Improving Multilabel Analysis of Music Titles: A Large-Scale Validation of the Correction Approach

By François Pachet and Pierre Roy.

IEEE Transactions on Audio, Speech, and Language Processing, 17(2), pp. 335–343, February 2009.

Read the paper (PDF)
Publisher record

About the paper

This paper tests whether relationships among music labels can improve automatic tagging on a large catalog. An additional layer of classifiers learns how to correct predictions for labels such as genre, mood, instruments, and vocal type.

Abstract summary

Pachet and Roy evaluate the correction approach on a database of about 30,000 tracks and roughly 600 labels. They report statistically significant improvements by using dependencies among labels to correct some errors made by the individual acoustic classifiers.

Related work

Signal + Context = Better Classification (ISMIR 2007) introduced the earlier context-based classification architecture. See Timbre Similarity and Hubness for the audio models, Feature Generation for work on task-specific descriptors, and Active Listening for applications to music browsing.

Back to Publications