From Low-level to Song-level Percussion Descriptors of Polyphonic Music

TitleFrom Low-level to Song-level Percussion Descriptors of Polyphonic Music
Publication TypeConference Paper
Year of Publication2009
Conference Name10th International Society for Music Information Retrieval Conference
AuthorsHaro, G. G., & Herrera P.
Conference LocationKobe, Japan
Keywordsdrums transcription, music information retrieval, semantic descriptors
AbstractWe address here the automatic description of percussive events in real-world polyphonic music. By taking a pattern recognition approach we evaluate more than 2,450 objectlevel features. Three binary instrument-wise support vector machines (SVM) are built from a training set of more that 100 songs and 10 genres. Then, we use these binary models to build a drum transcription system achieving comparable results with state of the art algorithms. Finally, we present 17 song-level percussion descriptors computed from the imperfect output of the transcription algorithm. We evaluate the usefulness of the proposed descriptors in music information retrieval (MIR) tasks like genre classification, danceability estimation and Western vs. non-Western music discrimination. We conclude that the presented song-level percussion descriptors provide complementary information to “classic” descriptors, that can help in the previously mentioned MIR tasks.
preprint/postprint documentfiles/publications/Song-level_Percussion_final_submission.pdf