Multimedia Communication Experiment Guide.docx

Multimedia Communication Experiment Guide.docx

《Multimedia Communication Experiment Guide.docx》由会员分享，可在线阅读，更多相关《Multimedia Communication Experiment Guide.docx（21页珍藏版）》请在冰豆网上搜索。

MultimediaCommunicationExperimentGuide

EXPERIMENT1AUDIOPROCESSING

（3hours）

1.Aims

◆Tolearntheproceduresofdigitizingsound,suchas:

●Sampling

●Quantization

●Encoding

◆Tolearnthesignificanceofsamplingandquantizationfortheaudioquality.

◆Tolearnhowtoeditandanalyzeaudio.

2.BasicTheory

2.1SoundFundamentals

Soundisaphysicalphenomenonproducedbythevibrationofmatter.Vibrationofmatterinducestheoscillationoftheairpressurearoundthematter.Then,thepressurevariationpropagatesinawave-likemotion,andwhenitarrivesatourear,wehearthesound.Withoutairthereisnosound,forexample,inspace.Sincesoundisapressurewave,ittakesoncontinuousvalues,asopposedtodigitizedoneswithafiniterange.Nevertheless,ifwewishtouseadigitalversionofsoundwaves,wemustformdigitizedrepresentationsofaudioinformation.

Thetwomostimportantparametersusedtocharacterizethesoundwaveformarefrequencyandamplitude.Frequencyisthereciprocalofperiodandrepresentsthenumberofperiodwithinonesecond.TheunitoffrequencyisHz.Thefrequencyrangeofspeechisfrom300Hzto3400Hz.Ontheotherhand,amplituderepresentstheloudnessofsound.

2.2DigitizationofSound

Asweknow,soundiscontinuousanalogsignal,andcannotbeprocessedbycomputerdirectly.Incomputer,soundisrepresentedbyaseriesofdigitalnumber.Wecalltheprocedureofconvertingtheanalogsoundwaveformintodiscretedigitalnumberasdigitizationofsound.Forthesakeofefficiency,thesedigitalnumbersshouldbeintegers.

Tofullydigitize,theanalogsoundsignalmustbesampledinbothtimedimensionandamplitudedimension,whichisreferredtoassamplingandquantizationrespectively,asdepictedinFig.1.Furthermore,filteringandencodingarealsoimportanttodigitizationofsound.Priortosampling,theanalogsoundsignalshouldbefilteredtoremovetheunwantedfrequencybyusingabandpassfilter.Afterquantization,encodingisperformedtoassignacodewordtoeachoutputlevelorsymbolofquantization,andabinarybitstreamisgenerated.Insummary,analogtodigitalconversion（ADC）systemcanbeillustratedbyFig.2asfollows.

Fig.1SamplingandQuantization

（a）Samplingtheanalogsignalinthetimedimension.

（b）Quantizationissamplingtheanalogsignalintheamplitudedimension.

Fig.2AnalogtoDigitalConversion（ADC）

2.2.1Sampling

Samplingmeansmeasuringtheamplitudeofthesoundwaveformatevenlyspacedtimeintervalstoproduceaseriesofvalues,andtherateatwhichitperformediscalledthesamplingfrequency.Foraudio,typicalsamplingratesarefrom8kHz（8000samplespersecond）to48kHz.Thehumanearcanhearfromabout20Hztoasmuchas20kHz;abovethislevel,weentertherangeofultrasound.Thehumanvoicecanreachapproximately4kHzandweneedtoboundoursamplingratefrombelowbyatleastdoublethisfrequency.Thuswearriveattheusefulrangeabout8to40orsokHz.

Inordertorecoverthesignalcorrectly,wemustuseasamplingrateequaltoatleasttwicethemaximumfrequencycontentinthesignal.ThisiscalledtheNyquistrate.TheNyquistTheoremisnamedafterHarryNyquist,afamousmathematicianwhoworkedatBellLabs.Moregenerally,ifasignalisband-limited—thatis,ifithasalowerlimitf1andanupperlimitf2offrequencycomponentsinthesignal,thenweneedasamplingrateofatleast2（f2-f1）.Thehigherthesamplingfrequencyis,thebettertheaudioqualityis,andthelargerstoragevolumeis.Typicalsamplingrateiseither44.1kHz,equivalenttoaboutCDquality,or8kHz,fortelephonequality.

2.2.1Quantization

Samplingintheamplitudeorvoltagedimensioniscalledquantization,anditincludesuniformquantizationandnon-uniformquantization.Typicaluniformquantizationratesare8-bitand16-bit;8-bitquantizationdividestheverticalaxisinto256levels,and16-bitdividesitinto65536levels.

Fordigitalsignals,onlyquantizedvaluesarestored.Foradigitalaudiosignal,theprecisionofeachsampleisdeterminedbythenumberofbitspersample.Asidefromanynoisethatmayhavebeenpresentintheoriginalanalogsignal,additionalerrorresultsfromquantization.Thatis,ifvoltagesareintherangeof0to1butwehaveonly8bitsinwhichtostorevalues,weeffectivelyforceallcontinuousvaluesofvoltageintoonly256differentvalues.Inevitably,thisintroducesaround-offerror.Althoughitisnotreally“noise”,itiscalledquantizationnoise,orquantizationerror.

Thequalityofthequantizationischaracterizedbythesignal-to-quantization-noiseratio（SINR）.Quantizationnoiseisdefinedasdifferencebetweenthevalueoftheanalogsignal,fortheparticularsamplingtime,andthenearestquantizationintervalvalue.Atmost,thiserrorcanbeasmuchashalfoftheinterval.

ForaquantizationaccuracyofNbitspersample,therangeofthedigitalsignalis-2N-1to2N-1-1.Thus,iftheactualanalogsignalisintherangeform-Vmaxto+Vmax,eachquantizationlevelrepresentsavoltageof2Vmax/2N,orVmax/2N-1.SQNRcanbesimplyexpressedintermsofthepeaksignal,whichismappedtothelevelVsignalofabout2N-1,andtheSQNRhasasdenominatorthemaximumVquan_noiseof1/2.TheratioofthetwoisasimpledefinitionoftheSQNR:

Ontheotherhand,ifweassumethattheinputsignalissinusoidal,thatquantizationerrorisstatisticallyindependent,andthatitsmagnitudeisuniformlydistributedbetween0andhalftheinterval,wecanshowthattheexpressionfortheSQNRbecomes

Sincelargerisbetter,thisshowsthatamorerealisticapproximationgivesabettercharacterizationnumberforthequalityofasystem.

Typicaldigitalaudiosampleprecisioniseither8bitspersample,equivalenttoabouttelephonequality,or16bits,forCDquality.Infact,12bitsorsowouldlikelydofineforadequatesoundreproduction.

2.2.2Encoding

Encodingmeansassigningacodewordtoeachoutputlevelorsymbolofquantization.Thiscouldbeafixed-lengthcodeoravariable-lengthcode,suchasHuffmancoding.Ingeneral,producingquantizedsampledoutputforaudioiscalledPulseCodeModulation,orPCM.ThedifferencesversioniscalledDPCM,andtheadaptiveversioniscalledADPCM.

InPCM,atthesender,aband-limitingfilterisusedtoremovethehighandverylowfrequencycontentfromtheanaloginputsignalfirstly,andthenperformanalog-to-digitalconversion,thusproducethePCMsignals,asillustratedinFig.3（a）.Atthereceiver,performdigital-to-analogconversionandthenconstructanoutputanalogsignalwhichisstaircaseshowninFig.3（b）.Thistypeofdiscontinuoussignalcontainsnotjustfrequencycomponentsduetotheoriginalsignalbut,becauseofthesharpcorners,alsoatheoreticallyinfinitesetofhigher-frequencycomponents.Therefore,theoutputofthedigital-to-analogconverterisinturnpassedtoalow-passfilter,thustheoutputbecomingsmoothed,asFig.3（c）shows.

Fig.3PulseCodeModulation（PCM）

（a）OriginalanalogsignalanditscorrespondingPCMsignals.（b）Decodedstaircasesignal.

（c）Reconstructedsignalafterlow-passfiltering

Generally,ifatime-dependentsignalhassomeconsistencyovertime（temporalredundancy）,thedifferencesignal—subtractingthecurrentsamplefromthepreviousone—willhaveamorepeakedhistogram,withamaximumaroundzero.Consequently,ifwethengoontoassignbit-stringcode-wordstodifferences,wecanassignshortcodestoprevalentvaluesandlongcode-wordstorarelyoccurringones.Predictivecodingsimplymeanstransmittingdifferences.Wepredictthenextsampleasbeingequaltothecurrentsampleandsendnotthesampleitselfbuttheerrorinvolvedinmakingthisassumption.DPCMisexactlythesameaspredictivecoding,exceptthatitincorporatesaquantizerstep.Furthermore,adaptiveDPCMtakestheideaofadaptingthecodertosuittheinputmuchfurther.ADPCMcanadaptivelymodifythequantizer,bychangingthestepsizeaswellasdecisionboundariesinanon-uniformquantizer.

3.Steps

●RecordandEditAudio;

1.Installthesoftwareincomputer（chooseapropertyfolder）andthenclickthe“AudacityPortable.exe”intheinstallfoldertorunthesoftware.

2.Makesurethatthemicrophoneworkswell,andthenclick

torecordasectionofaudiofile.Click

tostoprecording.

3.Click

toplaytheaudio,ifyoucannothearitclear,clickpop-upmenu“Amplify”onthe“Effect”menutoamplifythevoice.

4.NoticethattheaudiogeneratedbyrecordingisMono,pleaserecordtwoormoreaudiofiles.

5.ChoosetwoMonoTracksandmergethemintoaStereoone（Fig.4）.OpenanaudiofileandthensplittheStereoTrackintotwoMonoTracks（Fig.5）.

Fig.4MaketwoMonoTracksintoaStereoone

Fig.5SplitaStereoTrackintotwoMonoTracks

6.Loadastereoaudioclipandtrytoalignstereotracks,sothatthelefttrackislaterthantherightonewith3seconds.Describethestepsandtheresults.

7.Chooseasectionoftheaudio,andthenClickpop-upmenu“Cut/Copy/Paste…”onthe“Edit”menuor

toeditthefileasyouwish.Press“Ctrl+Z/Ctrl+Y”orclick

toundo/redoyoulastoperation.

8.Clickpop-upmenu“Duplicate”onthemenu“Edit”or“Ctrl+D”tomakeacopyofthechosenfile.Clickpop-upmenuonthe“Effect”menutoaddeffecttothecopyfile.

9.Clicksubmenuofthe“Generate”menutoedityouraudiofile.Youcanaddsomeeffecttothefile.Analyzethedifferencebetweenatleastthreeeffects.

●SamplingandQuantization

1.Clickpop-upmenu“Open”onthe“File”menutoopenaStereo/MonoTrack,alsoyoucanoperateonyourownfile.

2.Clickpop-upmenu“Duplicate”onthe“Edit”menuor