基于波原子变换的语音信号去噪

《基于波原子变换的语音信号去噪》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、Contents1.INTRODUCTION12.WAVEATOMTRANSFORM42.1DefinitionofWaveAtomTransform42.21-DCharacteristicsinWaveAtomTransform62.3TheStructureof1-Dfilter-bankinWaveAtom……………………………….…103.THRESHOLDINGMETHODFORNOISEREDUCTION173.1HardandSoftThresholdings173.2DeterminationofThreshold204.EXPERIMENTSANDDISCU

2、SSIONS……………………………………………..274.1ExperimentalConditions…………………………………………………………..274.2ResultsandDiscussions285.CONCLUSION37REFERENCE38ABSTRACT(INKOREAN)40i401.INTRODUCTIONSpeechisoneofthemostnaturalandconvenientwaysofintercommunication.Thus,speechsignalprocessingtechniquesforaman-machineinterface

3、likespeechrecognition,speechsynthesisandspeechcodinghavedrawnmuchattention.Asspeechprocessingmovedfromthelaboratorytothefield,ithasbecomeincreasinglyimportanttodealwiththeambientnoise.Sincethequalityandintelligibilityaredegradedbytheambientnoise,theproblemofreducingnoisecomponentsofthenoisys

4、peechisstillregardedasanimportantissueinthefieldofspeechresearch[1-3].Mostclassicalmethodsforhandlingambientwidebandnoisewerevariantsofanapproachbasedonsubtractinganestimateofthespectrumofthenoisefromthatofthenoisyspeech[4-7].Andthenanovelapproachfornoisereductionusingthewavelettransformhasb

5、eenproposedbyD.L.Donoho[8].ItemploysthethresholdingtechniqueinthewaveletdomainandhasshowntohavegoodpropertiesforawideclassofsignalscorruptedbyadditivewhiteGaussiannoise.Overthepasttwodecades,numeroussignalprocessingtechniqueswiththewavelettransformhavebeendevelopedtomakeuseofitssuperiorityto

6、theconventionalFouriertransform.Recently,waveatomtransformhasbeenproposed,andhasshownits40potentialindenoisingexperimentsonbothseismicandfingerprintimages[1].Waveatomwasintroducedasavariantof2Dwaveletpacketsobeyingtheparabolicscaling.Inanutshell,waveatomsinterpolateexactlybetweendirectionalw

7、aveletsandGabor,inthesensethattheperiodofoscillationsofeachwavepacket(wavelength)islinkedtothesizeoftheessentialsupport(diameter)bytheparabolicscaling,i.e.,wavelength~(diameter)2[1,2].Inthisthesis,weapplythe1Dwaveatomtransformtonoisyspeechsignalsfo