Content-Based Music Analysis

Abstract:

The quantity of music available ubiquitously is growing rapidly. There is thus a need for analysis techniques to automatically organize vast audio repositories. In this talk, we describe several of our efforts in this direction: automatic music summarization and automatically determining music similarity. Our approaches bring together knowledge from the signal processing, machine learning and information retrieval fields.

Both of our techniques extract spectral features from each song and learn statistical models of these. Our music summarization technique then automatically chooses a representative phrase for each song using the segmentation provided by its model. Our music similarity technique compares the models for each pair of songs using the Earth Mover's Distance (Rubner1998) to form a distance matrix. Both approaches show great promise, evidenced by objective and subjective results and demonstrations.

Beth Logan received the BSc. and B.E. degrees from the University of Queensland, Australia, in 1990 and 1991 respectively. She received the PhD in engineering from the University of Cambridge, United Kingdom, in 1998, completing a dissertation on speech enhancement. Since 1998, she has been a research scientist at HP Labs (formerly Digital) in Cambridge Massachusetts. Her work here has focused on scalable organization of digital content, primarily looking at indexing and modeling of speech and music.

More information available at http://www.hpl.hp.com/personal/Beth_Logan/

Parking: Visitors requiring a parking pass are asked to contact us at least one working day before the seminar.

Enquiries: sals@ics.mq.edu.au