This study aims to use supervised learning – specifically, support vector machines – as a tool for a music mood classification task. Four audio and MIDI datasets, each containing over four hundred files, were composed for use in the training and testing processes. Mood classes were formed according to the valence-arousal plane, resulting in the following: happy, sad, relaxed, and tense. Additional runs were also conducted with the linear discriminant analysis, a dimensionality reduction technique commonly used to better the performance of the classifier. The relevant audio and MIDI features were carefully selected for extraction. MIDI datasets for the same music generated better classification results than corresponding audio datasets. Furthermore, when music is composed with each mood associated with a particular key instead of mixed keys, the classification accuracy is higher.
(See document for exact affiliation information.)
https://aes2.org/publications/elibrary-page/?id=22282
(395KB)
Click to purchase paper as a non-member or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member Join the AES. If you need to check your member status, login to the Member Portal.
