Automatic Labeling of Unpitched Percussion Sounds

TitleAutomatic Labeling of Unpitched Percussion Sounds
Publication TypeConference Paper
Year of Publication2003
Conference NameAES 114th Convention
AuthorsHerrera, P., Dehamel A., & Gouyon F.
AbstractWe present a large-scale study on the automatic classification of sounds from percussion instruments. Different subsets of temporal and spectral descriptors (up to 207) are used as features that several learning systems make use of to learn class partitions. More than thirty different classes of acoustic and synthetic instruments and near twothousand different isolated sounds (i.e. not mixed with other ones) have been tested with ten-fold or holdout crossvalidation. The best performance can be achieved with Kernel Density estimation (15% of errors), although 1- Nearest Neighbors and boosted rule systems yielded nearly similar figures. Multidimensional scaling of the classes provides a graphical and conceptual representation of the relationships between sound classes, and facilitates the explanation of some types of errors. We also explore some options to expand the sound descriptors beyond the class label, as for example the manufacturer-model label and confirm the feasibility for doing that. We finally discuss methodological issues regarding the generalization capabilities of usual experiments that have been done in this area.
preprint/postprint documentfiles/publications/AES114-Herrera2003.PDF