Abstract
Can machine learning algorithms be trained to recognise beauty in music?To what extent is human recognition of beauty in music cultural, or cross-cultural? Music is prevalent in all human cultures. Music information retrieval is a growing field in which computational techniques have been applied to many musical problems such as genre recognition and measuring musical similarity. Computational ethnomusicology is rarer because the acquisition of non-Western music is difficult. Beauty in music has been little investigated with scientific methods, though there are some examples on which this thesis builds. The effect of timbral and 12-step chroma audio features, and a wide variety of diferent machine learning algorithms techniques were tested, with the combination of all the MARSYAS features and Support Vector Machines performing well. Predicting beauty was first investigated with a small Last.fm set and later with a larger world music survey with Singaporean participants. Beauty was predicted based on a small selection of Last.fm tags with good accuracy. Beauty ratings from the survey, conducted in Singapore, were predictable by machine learning using similar methods. Predicting the geographical origin of world music from audio features was attempted. Some promising results emerged, and novel methods for predicting points on the surface of the Earth were developed. An investigation into the link between beauty ratings and location was conducted. The Singaporean beauty ratings were predicted from audio content, geographic content and a combination of both, showing strong correlations between longitude, distance, and timbral features with the beauty ratings, which were statistically very closely linked with distance from Singapore. From this beauty in music is concluded to be culturally related and timbre is shown to be a good pointer to cultural differences,
| Date of Award | 21 May 2013 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Sponsors | Engineering and Physical Sciences Research Council |
| Supervisor | Mark Neal (Supervisor) & Maria Liakata (Supervisor) |