A computational approach to classify music by emotion

TitleA computational approach to classify music by emotion
Publication TypeConference Paper
Year of Publication2007
Conference NameEuropean Computing Conference
AuthorsLaurier, C., & Herrera P.
AbstractOne of the main motives for people to listen to music is to modulate their emotional state. Some research efforts in music information retrieval target the development of a system able to detect emotions in music, which, besides is a rather complex task involving the extraction of musical information from acoustical data and the exploitation of machine learning techniques to link emotional (textual) descriptions to musical (acoustic) descriptions. In this paper we present an approach to achieve this challenging work.
The first issue is to define a Taxonomy, a representation of the emotions that should be flexible enough to be relevant for most people. Then we have to organize a music collection, so called "ground truth", that will represent our way to categorize music by emotion. It should be large and typical because it will be used to train a system capable of learning from examples.
From this "ground truth" we will extract musical information from the audio representing rhythmic, spectral, or tonal aspects of the music. Other descriptors can also be designed according to existing knowledge about music and emotions from fields such as musicology or psychology. Finally these descriptors will be used to train a classification algorithm. The results from this research helps us to better understand how humans categorize music, and to grasp some more links between acoustics, music and emotions.
preprint/postprint documentfiles/publications/487cb2-EEC2007-Laurier-Herrera.pdf
intranet