Genre classification based on tone objects (mirex 2013 submission)

This extended abstract details the submission to the 2013 Music Information Retrieval Evaluation eXchange in the Audio Classification Train/Test task. The proposed system is designed to perform improved genre classification by combining temporal segmentation and source separation of music signals into so-called tone objects . The audio features extracted from these tone objects are low-level an d well-known. Gaussian Mixture Models and Support Vector Machines are used for classification. This submission is motivated by promising results obtained on other genre data sets.