资源描述:
《强噪声环境下的语音增强算法-研究与其dsp实现》由会员上传分享,免费在线阅读,更多相关内容在行业资料-天天文库。
1、重庆大学硕士学位论文II英文摘要ABSTRACTNoiseseriouslyaffectsthequalityofspeechsignalundercomplexenvironment,leadinguscannotconveythesemanticcorrectly,sospeechenhancementprocessingbecomesverynecessary,andhasalreadybecomeahotresearchtopiconspeechsignalprocessingathomeandabroad.Atpresent,therearelotsof
2、algorithmsonspeechenhancementtechnology,andthenoisecancellationtechnologybasedonadaptivefilteringtechniquehasbecomeawidelyusedspeechenhancementmethodforitsstrongself-learningcharacteristicandexcellenttrackingperformance.However,thetraditionaladaptivefilteringtechnologyexistproblemsasf
3、ollows:pooradaptability,slowconvergencespeedwhentheinputsignalswereheavilycorrelated,whichlimititsdevelopmentonacertaindegree.Inordertosolvethisproblem,animprovedspeechenhancementalgorithmwasproposedafteranalysistheprincipleandperformanceofthestandardleastmeansquare(LMS)algorithm,itus
4、esadaptivefiltertheoryandcomprehensivestheadvantagesofvariablestepsizeLMSalgorithmandde-correlationprinciple.Thenewalgorithmusestongue-likecurvetoupdatethestepsizeofadaptivefiltering,atthesametimeitalsointroducesthede-correlationprincipleforspeechde-noising.Moreover,comparedthealgorit
5、hmproposedinthispaperwithotheralgorithmsontraceability,computationalcomplexityandspeechenhancementeffect,whichprovedthatthealgorithmproposedinthispapercanimprovetheadaptabilityandconvergencerateandreducesthesteady-stateerroroftraditionalalgorithm.Consideringthehardwareimplementation,t
6、hisarticlealsotakestheDSPchipwiththehighspeedframeprocessingabilitytobethecontrolcoreofthesystem,meanwhile,usesitsmulti-channelbufferedserialport(McBSP)toconnectwiththeextendedaudiointerfacechipTLV320AIC23,whichcanrealizeseamlessconnection.Sothecomplexityofthesystemdesignissimplified,
7、alsoitmakesthesystemwithhighspeedacquisition、de-noisingprocessingandoutput.Atthesametime,de-noisingeffectwastestwithdifferenttypesofspeechsignalsbasedontheDSPsystem.Thefinalexperimentalresultsshowthat,thesignal-to-noise(SNR)rateisimprovedsignificantlywiththenewalgorithm.Atthemeantime,
8、theli