5^th symphony opening

This "short-short-short-long" motif appears throughout the piece.

Problem: The main problem is that I would like to identify the motifs present in different musical styles. I would also like to use statistical information such as motif frequency and position in the piece.

After that, I have two applications in mind:

(1) test whether motifs can be used as a basis for classification of a musical style

(2) attempt to recombine these motifs to compose new work of a particular style

Methods:

Part I: There is a pattern identification method detailed in this paper "Mining transposed motifs in music"

http://www.ugr.es/~miguelmolina/publications/jimenez-jiis11.pdf

There are also many potential techniques from DNA sequence analysis:

(1) http://www.cs.cmu.edu/~epxing/Class/10810-05/Lecture6.pdf

(2) motif clustering using SVM: http://math.bu.edu/people/mkon/O6.pdf

(3) http://users.cis.fiu.edu/~giri/papers/ICBA04_Sun.pdf

Part II: Use the motif structures and overall statistics to define a style feature set to use supervised learning algorithms.

Here is a paper evaluating some of the potential methods:

"Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms"

http://www.cs.iastate.edu/~honavar/dietterich98approximate.pdf

Part III: Use a genetic algorithm with an appropriate fitness function to attempt to create a new work of the style. I plan to use these two papers as starting points:

"Towards Melodic Extension Using Genetic Algorithms"

http://eprints.qut.edu.au/169/1/towsey.pdf

"Improving algorithmic music composition with machine learning"

http://homepages.inf.ed.ac.uk/s0786354/publications/icmpc06.pdf

Data Sets: I plan to analyze the information encoded in MIDI files available online.

There is a toolbox available online which will load the content of MIDI files directly into Matlab:

https://www.jyu.fi/hum/laitokset/musiikki/en/research/coe/materials/miditoolbox/

Here are a few MIDI repositories that I plan to start with:

http://www.classicalarchives.com/midi.html

http://freemidi.org/

http://www.xdigits.com/midi/styles.html

Objective for Milestone: I plan to have generated a style database of motifs and statistics. Then I would be in a good position to perform style classification and composition.