Manipulation of music for melody matching

Many types of user would find it valuable to search collectionsof music via queries representing music fragments, but suchsearching requires a reliable technique for identifying whether aprovided fragment occurs within a piece of music. The problem ofmatching fragments to music is made difficult by the psychology ofmusic perception, because literal matching may have litile relationto perceived melodic similarity, and by the interactions betweenthe multiple parts of typical pieces of music. In this paper weanalyse the propeties of music, music perception, and musicdatabase users, and use the analysis to propose alternativetechniques for extracting monophonic melodies from polyphonicmusic; we believe that such melodies can subsequently be used formatching of queries to data. We report on experiments with musiclisteners, which rank our proposed techniques for extractingmelodies.