For someone well acquainted with a historically significant body of musical work, telling a Bach fugue from a Mozart piano sonata is the work of a moment's listening. The act on the part of the human is almost without thought. The instruments used, the chords employed, the specific melodic devices, and so on trigger identifications in the human brain. Our goal is to allow machines the same ease of classification. We will focus on classifying pieces of music in their symbol representation (we will be using MIDI files) rather than in their auditory representation. This avoids the complexities of digital audio processing and essentially attempting to reverse engineer notes and instruments from digitally encoded frequencies. As well as performing the work of a human in classifying musical pieces, we are also interested to see what classifications unsupervised learning may come up with
Unlike most prior work in this area, we focus mainly on music which was created before 1900. The twentieth century was a time of musical explosion, and the works of a single composer or band often did not fall within a single genre. Indeed, a single work often spanned multiple genres. While classifying music of the twentieth century is certainly an important problem for music sales and recommendation services, we choose to focus on the problem which has received less attention – classifying older music which is more likely to uphold certain strict forms. Additionally, working with older music avoids copyright issues.
Our goal then is to find an appropriate model and an appropriate feature vector for correctly identifying the genre of each of a collection of works. Previous research into this area has yielded techniques for extracting as many as 160 features from a piece of music 1. We wish to evaluate the effect that each of a variety of features (drawing from prior work and from other features we may view as important) has on the genre predictions made. As humans do by no means evaluate anywhere near 160 data points when classifying the genre of a piece of music, so there must be features which are significantly more important than others, and we aim to find these.
Prior work has shown success applying various models of neural networks to the problems of classifying and clustering music by various features. 2 used k-nearest neural networks and feedforward neural networks, respectively on individual features and on collections of features drawn from a large source using genetic algorithms to optimize the feature selection. However, 2 refrained from any unsupervised learning, remarking that any categorization determined by a machine is unlikely to be recognizable by humans. But we may nevertheless be curious to see what machines come up with as the more distinctive features and clusters. A number of related papers have successfully used the algorithms of 2 for other sorts of classification, such as the mood of pieces.
In 3, Mostafa and Billor used feedforward neural networks, Kohonen self-organizing maps (SOMs), and standard statistical analysis such as k-means and hierarchical clustering to classify and cluster musical pieces. Furthermore, while 3 limited itself to a five different feature variables, it observed that SOMs enable visual display of the distribution of features across the clusters, which we might use to interpret the efficacy of particular features from 2 with more domain-specific knowledge than the genetic algorithms used in 2.
There are a plethora of freely available MIDI files for pre twentieth-century music (music that is often lumped under the term "classical" by those not aware of the finer delineations). This is another advantage to avoiding the classification of modern (copyrighted) music. We have identified two large sources of MIDI files appropriate for our work. The first is the Mutopia Project 4 which has a large variety of music, sorted by composer, ranging from the late 1600s to 1900. The second is a website calling itself The Classical MIDI Connection 5, which has a large variety of music arranged by period.
There has been a reasonable degree of work in this domain already. Basili et al perform an evaluation similar to ours except that they consider only a few genres and consider more modern music 6. McKay and Fujinaga introduce classification systems which make use of a large number of features extracted from MIDI files and theorize about what makes a good feature 2 1. They do not, however, present any experimental determination of the importance of various features and they do not consider a dataset similar to ours. Very recently, Mostafa and Billor used neural networks to cluster and identify genre of songs 6. They focused only on three genres, however and did not use such a large corpus of music.
1 Cory McKay and Ichiro Fujinaga. "Style-independent computer-assisted exploratory analysis of large music collections". Journal of Interdisciplinary Music Studies 1 (1): 63–85. 2007.
2 Cory McKay and Ichiro Fujinaga. "Automatic Genre Classification Using Large High-Level Musical Feature Sets". In Proceedings of the International Converence on Music Information Retrieval, 2004.
3 Mohamed M. Mostafa, Nedret Billor. "Recognition of Western Style Musical Genres Using Machine Learning Techniques". Expert Systems with Applications, Volume 36, Issue 8, October 2009, Pages 11378-11389.
4 http://www.mutopiaproject.org
5 http://www.classicalmidiconnection.com/
6 Roberto Basili, Alfredo Serafini, and Stellato Armando. "Classification Of Musical Genre: A Machine Learning Approach". In Proceedings of the International Conference on Music Information Retrieval (ISMIR), Barcelona, Spain, October 2004.
Date: 2010-04-13 22:14:30 EDT
HTML generated by org-mode 6.21b in emacs 23