Abstract | ||
---|---|---|
Inference of musical genre, whilst seemingly innate to the human mind, remains a challenging task for the machine learning community. Online music retrieval and automatic music generation are just two of many interesting applications that could benefit from such research. This paper applies four different classification methods to the task of distinguishing between rock and classical music styles. Each method uses the Minimum Message Length (MML) principle of statistical inference. The first, an unsupervised learning tool called Snob, performed very poorly. Three supervised classification methods, namely decision trees, decision graphs and neural networks, performed significantly better. The defining attributes of the two musical genres were found to be pitch mean and standard deviation, duration mean and standard deviation, along with counts of distinct pitches and rhythms per piece. Future work includes testing more attributes for significance, extending the classification to include more genres (for example, jazz, blues etcetera) and using probabilistic (rather than absolute) genre class assignment. Our research shows that the distribution of note pitch and duration can indeed distinguish between significantly different types of music. |
Year | DOI | Venue |
---|---|---|
2003 | 10.1007/978-3-540-24581-0_91 | AI 2003: ADVANCES IN ARTIFICIAL INTELLIGENCE |
Keywords | Field | DocType |
minimum message length,standard deviation,neural network,decision tree,statistical inference,machine learning,unsupervised learning | Decision tree,Minimum message length,Classical music,Computer science,Inference,MIDI,Unsupervised learning,Statistical inference,Artificial intelligence,Artificial neural network,Machine learning | Conference |
Volume | ISSN | Citations |
2903 | 0302-9743 | 1 |
PageRank | References | Authors |
0.34 | 8 | 2 |
Name | Order | Citations | PageRank |
---|---|---|---|
Adrian C. Bickerstaffe | 1 | 1 | 0.34 |
Enes Makalic | 2 | 55 | 11.54 |