简介:Theanalysisoftheradiatednoiseofvesselsgiveninthispapershowssomestrongsuperposedlinecomponentsinlowfrequencyspeetrumbelow100Hzoccurringatdiscretefrequencieswhichcorrespondwiththerotationspeedofpropellershaft,orpropellerbladefrequency,ortheirharmonicfre-quencies.sincethelinecomponentsreflectpropeller'workingcharacteristics,thepropller'sfeaturescanbeextracteddirectlyfromlow-frequencylinecom-ponentsinadditiontodemodulatedlinecomponent.Sotherearetwowaystoextractthefeatures,oneisdirectway,theotherisdemodulationway.Detec-tionperformanceofthelinecomponentinbackground-noiseisdiscussedinthispaper.ThesignallevelisdefinedasthedifrerencebetweenthePDF's(ProbabilityDensityFunction)meanofthepeakofthelinecomponentandPDF'smeanorthebackground-noise.Indircetwaythesignallevelofthelinecomponentisproportionaltothesignalnoiseratio(S/N).Indemodulationwaythesignallevelofdemodulatedlinecomponent
简介:Usingtheextremedifferenceofself-similarityandkurtosisatlargelevelscaleofwavelettransformapproximationbetweenthePTFM(PulseTrainsofFrequencyModulated)signalsanditsreverberation,afeature-basedmatchedfiltermethodusingtheclassify-before-detectparagriamisproposedtoimprovethedetectionperformanceinreverberationandmultipathenvironments.Processingthedataoflake-trailsshowedthattheprocessinggainoftheproposedmethodisbiggerthanthatofmatchedfilterabout10dB.Inmultipathenvironments,detectionperformanceofmatchedfilterbecomebadlypoorer,whilethatoftheproposedmethodisimprovedbetter.Itshowsthatthemethodismuchmorerobustwiththeeffectofmultipath.
简介:Soundsourcerecognitionisapartofenvironmentalsoundrecognition,whichisoneofthemostimportantresearchareasinpatternrecognition.Impactsoundscarrymuchphysicalinformationassociatedwiththesoundsources,whichmakesimpactsoundbasedsoundsourcerecognitionanimportantapproachtoimproverecognitionperformance.Inthisstudy,theimpactsoundcontinuumsynthesizedwithaball-platecollisionmodelisusedformaterialrecognitionoftheimpactedplates.Thebasisfunctionlearningmethodandtime-frequencyrepresentationmethods,includingtheshorttimeFouriertransformandthewaveletpackettransform,areappliedintoclassificationandtherecognitionresultsarecompared.TheresultshowsthatthefeaturesobtainedbyusingthebasisfunctionlearningperformbetterformaterialclassificationoftheimpactedplatesthanthatbyusingtheshorttimeFouriertransformandthewaveletpackettransform.Thisdemonstratesthehighefficiencyandsuperiorityofthismethodinmaterialrecognitionofsoundsources.
简介:InordertoimprovetheperformanceofdeceptiondetectionbasedonChinesespeechsignals,amethodofsparsedecompositiononspectralfeatureisproposed.First,thewaveletpackettransformisappliedtodividethespeechsignalintomultiplesub-bands.Bandcepstralfeaturesofwaveletpacketsareobtainedbyoperatingthediscretecosinetransformonlogarithmicenergyofeachsub-band.ThecepstralfeatureisgeneratedbycombingMelFrequencyCepstralCoefficientandWaveletPacketBandCepstralCoefficient.Second,K-singularvaluedecompositionalgorithmisemployedtoachievethetrainingofanover-completemixturedictionarybasedonboththetruthanddeceptivefeaturesets,andanorthogonalmatchingpursuitalgorithmisusedforsparsecodingaccordingtothemixturedictionarytogetsparsefeature.Finally,recognitionexperimentsareperformedwithvariousclassifiedmodules.Experimentalresultsshowthatthesparsedecompositionmethodhasbetterperformancecompariedwithconventionaldimensionreducedmethods.Therecognitionaccuracyofthemethodproposedinthispaperis78.34%,whichishigherthanmethodsusingotherfeatures,improvingtherecognitionabilityofdeceptiondetectionsystemsignificantly.
简介:Aspeakermodelcalledcompletefeaturecorpus(CFC)anaanevaluationofmutualinformation(MIE)areproposedfortext-independentspeakeridentification.TheCFCmodelrepresentsthespeechandpronunciationcharacteristicsofspeakerbyafeaturevectorcorpuswhichwastrainedfromsometypicalspeechsamples.Ithiresmulti-stepmini-maxsearchmatchingschemeforMIEalgorithmtoevaluatethesimilarityofspeechfeaturesbetweeninputspeechandthemodelsilldistanceandinformationspace.Maximummutualinformation(MMI)decisioncriterionisusedtodecidetheidentityofspeaker.ExperimentsonperformanceanalysiswithcomparisontoGMMmethodshowthatproposedmodelandevaluationalgorithmarequiteeffectiveandpresentedahigherperformancethanordinaryGMMmethod.
简介:
简介:Thesafetyofrailisveryimportantforthedevelopmentofhighspeedrailway,anditisnecessarytoinvestigatethefeaturesofinnercracksinrail.InordertoobtainthefeaturesofAcousticEmission(AE)sourcesofinnercracksinrail,AEsourceswithdifferenttypes,depthsandpropagationdistancesareexaminedforcrackinrail.Thefiniteelementmethodisutilizedtomodeltherailwithcracksandtheresultsofexperimentdemonstratetheeffectivenessofthismodel.WavelettransformandRayleigh-LambequationsareutilizedtoextractthefeaturesofcrackAEsources.TheresultsillustratethattheintensityratioamongAEmodescanidentifytheAEsourcetypesandtheAEsourceswithdifferentfrequenciesinrail.ThereareuniformAEmodefeaturesexistingintheAEsignalsfromAEsourcesinrailweb,howeverAEsignalsfromAEsourcesinrailheadandrailbasehavethecomplexandunstableAEmodes.DifferentAEsourcetypeshavethedifferentpropagationfeaturesinrail.ItishelpfultounderstandtherailcracksanddetecttherailcracksbasedontheAEtechnique.
简介:Foraccuracyandrapidityofaudioeventdetectioninthemass-dataaudioprocessingtasks,agenericmethodofrapidlyrecognizingaudioeventbasedon2D-HaaracousticsuperfeaturevectorandAdaBoostisproposed.Firstly,itcombinescertainnumberofcontinuousaudioframestobean'acousticfeatureimage',secondly,usesAdaBoost.MHorfastRandomAdaBoostfeatureselectionalgorithmtoselecthighrepresentative2D-Haarpatterncombinationstoconstructsuperfeaturevectors;thirdly,analyzesthecommonalityanddifferencesbetweensubcategories,thenextractscommonfeaturesandreducesdifferentfeaturestoobtainagenericaudioeventtemplate,whichcansupporttheaccurateidentificationofmultiplesub-classesanddetectandlocatethespecificaudioeventfromtheaudiostreamaccurately.Experimentalresultsshowthattheuseof2D-Haaracousticfeaturesupervectorcanmakerecognitionaccuracy5%higherthanonesthatMFCC,PLP,LPCCandothertraditionalacousticfeaturesyielded,andcanmakethetrainingprocessing7-20timesfasterandtherecognitionprocessing5-10timesfaster,itcanevenachieveanaverageprecisionof93.38%,anaveragerecallof95.03%undertheoptimalparameterconfigurationfoundbygridmethod.Aboveall,itcanprovideanaccurateandfastmass-dataprocessingmethodforaudioeventdetection.