Mood Cloud

Cyril Laurier’s PhD thesis · 2006-2011 · Universitat Pompeu Fabra, Barcelona

Our role PhD thesis, machine learning, audio analysis.

Cyril Laurier’s PhD thesis, “Automatic Classification of Musical Mood by Content-Based Analysis”, defended in 2011 at Universitat Pompeu Fabra in Barcelona. It deals with classifying music automatically by emotion (mood), using models that listen to the audio signal. It draws on signal processing, machine learning and information retrieval.

The main contributions:

Modelling emotion a representation model of musical emotions, derived from analysing the tags of a music social network.
Automatic classification a method for classifying music by emotion, with an analysis of what each audio descriptor contributes and how it relates to perceived emotion.
Combining audio and lyrics a multimodal approach, which also advances information retrieval in text.
Linking emotion and genre after showing the relationship between emotion and musical genre, a new approach integrates automatic genre classification. These classifiers are more accurate than standard audio models.
Making models readable a rule extraction technique makes the models explicit.

The results of this thesis have been used commercially by Bmat, Eventide and Sonaris Systems.