学术交流英语final Presentation.docx

学术交流英语final Presentation.docx

《学术交流英语final Presentation.docx》由会员分享，可在线阅读，更多相关《学术交流英语final Presentation.docx（9页珍藏版）》请在冰豆网上搜索。

学术交流英语finalPresentation

ATranscriptforaConferencePaperPresentation

Slide1（Title）:

Time:

35seconds（0:

00-0:

35）

Say:

Goodmorning!

Mymajoriselectricengineering.Myresearchfieldissignalandinformationprocessing.ThetopicIconcernisaboutspeechenhancement.TodayIwillgiveapresentationwiththetitle“ATwo-stageBeamformingApproachforNoiseReductionandDereverberation”.ThisworkisdonebyHabetsandBenestyfromUniversityofQuebec,Montreal,Canada.

Slide2（Introduction）

Time:

55seconds（0:

35-1:

30）

Say:

First,Iwillgiveabriefintroductionofmicrophonearraysandmakeyouhaveapreliminaryunderstandingoftheresearchproblem.Simply,whenyouplaceseveralmicrophonesaccordingtocertaingeometricshapes,yougetamicrophonearray.Hereisalinerarrayandacirclearrayisalsogeneral.Let’stakethispictureasanexample.Thereisanoisesourceatthelocationofthestar.Whenthegirlspeaks,hersoundwillgetcapturedbyallmicrophones.Inthesametime,receivedsignalsarepollutedbytheundesirednoiseandwecan’thavethecleanspeech.

Slide3（Introduction）

Time:

55seconds（1:

30-2:

25）

Say:

So,weneeddosomethingtosolvethisproblem.Inthisslide,thebackgroundandsignificanceofspeechdenosinganddereverberationareintroduced.Inmanyapplications,suchasspeechrecognitionandteleconferencing,weneeddistantorhand-freeaudioacquisition.However,inmanycases,wejustreceiveanoisecorruptedorreverberantversionofdesiredspeechsignals.Toachievehigh-qualityhuman-to-humanorhuman-to-machinespeechcommunication,weneedtodevelopefficientnoisereductionanddereverberationalgorithms.Microphonearrayscanbeveryusefulinthesesituation.

Slide4（Body）

Time:

50seconds（2:

25-3:

05）

Say:

First,wecanusebeamformingforthemicrophonearraysduringtheprocessingofreceivedsignals.Whatisthebeamforming?

Hereareitsdefinitionandworkingprinciple.Beamformingisasignalprocessingtechniquethatappliesmicrophonearraysfordirectionalsignaltransmissionorreception.Byoperatingonreceivedmultichannelsignals,beamformingallowsustorecoversignalsfromaparticulardirectionandsuppressnoisesignalsfromundesireddirections.Thistechniqueissocalled“beamforming”.

Slide5（Body）

Time:

45seconds（3:

05-3:

55）

Say:

Inthispaper,weusebeamformingtoachievenoisereductionanddereverberation.Noisereductionisimportantsincenoiseiseverywherearoundus.Somecommonnoiseincludesmachinenoise,vehiclenoise,musicnoise,babblenoise,andsoon.Ontheotherhand,thereverberationiscreatedwhenasoundisproducedinanenclosedspacecausingalargenumberofechoestobuildupandthenslowlydecayasthesoundisabsorbedbythewallsandair.

Slide6（Body）

Time:

60seconds（3:

55-4:

55）

Say:

Toachievebothnoisereductionanddereverberation,thetwo-stageapproachisproposedinthispaperandbeforethenoisereductionstage,adereverberationstageisneeded.Hereistheprinciplediagram.TheseyrepresentthereservedsignalsbythemicrophonearrayandwehaveNmicrophone.TheseQandHrepresentweightingcoefficientsoftwodifferentbeamformingstages.TheZrepresentsthefinalsignalafternoisereductionanddereverberation.Inthenextfewslides,thedetailsabouthowthealgorithmworkaregiven.

Slide7（Body）

Time:

45seconds（4:

55-5:

45）

Say:

Thefirststageisdereverberationstage.Inthisslide,thecomputationalprocessofdereverberationstageispresented.Allchannelinputsareweighted.Theweightedchannelinputsaresenttothenextstagefornoisereduction.Sothekeyistofindproperweightssothatthereverberationcomponentsareminimized.Thisisimplementedbycomplexmathematicscomputation.Thefinalweightsareindependentonsignals.

Slide8（Body）

Time:

45seconds（5:

45-6:

25）

Say:

Onthebasis,furtheranalysisofthedereverberationstageisneeded.Thefirststagecomprisesasignal-independentbeamformerthatgeneratesareferencesignalthatcontainsadereverberatedversionofthedesiredspeechandresidualinterference.Ingeneral,thedesiredspeechcomponentattheoutputofthebeamformercontainslessreverberationcomparedtoreverberantspeechsignalreceivedatthemicrophones.

Slide9（Body）

Time:

40seconds（6:

25-6:

55）

Say:

Thesecondstageisnoisereductionstage.Inthisslide,thecomputationalprocessofnoisereductionstageispresented.Theweightedinputsobtainedinthefirststageareagainweightedandsummed.Theweightsarecomputedsothatthesignal-to-noiseratio（usuallycalledSNR）ismaximized.SinceSNRisindependentondesiredsignals,theweightsareindependentonsignalsaswell.

Slide10（Body）

Time:

45seconds（6:

55-7:

25）

Say:

Furtheranalysisisalsoappliedtothenoisereductionstageisdereverberationstage.ThesecondstageusesthefilteredmicrophonesignalsandthenoisyreferencesignaltoestimatethedesiredspeechcomponentattheoutputoftheDSbeamformer.Amajoradvantageoverclassicalapproachesisthattheproposedapproachisabletodereverberatethereceiveddesiredsignalwithverylowspeechdistortion.Thisisthewholeprocessoftheproposedalgorithm.

Slide11（Body）

Time:

20seconds（7:

25-7:

45）

Say:

Let’sseetheperformance.Herearetwographswhichrepresentthereceivedspeechsignalbyonemicrophoneandtheprocessedspeechsignalbytheproposedtwo-stageapproach.Inthefirstgraph,thefuzzypartsdenotenoiseandreverberation.Theyhasbeenweakenedinthesecondgraph.Inthisway,betterperformanceisachieved.

Slide12（Conclusion）

Time:

20seconds（7:

45-8:

05）

Say:

Inconclusion,ourgoalistofindamethodwhichcanachievebothdereverberationandnoisereductionwhilecausinglowspeechdistortionasmuchaspossible.Aftertheintroductionofthewholeprocessoftheproposedalgorithm,let’ssummarizewhatwehavegot.

Slide13（Conclusion）

Time:

10seconds（8:

05-8:

25）

Say

First,atwo-stagebeanformingapproachisdesigned.Thefirststageisasignal-independentbeamformerthatgeneratesareferencesignalwhichcontainsdereverberatedversionresidualinterferenceandthesecondstageismultichannelnoisereductiontoestimatethedesiredspeechcomponentattheoutputofthefirststage.Finally,betterperformanceisobserved.

Slide14（Conclusion）

Time:

25seconds（8:

25-8:

50）

Say:

Whyisthissignificant?

Ontheonehand,weneedmorerecognizableandclearerspeechinsteadofanoisyworld.Ontheotherhands,thisworkissignificantinthefutureapplicationofspeechtechnologies.Forexample,youcanmakeatelephonecallwithoutthemobilephoneinhand.Youcancontrolyoursmartdevicesbyvoiceevenwhenyouarefewmetersaway.

Slide15（Conclusion）

Time:

15seconds（8:

50-9:

00）

Say

Inthefuture,wewillimplementproposedalgorithmanddesignbetterspeechenchantmentalgorithms.Thankyou,arethereanyquestions?