Timbre Similarity Search with Metric Data Structures

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

Similarity search is essential in musiccollections, and involves finding all the music documents ina collection, which are similar to a desired music, based onsome distance measure. Comparing the desired music toall the music in a large collection is prohibitively slow. Ifmusic can be placed in a metric space, search can be spedup by using a metric data structure. In this work, weevaluate the performance of the timbre range query inmusic collections with 6 metric data structures (LAESA,GNAT, VP-Tree, HDSAT2, LC and RLC) in 2 metricspaces. The similarity measures used are the city-blockand the Euclidean distances. The experimental resultsshow that all the metric data structures speeds the searchoperation, i.e. the number of distance computed in eachsearch process is small when compares to the number ofobjects in the database. Moreover, the LAESA datastructure has the best performance in the two metricspaces used, but the RLC data structure is the only datastructure that never degraded its performance andcompetes with the other metric data structures, in all theexperimental cases.
Original languageUnknown
Title of host publicationProc. of Workshop on Exploring Musical Information Spaces (WEMIS 2009)
Pages7-11
Publication statusPublished - 1 Jan 2009
EventWorkshop on Exploring Musical Information Spaces (WEMIS 2009), 13th European Conference on Research and Advanced Technologies on Digital Libraries (ECDL 2009) -
Duration: 1 Jan 2009 → …

Conference

ConferenceWorkshop on Exploring Musical Information Spaces (WEMIS 2009), 13th European Conference on Research and Advanced Technologies on Digital Libraries (ECDL 2009)
Period1/01/09 → …

Cite this