hemost important research direction is topick theright distancemeasure which Can wellreflectthespecialhuman perception resultamong SOmany ones.In thispaper,we taketheConstant-QTransform(CQT) feature and research andanalyze deeply monly used distance measures which are Euclidean distance measure.cosine distancemeasureand s缸ucture-based distance measure.At thesame time.musical lre tagging work pleted on adatabase of 0songs (including pop songs,children’S songsandfoll【songs)which allhave repeating patterns.Finally, we get a81.6percent uracy degreeand ageneral 83percent ofF1·measureevaluation.The tagging results show that structure-based distance measure which atfirstaims toreflectthe hearing similaritydegree ofdifferentmusical notes isstillsomewhat problematic on discussing melody similarity.
全文预览
