Show simple item record

dc.contributor.advisor© 2021, EURASIP
dc.contributor.authorZhao, Y
dc.contributor.authorWang, C
dc.contributor.authorFazekas, G
dc.contributor.authorBenetos, E
dc.contributor.authorSandler, M
dc.contributor.author29th European Signal Processing Conference (EUSIPCO)
dc.date.accessioned2021-06-09T12:40:32Z
dc.date.available2021-05-04
dc.date.available2021-06-09T12:40:32Z
dc.date.issued2021-08-23
dc.identifier.urihttps://qmro.qmul.ac.uk/xmlui/handle/123456789/72433
dc.description.abstractIdentifying performers from polyphonic music is a challenging task in music information retrieval. As a ubiquitous expressive element in violin music, vibrato contains important information about the performers' interpretation. This paper proposes to use vibrato features for identifying violinists from commercial orchestral recordings. We present and compare two systems, which take the same note-level melodies as input while using different vibrato feature extractors and classification schemes. One system calculates vibrato features according to vibrato definition, models the feature distribution using histograms, and classifies performers based on the distribution similarity. The other system uses the adaptive wavelet scattering which contains vibrato information and identifies violinists with a machine learning classifier. We report accuracy improvement of 19.8% and 17.8%, respectively, over a random baseline on piece-level evaluation. This suggests that vibrato notes in polyphonic music are useful for master violinist identification.en_US
dc.format.extent? - ? (5)
dc.publisherEURASIPen_US
dc.rightsThis is a pre-copyedited, author-produced version of an article accepted for publication in 29th European Signal Processing Conference following peer review.
dc.titleViolinist identification based on vibrato featuresen_US
dc.typeConference Proceedingen_US
pubs.notesNot knownen_US
pubs.publication-statusAccepteden_US
pubs.publisher-urlhttps://eusipco2021.org/en_US
dcterms.dateAccepted2021-05-04
rioxxterms.funderDefault funderen_US
rioxxterms.identifier.projectDefault projecten_US


Files in this item

Thumbnail

This item appears in the following Collection(s)

Show simple item record