ECTE401ZZUFinal Exam Solutions.docx
- 文档编号:4337738
- 上传时间:2023-05-07
- 格式:DOCX
- 页数:13
- 大小:63.22KB
ECTE401ZZUFinal Exam Solutions.docx
《ECTE401ZZUFinal Exam Solutions.docx》由会员分享,可在线阅读,更多相关《ECTE401ZZUFinal Exam Solutions.docx(13页珍藏版)》请在冰点文库上搜索。
ECTE401ZZUFinalExamSolutions
UNIVERSITYOFWOLLONGONG
SCHOOLOFELECTRICAL,COMPUTERANDTELECOMMUNICATIONSENGINEERING
ECTE401MULTIMEDIASIGNALPROCESSING
FINALEXAMINATION,2007
THISPAPERISFORSTUDENTSSTUDYINGATZHENGZHOUUNIVERSITY
TimeAllowed:
THREE(3)hoursand
FIFTEEN(15)minutes
NumberofQuestions:
PARTA:
SIX(6)-Eachquestion
mayhavemultipleparts
PARTB:
SIX(6)-Eachquestion
mayhavemultipleparts
DIRECTIONSTOALLCANDIDATES:
AttemptALLquestions.Marksallocatedforrespectivequestionsareshown.Allocatetimeaccordingly.
PARTA–Usetheanswerbookprovided–begineachquestiononanewpage.
PARTB–Usetheanswerbookprovided–begineachquestiononanewpage.
Note:
MakesureyouranswersareCLEARandREADABLE.
CalculatorsPermitted:
ThenormalSchoolPolicyisthatONLYthecalculatorsapprovedbytheBoardofStudiesareapprovedforuseinexaminations.(http:
//www.boardofstudies.nsw.edu.au/manuals/calculators_hsc04.html)
DrawingInstrumentsarepermitted.
CANdidatesshouldnotethatquestionsaretobeansweredaswritten-noconsultation(individualorgroup)onquestionswillbegiven.Anyassumptionsmadeshouldberecordedwithyouranswer.
EXAMINATIONPAPERSMUSTBEWRITTENININK.PAPERSWRITTENINPENCILWILLNOTBEMARKED.
NOTE:
TheexaminationpaperisprintedonONEsideONLY.
**ALLEXAMPAPERSANDANSWERBOOKLETSMUSTBERETURNEDATTHEENDOFTHEEXAM**
PARTA:
ShortAnswerQuestions[50marks]-USETHEEXAMINATIONBOOKLETPROVIDEDFORYOURANSWERS.
AllPartAquestionsmustbeansweredinthespaceprovidedonthisexaminationpaper.
QUESTION1A[10marks]
a)Brieflyexplaintheadvantageoflogarithmicquantisationwhenappliedtospeechsignals?
(1mark)
Answer:
Logarithmicquantisationprovideslowerquantisationerrorsforlowamplitudevalues,whicharecommoninspeech.
b)Whatistheadvantageofbackwardadaptivequantisationcomparedwithforwardadaptivequantisation?
(1mark)
Answer:
Backwardadaptivequantisationdoesnotrequiretransmissionofthequantisationstepandhaslowerdelayasdoesnotrequirelookahead.
c)Asourcehasvaluesof1to21.Designa3bituniformquantiserforthissourceandwritedownthequantisationtable.Findtheerrorresultingfromquantisationofthevalue5usingthisquantiser.(3marks)
Answer:
Δ=(21-1)/8=2.5
Quantisationlevels=1+Δ/2=2.25,4.75,7.25,9.75,12.25,14.75,17.25,19.75
Eq(5)=5-4.75=0.25
d)DrawablockdiagramshowingthebasicstagesofanAnalysisbySynthesisspeechcoderincorporatingbothlongtermandshorttermprediction.Giveabriefdescriptionofeachblock.(5marks)
Answer:
Theexcitationgeneratorisresponsibleforproducingarepresentationoftheexcitation.ThisisthenfilteredbytheinverseLongTermPredictor(LTP)filter,whichgeneratesaresidualwithperiodicpulsesduringvoicedspeech.TheresultingresidualisthenfilteredbyaninverseLinearPredictor(LP)filtertoproduceasynthesisedspeechsignalŝ(n).Theerrorminimizationblockaimstominimisetheerrorbetweenthesynthesisedandoriginalspeechsignals.
QUESTION2A[10marks]
a)Describethedifferencebetweenlosslessandlossycompression.Givetwoexamplesofsignalswherelossycompressionisinappropriate.(2marks)
Answer:
Losslesscompressionperfectlyreconstructstheoriginalsignalwhilelossycompressionproducesanaccuraterepresentationoftheoriginalsignal.Lossycompressionisappropriateforspeechsignalsandaudiosignals.
b)Table1AliststheprobabilitiesofoccurrenceofthesymbolsAtoE.FindHuffmancodestorepresenteachofthesesymbols.
(4marks)
Symbol
Probability
A
0.16
B
0.29
C
0.05
D
0.36
E
0.14
Table1A
Answer:
D:
1
B:
01
A:
001
E:
0001
C:
0000
c)AssumewedesignaHuffmancoderforanalphabetof5symbolswithprobabilitiesandcodewordsgiveninTable2A.Whatistheaveragecodewordlength?
Whatistheresultingcompressionrate?
(4marks)
Symbol
Probability
Codeword
A
0.78
1
B
0.1
10
C
0.05
100
D
0.04
1000
E
0.03
0000
Table2A
Answer:
Averagecodewordlength=1*0.78+2*0.1+3*0.05+4*0.04+4*0.03=1.41bits/symbol
Compressionrate=1.41/3=0.47or47%compression
QUESTION3A[10marks]
a)Describethesourcefiltermodelofspeechproduction.Howis
thismodelusedinspeechcoding?
(3marks)
Answer:
Thesourcefiltermodelconsistsoffilterrepresentingthevocaltract,theinputofwhichiseitheraperiodicpulsetrain(forvoicedspeech)orrandomnoise(forunvoicedspeech),appropriatelygainscaled.Thesourceischosenusingaswitch.Thismodelisusedsynthesisespeechandmodelspeechproductionusingasetoffiltercoefficients,derivedbyLinearPrediction(LP)andarandomorperiodicsource,withtheperiodequaltothepitch.
b)Withregardtohumanhearing,definesimultaneousmasking.
(2marks)
Answer:
Thisiswhereatoneofgivenfrequencyismasked(notheard)byanothertoneorsetoftonesoccurringinthesamefrequencyregionasthesingletoneandatthesametime.
c)Brieflydescribewhichtypeofspeechsignalsaremostsuitedtolinearpredictionandexplainwhy.(1mark)
Voicedspeechsignalsaremostsuitedduetothepresenceofshorttermcorrelations.
d)Howistheperformanceofspeechandaudiocodersevaluated?
WhyisSNRinappropriateinspeechcoding?
Whyaresubjectivemeasuresusedalmostexclusivelyduringstandardizationactivities?
Giveanexampleofasubjectivemeasure.(4marks)
Answer:
Theperformancecanbeevaluatedusingsubjectiveandobjectivemeasures.Objectivemeasuresdeterminethemathematicalerrorbetweentheoriginalandsynthesisedspeechsignals.E.g.MSE,SNR,SegmentalSNR.Subjectivemeasuresdeterminetheperceptualerrorbetweentheoriginalandsynthesisedspeechsignalsandrequirehumanlisteningtests.E.g.MOS,A/Bcomparisontests,DiagnosticAcceptabilityTests.Subjectivemeasures,suchastheMOStest,areusedalmostexclusivelyduringstandardisationasthesegivethemostreliableanduniversallyacceptablemeasureofspeechquality.Thisisbecausethereisnofixedcorrelationbetweenobjectiveandsubjectiveerror–alargemathematicalerrorsuchasmeasuredwiththeSNRmaynotresultinaperceivableerror,whileasmallmathematicalerrorinacriticalregion(e.g.alowenergyfrequencyband)mayresultinalargeperceivableerror.
QUESTION4A[10marks]
a)ThemethodsusedtorepresenttheLPexcitation(residual)signalinspeechcoderscanbegroupedinto3broadmethods.Namethese3methodsandbrieflydescribetheoperationofeach.
(5marks)
Answer:
Direct–Directly_inimizeeachsampleusinge.g.uniform_inimized_d_e.g.G721usesAdaptiveDifferentialPulseCodeModulation(ADPCM).
Analysis-by-Synthesis–Quantiseeachsample,orsetofsamples,suchthattheerrorbetweentheoriginaland_inimized_dspeechsignalis_inimized.E.g.CELP.
Parametric–Modellingtheexcitationusinge.g.pulsesforvoicedspeechandrandomnoiseforunvoicedspeeche.g.FS1015speechcoder.
b)Whatisthepitchperiodofaspeechsignal?
Explainintermsofboththesignalandtheproductionmechanisms.(5marks)
Answer:
Thepitchperiodofaspeechsignalcorrespondstothemeasuredperiodofthequasi-periodicvoicedsectionsofaspeechsignal.Theperiodcorrespondstothepeaksinthetimedomainsignalandaretheinverseofthefundamentalperiod.Thisperiodicnatureiscausedbytheperiodicopeningandclosingofthevocalfoldsinthevocaltract
QUESTION5A[10marks]
a)MP3isbasedontheperceptualcodingparadigm.Brieflyexplainhowaudiocodersoperatingwithinthisparadigmattempttoachievehighaudioquality.(3marks)
Answer:
Audiocodersoperateinthisparadigmbyusingapsychoacousticmodelofthemaskingcurveoftheaudiospectrum.Themostperceptuallyimportantcomponents(thoseabovethemaskingcurve)arerepresentedmoreaccuratelythanthemaskedcomponents.
b)Drawablockdiagramshowingthemainstepsperformedinaperceptualaudiocoderandbrieflyexplaineachblock.(5marks)
Answer:
Analysisfilterbank:
Separatetheinputintofrequencybands(alsocalledspectralcomponentsorsub-bands)
Perceptualmodel:
Calculateapsychoacousticmodelfortheinputspectrum.Thisallowsthemaskingthresholdtobecalculated
Quantisationandcoding:
Quantisethespectralcomponentssuchthatthequantisationnoise(duetothequantisationerror)fallsbelowthemaskingthresholdforthefrequencyofthespectralcomponent
Encodingofbitstream:
Useentropycodingtofurtherreducethebitrate.
c)WhataretwokeydifferencesbetweenMP3andAACaudiocoding?
(2marks)
Answer:
Anytwoofthefollowing:
∙Afinerfrequencyresolutionfilterbankforfrequencydomaindecomposition.
∙TemporalNoiseShaping(TNS)
∙Accountsfortemporalmasking
∙Backward-adaptivelinearprediction
∙Jointstereocoding
∙Alargerangeofsamplingrates,bitratesandchannelnumbers
PARTB:
ShortAnswerQuestions[50marks]-USETHEEXAMINATIONBOOKLETPROVIDEDFORYOURANSWERS.
QUESTION1B[10marks]
(a)Brieflyexplainthedifferencebetweenagreyscaledigitalimageandacolourdigitalimage.(2marks)
Greyscaleimageisblackandwhite.Onlyonecolourcomponentperpixelwitharangeofvaluesfrom0to255.
Colourimageshavemultiplecomponentsperpixel,representingRed,GreenandBlue.
(b)Whatisthepurposeoftransforminginimagecompression?
(2marks)
Transformingaimstocompacttheenergyoftheimageintoafewsignificantcoefficients.Thisallowsforefficientcoding.
(c)Whatcriteriashouldbeusedwhenchoosingtheresolution(i.e.numberofpixels)foradigitalimage?
(2marks)
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- ECTE401ZZUFinal Exam Solutions