AUTOMATIC MUSIC TAG CLASSIFICATION BASED ON BLOCK-LEVEL FEATURES

Publication Type:

Conference Paper

Source:

SMC Conference 2010 (2010)

URL:

files/proceedings/2010/19.pdf

Abstract:

In this paper we propose to use a set of block-level au- dio features for automatic tag prediction. As the proposed feature set is extremely high-dimensional we will investi- gate the Principal Component Analysis (PCA) as compres- sion method to make the tag classification computationally tractable. We will then compare this block-level feature set to a standard feature set that is used in a state-of-the- art tag prediction approach. To compare the two feature sets we report on the tag classification results obtained for two publicly available tag classification datasets using the same classification approach for both feature sets. We will show that the proposed features set outperform the stan- dard feature set, thus contributing to the state-of-the-art in automatic tag prediction.