09 Sound Isolation by Harmonic Peak Partition For Music Instrument Recognition

Description

Fundamenta Informaticae XXI (2001) 1001–1017 1001IOS PressSound Isolation by Harmonic Peak Partition For Music InstrumentRecognitionXin ZhangDepartment of Computer Science,University of North Carolina, CharlotteNC 28223, [e-mail]´Zbigniew W. RasDepartment of Computer ScienceUniversity of North Carolina, CharlotteNC 28223, [e-mail]. Identification of music instruments in polyphonic sounds is difficult and challenging, es pecially where heterogeneous harmonic partials are overlapping with each other. This has stimulatedthe research on sound separation for content based automatic music information retrieval. Numer-ous successful approaches on musical data feature extraction and selection have been proposed forinstrument recognition in monophonic sounds. Unfortunately, none of those algorithms can be suc cessfully applied to polyphonic sounds. Based on recent successful researches in sound classificationof monophonic sounds and studies in speech recognition, Moving Picture Experts Group (MPEG)standardized a set of features of the digital audio content data for the purpose of interpretation of theinformation meaning. Most of them are in a form of large matrix or vector of large size, which are notsuitable for traditional data mining algorithms; while other features in smaller size are not sufficientfor instrument recognition in polyphonic sounds. Therefore, these acoustical features themselvesalone cannot be successfully applied ...

Informations

• Publié par: Ornyo
• Nombre de lectures: 15
• Langue: English

Extrait

Fundamenta Informaticae XXI (2001) 1001–1017 IOS Press

1001

Sound Isolation by Harmonic Peak Partition For Music Instrument Recognition

Xin Zhang Department of Computer Science, University of North Carolina, Charlotte NC 28223, USA [e-mail] ZbigniewW.Ras ´ Department of Computer Science University of North Carolina, Charlotte NC 28223, USA [e-mail]

Abstract. Identification of music instruments in polyphonic sounds is difficult and challenging, es- pecially where heterogeneous harmonic partials are overlapping with each other. This has stimulated the research on sound separation for content-based automatic music information retrieval. Numer- ous successful approaches on musical data feature extraction and selection have been proposed for instrument recognition in monophonic sounds. Unfortunately, none of those algorithms can be suc- cessfully applied to polyphonic sounds. Based on recent successful researches in sound classification of monophonic sounds and studies in speech recognition, Moving Picture Experts Group (MPEG) standardized a set of features of the digital audio content data for the purpose of interpretation of the information meaning. Most of them are in a form of large matrix or vector of large size, which are not suitable for traditional data mining algorithms; while other features in smaller size are not sufficient for instrument recognition in polyphonic sounds. Therefore, these acoustical features themselves alone cannot be successfully applied to classification of polyphonic sounds. However, these features contains critical information, which implies music instruments’ signatures. We proposed a novel music information retrieval system with MPEG-7-based descriptors and we built classifiers which can retrieve the important time-frequency timbre information and isolate sound sources in poly- phonic musical objects, where two instruments are playing at the same time, by energy clustering between heterogeneous harmonic peaks.

Keywords: Music Instruments Detection, MPEG-7 descriptors, Musical Sound Separation, Energy

utilniedenilcsrae,alicwhschiselo ronaFFpTiotn fohtelocalmaximumamp dnufehtfflatnemacyenqureths,hu.T TFopotFatiahniwtgerminteple o ulti amfranecinntcheaebonicpeakejthharmnitsuremfohtkeht d). Amplitudeinli (tomonhpnociosnu3AofflaB-latcneri snartnei tatsafoe ectrerspthetuminugeriFpewo.1hT arscale.deinlineA . pmilut ciosnu)d teburodadi ly rist eromneverenEsiyg rs(ssifie aks.icpemrnolaahverenusd 13e[ses(eeTronsiiceD,)]5[,]71[ee siidtsirubet a d or nearscale.Energy skae giF . 2eruehT. daunwhfemoarcpni sa-iehuqyd ts tsaerspepowemintctru (teniralnohponomoffaeoattcflaB-3A c lac ssal areiihcr ocpruredcaifiontirehtkeN-seoteghthmandothNalgoritevahsrefiissalcrediinedorplexenbe rdceosnuertn -f efems(systtionogni ,D])crisinimtAanylan,sishgiHrOrederStatistics,Aritcfiai N l uearNlteSus,rkwoectVorppihcaMrotguoR,sensandhSetenMaHiddoMedkrvoctH.sle, icustmenerffdiofsroivaheberutaeftyofarieidevthew euotred,wovesmH. anthheotbelyerttfiin - tnac adetgisseenreporifierhasbn,coalsslaosnusd ehtdesae cfiissalc swerifisscrinchhiperorerecsaletad18][see[andw9]), issatacfisnoietsyhin-arericchclalmoapirostntoehonationaccuracyinc neyceruqatflmanectrurspepowe.Inaegetnierascinomrndfuofespltiulrm e,engreiarsnofmrributedcsaredist oroTrehtorfmFTSm urfortiescditere fererdifassintclotlcefisryfhtsaislcalinrshu.Tesascevahewsuofnesoh arch. 2.SoundIsoltaoiAnglrotimh aH ep e satarmuedcasiuoslisdnruoneser ndas net ain4Cound id3nosnualiroAcf esthatede,imetamniloivfoyalperew appingwitheachots t uocrseraoeevlrsonadmunwhixeaer.reheroFpmaxi,el teoslpseluitsem,itchplepultiithmnemurtsnitnereffdiomfrcsnimoarfh enraaehcTFopnistseveralFloselyon osciwdnuylopnohpakpena.I rmhaicon iopTFFfo eht nistn perswepoytmbructegcakaniominhhrakasacpeatercclus tnI.rsihaese,hcrapweieplcldateusirgnethcinuqtesoeparatetheenergy Hz40f4soleipltmusdaerpsrehtoeht;rmultegendinarou 52zH25.3 sefoitlp ostwsoeterthreeao:scesenrahfinom ibutedne t isdistriltngereraesevar verestnty.raehThmmaHwgniodnipotwtherintime,withaalppniwgtiehcaoh hwicwhs,r-veeoerbuswotnitnemges-discsbyaSTFTretemalpidsoemtnsegessecuadeW .)3 orpe ,(4),(5)figeee(urciepkaS.hcahmrnoaroundeaheenergytgniretsulcybdetenemplimenbe asnhtaoiperanusd,eoseforTheroid.entr ciepkawsreeetsmiatedbysearchingf-rewt-founowrebmFF(N.TT)haheonrm oiwnlotu84sasa02ext-then erpolarg lim03ottdnocesileterwhs,esTrFFhezeofeach on.The sinewtsaesus-besmg uiacoftymptealorcrepitpeamuhraeneraninthwithchisfsroametseitegfo 10toetsswazesiop ihw,sdnocesillim

en. Figure4.Onese ecuderhslablteka e c th stluinerrogp ethggerdwnoiereh encyrequ ure5 . Fig ufdnonsuatflmane itswicon gemo o-hhrapesfotmrahdeta fleruqnednmaneatgenousfuwithhomo omrascinaraphdetetrsSeofhe.Theot ( » )j r S , º 2 gewd 2 ol j)l k ∈ " ¹ ( )),n(χ(nxam= j k h .)01(alumornfditepuom . c cy .3ermraHugiFstluilreyancueeqilhgihahtiwdetarfferwodisoftoniclarfemtnnuadneft onrmhaofstakpeiclus -ereh stestnatceofquensH p j peakehtrgoteamesforoasanndceg o in errd htiwremaosegnitr nthissequenceofp ybhtfeeruqneyc I . ,skaeprobhgienfoirpacheaor,fkseano:sauitsetiislbeposthreearether srobferankaehgiedimeepat e t th imwoteowmiemuocr;ehtmesoundsromthesa ol 2 g k l d 2 # ¼ 1( ew r S rel k ) wheefunisthlatnemadcneuqerfktheftyoumtrnshine,t(χ)nsihtFeTFcoefficient.Herewmile ht ti narefoegarsefochloralpcackinidpioavoeaktkafoybepenraugapursondoursheotan ,yltneuqesnoC .ec tinstrumkdominanerustlninestiwllmoarcpnietksfhsoew,nttupskaeehT. refo.Thercesdsouosnut e ow htsoskfoeacpnimoarshouvierpehtfooitarehtordingtourcesaccsouodnoseetnehwtgyertwbe ntio en heaptstitrtsysrfime teretahe onsu go-eeentbetwmediwoimnoitarapygrenefonueesbhaseordfsetsreniagglrotimh processed. Theclu onegepsuiwkaebll,oreytnlheherote ehT.ecruos dnuose touesdeiasdcirth ehprehtonatdepkaesamomthrefreakaosdnecrunereuosteoonheftpas;ofrt nkiehgobidtapeae mtwodiff rsare fro sondoutsIns.ceurnecssihteht,oirantaldameuencfreqtfoweiosreneidff ncuestyimuheipltacilnoithtfo nuf etwooverlappingpeka,shwreteehrfqe tnufdnmadffieferental(sonepkabhroengi p j+1 andHsH p j tedaisniatnochcihw,)soleipltmueregnt