title multivariate statistical analysis of speech ….../rl /1 cv cvvv vcv cvcv \(t) speechsound ・...
TRANSCRIPT
![Page 1: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/1.jpg)
Title Multivariate Statistical Analysis of Speech Sounds(Dissertation_全文 )
Author(s) Tabata, Koh-ichi
Citation Kyoto University (京都大学)
Issue Date 1973-11-24
URL https://doi.org/10.14989/doctor.r2408
Right
Type Thesis or Dissertation
Textversion author
Kyoto University
![Page 2: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/2.jpg)
」χ4f
MULTIVARIATE STATISTICAL ANALYSIS
OF
SPEECH SOUNDS
KOH-ICHI TABATA
Department of Information Science
Kyoto University
March,1973
![Page 3: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/3.jpg)
-・
'1゛"
ヒ1‐乙
MULTIVARIATESTATISTICALANALYSIS
OF
SPEECHSOUNDS
DOC
1973
10
電気系
KOH-ICHITABATA
`DepartmentofInformationScience
KyotoUniversity
March,1973
![Page 4: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/4.jpg)
PREFACE
VJhenweexplainphenomenainthenatureand,furtニher,tryto
establishlawswhichgovernthem,itisnecessaryasoneofthe
guidingprinciplestoworkaccordingtoapattern
1-
observation,
qualitativeexplanation,modelestablishmentandthenquantitative
explanation.However,itisinsomeaspectsdoubtニfulwhetherwe
canbringlighttocharacteristicsofspeechsoundonlybythat
patニtニern.Ideallywemayhavetowaituntil"Advancedinformation
science"becomesdeveloped.Underpresentconditionsitseemsthatニ
wehavenochoicebuttoproceedwitニhthestudyofspeechsounds
accordingtotheguidingprinciplementionedabove。
Now,thephysicalcharacteristicsofvoiceofeachindividual
aredifferentfromthatofeveryother,andwecannotexpectthat
thephysicalaspectoftwoutterancesofthesamewordbyonespeaker
isidenticalevenunderthesamecircumstances。
Therefore,ifwearegoingtofindacertaincharacteristic,
wehavetoconfirmitqualitative:lyorquantニitativelyonthebasis
ofsufficientdatainordertoassertthatthediscovery1Snever
accidental.
Thisresearchwasperformedfromsuchastandpoint
lappliedmultivariatestatisticalanalysistospeechspectraobtain-
edbyphysicalobservation,andthenclarifyvariouscharacteristics
ofspeechsoundonthestatisticalfoundation。
Tobebrief,themethodofmultivariatestatisticalanalysis
isconcernedwithmulti-dimensionalvectorwhosecomponentsarenotニ
independentofeachother.Ifweassumethatcomponentsofspeech
spectrumarethoseofmulti-dimensionalvector,itニbecomesauseful
meansforspeechanalysis.Althoughthevalueofmultivariate
analysisofvarianceaboutpracticalapplicationshasbeenneglected
despit二ethemathematicalresults.ofitstheory,Itwasveryuseful
inthisstudv.Thisfactrevealsthatthepracticalvaluewasfor
thefirsttimecertifiedhere.Further,thisstudycontributesto
![Page 5: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/5.jpg)
thetheoryofmultivariatestatisticalanalysisaswellasits
practica:Lusage。
Characteristicsofspeechsoundsclearedherearemainlyconcern-
ingco-articulation,individualdifferencesofspeakersandcharactニer-
isticsofconsonants,eachofwhichhasbeenunknownuptonowand
isinterestingtospeechscience。
Formantfrequencyhastraditionallybeenusedinthiskindof
research.Thisthesisgivesusacluewhichwillreleaseusfrom
thetraditionalresearchmethod,whichhasgovernedwitニhaniron
handandconfineduswithinsomelimits。
Atニtheendofpreface,wewillinsist,"Thisstudyhasbeen
developedasaconsequenceofseriousref:Lectiononthetraditional
wayofresearchinwhichresearcherswereinahurrytotryspeech
recognitionwithouteχaminingsufficientlybasicacoustic-character-
isticssuchasco-articulation,individualdifferencesofspeakers,
etc..
Weusedacomputertニhroughouttheprocess fromthe
collectionofdataofvoicetotheinvestigationofanalvzedresults.
Ifwedidnothaveacomputer,thisstudycouldnotbeaccom-
plished."
-2-
Koh-ichiTabata
March,1973
![Page 6: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/6.jpg)
PREFACE
GLOSSARYOF
CHAPTER:L
2.3
CHAPTER
2.6
3
3.1
3.2
3
3
3
3
4
4
3
4
5
6
1
2
CONTENTS
1<£>
288
111
20
20
20
21
22
25
27
27
29
33
34
37
37
3
4
4
4
8
0
4
5
46
50
50
工nt
2.2
2.2.1
2.2.2
2.2.3
2.2.4
2.4.1
2.4.2
CHAPTER2MultivariateStatisticalAnaly
2.1Introduct
ofVari
MultivariateDistribut
Introduct
NormalMultivariateDistribution・・・・・・・・・・・・・丿・・・・・・・
ProbabilityEllipsoid‥‥‥‥
ConcentrationEllipso
SamplesfromtheMultinormalPopulation‥‥‥‥‥
MultivariateLinearRegression‥‥‥ ● ● ● ● ● ●
2。4MultivariateAnalysisofVariance‥・‥‥‥・‥‥・‥‥‥‥・‥・
TheLikelihoodRatio
2.4.3MethodofMultivariateAnalysisofVariance…30
2.5TwoProposalsforMultivariateStatisticalAnalysis
Principal-ComponentAnaly
Facilit:iesforSpeechAnalysisandSpeechSpectrum
NonreverberantRoom,Microphone,Taperecorderand
Pre-emphasis
Observat
1/4-OctaveFilterAnalyze
SpeechSpectra…‥
On-line,RealTimeDataInputSystニem
SegmentationofConnectedSpeechbyVisual
CHAPTER4AnalysisofJapanesevcvUtterancesofFiveMale
工ntroduct
EstablishmentofExperimentalObjectsandFactors51
-3-
![Page 7: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/7.jpg)
4
4
3
4
4.5
4
4
4
4
CHAPTER5
CHAPTER
6
7
8
9
lCN
inin
5.3
5.4
5.5
5.6
5.7
6
6.1
6.2
6
6
3
4
6.6
700
66
LinearModelandMultivariateAnalysisofVariance53
RelationbetweenNormalizedCriterionり'and
DiscriminationScore …………………………56
GeometニricRepresentationofMulti-factor
Distribut
]:ntroduct
Conc:Lus
Introd
5
6
6
6
9
3
6
8
70
72
74
75
n4″り
詣88
89
93
95
96
102
104
105
108
111
ComparisonWitニhPrincipal-ComponentAnalysis・・・・・・
Discussionaboutニ
AnalysisofVar
4.10Conclusion
AnalysisbyFormantFrequency
RegressionEstimatebyUsingFinalVowelsand
AnalysisofJapanesecvcvUtterances‥‥‥‥‥‥‥,‥‥‥‥‥74
MultivariatニeAnalysisofVarianceforFour-Factor
DesignwitニhRepeatedMeasurements
EstablishmentofExperimentalObjectsandFactors82
ResultofAnalysisofVariancefor・Four-FactorDes
Examinationof
EvaluationofInteract
AnalysisofJapanesecvandvvUtterances,・・・・・・‥‥‥‥95
MultivariateAnalysisofVariancesforDivided-Type
Two-FactorDesign‥‥‥‥
Two-FactorDes
Divided-Type
Discuss
Conclus
ResultofAnalysisofVarianceforDivided-Type
6.5RelationbetweenNormalizedCriterionり'and
DiscriminationScore
GeometricRepresentationofAnalysisofVariancefor
-4-
![Page 8: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/8.jpg)
CHAPTER7
7.1
7.2
7.3
7.4
CHAPTER8
ACKNOWLEDG
REF
Three-DimensionalRepresentationofJapanesePhonemes113
工nt113
114
116
118
122
128
131
135
137
138
139
146
147
156
165
cvcvUtterancesandtheirSpe
ConsonantSpe
フ.6Conclus
Conclusi
ClassificationfromtニheViewpointニofMannerand
PlaceofArticulat
フ.5ExpressioninThree-DimensionalSpace
OTHERREFERENCES
THEAUTHOR'SPAPERSRELATEDTOTH工STHES工S
APPENDIXATheAsymptoticDistributionoftheLikelihoodRatio
TestニCriterioninMultivariateAnalisisofVariance
forFour-FactorDesignwithSingleObServation.‥‥‥141
APPEND工XBTheIntersectionProducedbytheEllipsoidandthe
APPEND工XC
APPEND]:XD
APPEND]:XE
APPEND工XF
StraightLineorthe
VarianceofFactor
Measur
TheVectorSpaceNormalizedbytheResidualR.‥‥‥149
TheKolmogorov-SmirnovOne-SampleTest.・・・・・・・・・・・・・・・・・・150
TheAsymptoticDistributionoftheLikelihoodRatio
TestCriterioninMultivariateAnalysisofVariance
forFour-FactorDesignwithRepeatedMeasurementsin
whichA11InteractニionsAreDiSregarded.‥‥‥‥‥‥‥‥‥152
APPENDIXGTheAsymptoticDistributionoftheLikelihoodRatニioTest
CriterioninMultivariateAnalysisofVariancefor
Divided-typeTwo-FactorDesignwithRepeated
APPENDIXHMatrixofSumsofSquaresandCrossProducts
inSubs
-5-
![Page 9: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/9.jpg)
/1
/rL
CV
VV
CV
vcv
cvcv
\ (t)
Speechsound
・●
●
●
●
●
●●・一@一
・一
@一
Matrix
A(pxq)
A(pxq+1)
A゛
A‾1
国
tr(A)
rank(A)
lp
j'
・す
一1
・・
●
●
●
●
●一●一
●●
・
●
GLOSSARYOFSYIIBOLS
Phonemicsymbol.
Phoneticsymbol.
Consonant.
Vowel.
Phonemesequence
Phonemesequence
Phonemesequence
一・
一一@一
Phonemesequence:
Consonant-Vowel.
Vowel-Vowel.
Vowel-Consonant-Vowel.
ConsonantニーVowe:L-Consonant
-Vowel.
Thei-thoutputof20-channelfilteranalyzer.
pxqmatrixA.
px(q+l)matrixA.
ThetransposeofmatrixA.
TheinverseofmatrixA.'‘
ThedeterminantofmatrixA.
ThetraceofmatrixA.
TherankofmatrixA.
p-dimensionalunitmatrix.
LetR(pxp)beasjnnmetricpositivedefinitematrix,
thentheorthogona:LmatrixTeχistssuchthat
λ0
F~LL
T=R
-6-
T
j
p
oλ
![Page 10: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/10.jpg)
f±
0
.X°
Z
へy
ぷ
一Se・
`゛here入1>・・・,λ (>O)aretheeigenvaluesofR-・・〃
Define
汗
as
p
follows:
jo呵
瓜o
lけ
まiSalsoasymmetricpositivedefinitematrix,and
lによ.診.
R‾亀(Ry)‾1
Zeromatrix
Randomvector,probabilityanddistribution
e一
S●
e●
Pr{E}
pHeIh}
e(x)
μA
N(m,�)
N(μ,Λ)
0
X2
・S・S
X(IXp)=(xi,X2,x;:p-dimensionalrandomvector
,・(pxl)りFI
謡光二竃二謡二三鶯三二ニ
●S
一e
・一
ExpectedvalueofZ
Expectedvalueofぷ
・●一
Covariancematrixof、z:
μ=e(x)
Λ=e(x-μ)'(r-μ)
Theunivariatenormaldistributionwithexpected
valuemandvariancea^.
Thep-dimensionalnormaldistributionwith
expectedvalueμandcovariancematriχΛ.
Zerovector。
Chi-squaredistribution。
-7-
![Page 11: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/11.jpg)
Ap
W(Λ,P,r)
W
X
e(X)
χ.
Zi●
χ..
Q
一一
S・
・S
Chi-squaredistribut:ionwithp?egreesoffreedom.
Wishartdistribution.TheSumiEllji゛Uいwhere
p-dimensionalvectニorui'sareindependently
distributedaccordingtoN(0,Λ),hasW(A,p,r)
distribution. riscalledthedegreesoffreedom.
Theleft-handsidevariateisdistributedaccording
totheright-handsidedistribution.
Forexample, z~N(0,Λ)
Samplesfromthepopulation
e一
一一
・・
一一
S一
●・
]』
一一
Iχ・~一χ ●
一幽
χ一@・
●
ip
X np
j
●
● Thedatamatrix.
wherexlj°゛'゛xnaretherealizatニionsofp-dimensional
randomvector.
Expectedvalueofx一・ e(x)=
』
ThemeanvectorofXi(i=l,・‥,n)
Themeanvectorofぶり(i=l,--゜゛a・≪\
e(X,)
・I・
eぐ4)
一・ Z=
j
―一n
j°:L,‥
1
correspondingtothesuffixj:Xj.= -b
ぶ
,b)
bΣ一一一
j
Themeanvectorof・≪^ij(i°19¨゜9a;j°1゛'゜゜゛
1abz..=冨ふぶり
りj
SS
jb
Totalvariance.(Matrixofsumsofsquaresand
crossproducts.)Forexample,
ab
1=1j
EI(゛ij
゛‥)(゛ij-゛・.)
-8-
![Page 12: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/12.jpg)
●
S
〈μ〈Λ〈B
《〈B
〈〈A
〈〈μ
A
AB
A.
Q
Oj
"Qi
R
~R
・9
wherezij(i°111°・,a;卜1,・・・,b)arethesamplevectors.
Thesamplecovariancematrix.Forexample,
S°
lab
゛..)'(x,,-゛¨)'
wherezij(i°1゛'゜゜9a;j°19・・,b)arethesamplevector.
ThemaximumlikelihoodeStニimates
el
S●●Sle
Themaximumlikelihoodestimateofexpectedvalueμ.
ThemaximumlikelihoodestimateofcovariancematrixΛ.
ThemaximumlikelihoodestimateofparametermatrixB.
Themaximumlikelihoodestimateofμ,A,Bunderthe
nullhypothesisH.
Analysisofvariance
Se
S・S9・・
●一
一・
一一
・S
FactorA.
工nteractionfactorbetweenfactorAandfactorB.
Thei-thtreatment(level)offactorA.
Totalvariance.(Matrixofsumsofsquaresand
crossproducts.)
Thei-thvarianceobtainedbythebreakdownofQ.
Thei-thvariancenormalizedbytheresidualR.
-Q-°R焉iR斗9wherenisthesamplesize.
Residualvariance.(Matrixofsumsofsquaresand
crossproducts.)
n
士RR‾1°I,wherenisthesamplesize.
-9-
![Page 13: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/13.jpg)
μ
αi
'"ij
"ijk
BZ
Ha
L(∂)
(Mω
LL
入
l£J
り
りi
β
j
・一
@一
一・
ゆ・
S●
一●
Thegenerallevelvector;
ThemaineffectvectorscorrespondingtoAi,Bj
treatments(levels),respectively;
Theinteractionvectorcorrespondi叱to(Ai・Bj)
treatment,and;
Theresidualvector・;
inthefo:Llowinglinearmodel:
*ijk°μ十αi十βj十rk十りjk
Parametermatrix(unknown).
Designmatrixofexperiment,orMatrixofknown
vectorszainregressiontheory・
Testofhypothesis
一@
●●
一一
一●
e・
II
●一
・一
AnullhypothesisforfactorA.
Alikelihoodfunctionwithm-dimensiona1
parameterd.
Ln=mL(0
Lω゜;;2:L(∂),whereωiSthesubspaceofΩ
入=-μ:LikelihoodratiocritニerionfortheLΩ
hypothesisHO:∂.丘ωagainstthealternativeH:∂e,Ω-ω
ZZJ・こ 鴎帚:Likelihoodratiocriterionwhichisa
monotonicfunctニionofλ,wherenisthesamplesize・
LikelihoodratiocriterionbyBox,whichisdistributed
asaChi-squaredvariateasthesamplesizetend
toinfinity・
NormalizedcriterionofV:
V'=
. ¶
(Valueof1Ssignificantlevel0tIχ I、testcorre-
spondingtothedegreesoffreedomofiり)
-10-
![Page 14: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/14.jpg)
兌1
聡十聡
そ
{刈 H}
・e一
Others
・・一・
聡=rank(Qi)
刻十ii2=n-rank(R),wherenisthesamplesize.
isamemberof-
Asetofz'swhichsatニisfytheconditionH
-11
![Page 15: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/15.jpg)
CHAPTER1
]:ntroduction
Thespeechsoundisoneoftherepresentativemeansofinfor-
mationcommunication.Amongthevariouskindsofinformationconvey-
edbyit,thereareindividualityandemotionalexpression,andso
on.Themostnotableofallisphonemicinformation.Itseemsquite
asimp:Lephenomenonthatwecancommunicatevocallywitheachother
becausewehavebeenaccustomedtotalkingand:Listeningsincechild-
hood.Physically,however,itisamiracle.Thestニructureofthe
vocalorgansofeachspeakerisdistinctニfromthatofothers,and
nobodycanuttertheverysamevoiceintニhephysicalsense.Further-
more,thecharacteristicsofacousticsystemorelectro-acoustic
systemwhichinterlinksthespeakersandtheauditorynervesof
listnersareremarkablydifferentaccordingtotheirenvironment・
Consideringthatthevoicesofvariouspersonsareidentifiedexactly
thesamephonemedespitetheirbeingphysica:Llymu:Ltlfarious,(that
is,theyarealldifferentphysically),itSe゛゛lsimpossibleto
explainvoiceinformationinaphysicalfield.Thismayleadusto
believethatitisgovernedbyanabsolutelydifferentfield.If
weassumethisdifferentfieldtobethe"Informationscientific
field,"wemayhavetowaituntil"Advancedinformationscience"
becomesdevelopedinordertofullycomprehendit。
Inthepresentsituation,thereisnootheraltニernativebutthe
physicalobservationofphenomenainordertoexplainandtoestablish
thelawswhichgovernthem.Itisorthodox,butwouldbehopeful
totrytoapproachtheessenceoftheproblembyrepeatingthe
cycle-observation,qua:Litativeexplanation.modelestablishment
andthenquantitativeexplanation。
Atfirst.letamulti-dimensionalvectorcorrespondtospectrum,
-12-
![Page 16: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/16.jpg)
atdesiredtimepoint,obtainedfromspeechsignalsthroughphysical
observation.工notherwords,frequencycomponentsofspectrumare
consideredascomponentsofmulti-dimensionalvector.Postulating
varioustypesofmodelsbasedonthespectrumvectorsthusobtained.
wecantrytoexplainthecharacteristicsofvoicequalitativelyand
quantitatively。
Thecharacteristicsofsoundhaveessentiallyanaspectrelated
toprobabilitysinceavoiceappearsinmultifariouswaysgoverned
bychanceasdescribedabove.Therefore,tニhemodelisessentially
aprobabilityone。
Deviationsbetweentheactualdataandthispresumedmodelare
duetoprobabilitychancesaswellasimperfectionsinthemodel。
I'Thatismostニimportanthereisthattointroducethemodel
doesn'tmeanImmediatelytoconfirmthecharacteristicswhichweare
attemptingtoexplainwithit.Toconfirmthem,athoroughexamina-
tニionisneeded.Unlessthequantity,whichrepresentsacertニain
characteristicexplainedbytニhemodel,issufficientlylargerthan
theresidualerror,wecannotprovetheexistニenceofthecharacter-
istic.Besides,wecannotニassertthevalidityofthepresumed
modelitself.Toguarantニeeit,theintroductionofstatistical
procedurestestofsignificancybecomesapressingproblem。
Mu:Ltivariatestatistニicalanalysiswasadoptedforthisstudyas
ausefullanalyticalmethodwhichsatisfiesthisrequirement.
Briefly,multivariatestatisticalanalysis1Sastatisticalmethod
dealingwithmulti-dimensionalvectorwhosecomponentニSarenot
independentofeachother.Wecanschemeconsiderablyfreelythe
formationsofmodelsbasedonthevector,andgivetニhemmathematical
basesnecessaryforthetestofsignificancy。
Thespeechspectrumisoneofimportantmeansforexpressing
speechsignals.Almostphysicalinformationofspeechsignals1S
reflectedinit.Ascharacteristicsofthefilteranalyzer(whichare
representedby"Windowfunction"1naviewpointof"Short-time
Fourieranalysis")bearontheobtainedspectrumcomponents,it1Sa
-13-
![Page 17: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/17.jpg)
problemwhatcharacteristicsshouldbeadopted.Nevertheless,
spectrapreservemuchmorecontentsofprimarysignalsthanformant
frequenciesd0.Wecananalyze,infact,onlyvowelsbyusingformant
frequencies,whilebyusingspectrawecananalyzebothvowelsand
consonants。
Evidently,therespectivefrequencycomponentsofthespeech
spectrumarenotindependentofeachother.Therefore,thecomponents
ofthemulti-dimensionalvector,definedbytheabovecorrespondence,
arealsonotindependent.Todealwithaprobabilityeventwhich
hasthissortofvectorasarealization,multivariatestatistical
analysisisasuitablemethod。
Thoughmultivariateanalysisofvarianceappliedtニothisstudy
hasbeendevelopedinafieldofmultivariatestatisticalanalysis,
ithasbeenapttobeundervaluedaboutpracticalapplicationsin
spiteofthemathematicalresultsofitstheory。
However,thismethodwasrecentlyfoundtobeextremelyusefu:L,
sincewecouldclarifyconsiderably,bythismethod,patternsofCO-
articulationandofdifferencesbetweenindividua:Lspeakerswhich
werenotunderstoodquantitativelysofar.Thismeansthatthe
practicalworthofthemethodofmultivariateanalysisofvariance
was,forthefirsttime,confirmedhere。
Wedesignedvariousmodelsandexperimentstoclarifyproperties
ofspeechsounds.Tointroduceandproveexpressionsofthemodels
ofmultivariateanalysisofvariance,nolesslaborwasspent。
Meanwhile,twomatters,intheoryaswellas'inapplication,
werefoundtocontributetomultivariateanalysisofvariance:
(1)Notioaofdirection,besidesnotionofamountindicatedgenerally
bytheanalysisofvariance,wasnecessarytounderstandtheeffects
oftherespectivefactorsofanalysisofvariance,andoneofits
techniquesofexpressionwasdescribed;
(2)amodel0ftheanalysisofvariancefor"Divided-Type"modelwas
proposed.Thisisanup-to-datemodelinwhichwedevelopananalysis
ofvariancefortwo-factordesignbydividingoneoftwofactorsinto
14-
![Page 18: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/18.jpg)
morethantwoportions.
Thismodelwasintroducedtodoresearchonpatternsofsome
tニypesofco-articulations,andthenformulated.
Weshouldnowexplainthestandpointofthisstudyfroma
viewpointofspeechscience.
Amongthestatisticalcharacteristicscomparativelywellstudied
uptothistime,therearedistributionofspeechamplitudelevel,
distributionoflong-timespectrum,on-offrateofspeech,and
fundamentニalfrequency.Butmostofthesecharacteristicsareconcern-
edwitニhwhatiscalled"acousticcharacteristics."Therearesome
statisticaldataonthelinguisticcontentsofspeechconcerning
vowels.butfewconcernedwithconsonants.Especially,quantitative
explanationsonco-articulatニionarequitefew.Althoughalmosta11
attemptstoexplainitareconductedbyformantfrequencies,those
whichhavesufficientstatistニicalfoundationsarefew.
ThereportofBroadandFertig(,1)StateSthatthepatternsof
influencesofvariouskindsofinitial0rfinalconsonantsonthe
vowelregionofcvcsyllables(sequencesofconsonant-vowel-consonant:
Onlythevowe:L/I/wasused)arequantitativelydisplayedbyananaly-
sisofvariance(univariate).Ithasastatisticalground,sothat
itニisquiteinteresting.
Thefirst,secondandthirdformantsarestillseparatelydealt
withasunivariate,respectively,intheirreport.However,since
thereisnoguaranteethattheseformantsareIndependentofeach
other,andsinceitisnotclearhowmanypartsofsoundinformation
theformantsshare,theirreportisnotsufficienttoexplainCO-
articu:Lation.
Wecarefullystudiedtheprocessoftheirresearch,andthen
plannedfurtherresearchwhichmustgobeyondit.
Wedirectlywrestlewiththeproblemofco-articulationInclud-
ingconsonantsaswe1:Lasvowels,usingspeechspectrumwhichisa
simpleexpressionofphysicalspeechinformationasmentionedabove.
-15-
![Page 19: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/19.jpg)
And,wealsotriedtoevaluatetニheindividualdifferencescausedby
differentニspeakers・
Wegainedtheotherresultsconcerningtherelativespatial
distributionofconsonants
whichhasfourfactors
andfina:Lvowel
three-dimensionalrepresentationof
-16-
consonants.Eventhoughspatialrepresentationbasedonactual
speechsoundhadearlybeentriedonvowels,thespatialrepresent-
ationofconsonantshadbeencomposedbyphonetician'sinsightso
far.Physicalandanalyticexpressionsbasedonactニualspeech
soundare,forthefirsttime,triedhere.
Wewillstatetheframeworkofthisthesisattheendofthis
chapter。
工nChapter2,wegiveasummaryofmultivariatestatistical
analysisusedinthisthesis,andanexplanationofthemultivariate
analysisofvariance.Wealsotouchalittleontwotheoretical
proposalsforthemultivariateanalysisofvariance.
Acomputerwasusedinthisstudythroughouttheprocess
fromtheCo:Llectionofdataofspeechsoundtotheinvestigationof
theanalyzedresu:Lts.Thisstudycouldnotbeaccomplishedwithout
thecomputer.InChapter3,wetouchonthemeansofspeechana:lysis
includingthiscomputersystem,onspeechspectraandonthedefinition
ofvectorscorrespondingtospectra.Alsowedescribethemethod
ofsegmentationadoptedhere。
InChapter4,weperformmultivariatestatisticalanalysis,
speaker,initia:Lvowel,consonant(nasa:L)
onavcvword(vowel-consonant-vowel)spokenby
fiveadultmales,andmeasuretheamountandthedirectニionofthe
co-articulationsandthespeaker-effect.Bymultivariateregression
theoryweinvestigatethecorrelationbetweenfina:Lvowelandeach
spectrumsectionofword.Principal-componentanalysiswasalso
performedandcomparedwithmu:Ltivariateanalysisofvariance・
Thedifferencebetweenspectralinformationandformantinformation
isa:Lsoconsidered..・.・・..・
![Page 20: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/20.jpg)
Chapter5dealswithanalysisofvarianceforfour-factordesign
withrepeatedmeasurementsonwords―C V,C2V2‘―spokenbyanadult
male.AssigningfourfactorstoCI゛V1゛C2andV2irespectively,we
observeconsiderablyindetニailtheirinteractionsaswellastheir
maineffects.Particularly,wemeasurethedepthofco-articulations
howfartheinf:Luenceofonephonemeextendstotheothersin
phonemicsequence。
Chapter6zTocomparepatternsofco-artIculationswitheach
otherintwotypesofphonemicsequences-C-V2(consonant-vowel)
andv1-V2(vowel-vowel);wescrutinizethemagnitudeofthe
influenceoftheprecedingconsonantCandvowelV10nthefollowing
vowelV2.Forthatpurpose,weextendthemodel0fusualanalysis
ofvariancefortwo-factordesign。
Chapter7:Weanalyzenineconsonants,andobserverelatニions
oftheirrelativedistニributioninatニhreedimensionalspace.Asa
result,thephonemesarenearlyrepresentedwithatriangularprism.
whichisstatistica:Llyguaranteedfromvariousaspects.Theexperi-
mentofthischapterwasperformedtogetherwiththeexperimental
schemeofChapter5。
SofarastheAppendicesareconcerned,wedescribemathematical
deductionsandproofswhichwillberedundantinthetext.
-17-
![Page 21: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/21.jpg)
CHAPTER2
MultivariateStatistica:LAnalysis
2.1工ntroduction
Multivariatestatisticalanalysisdealswithatheoryof
statisticsinwhich;
(:L)Whenweconsiderpcharacteristics(p>l)concernedwithan
indiv工duala,theirrespectivemeasuredvaluesareexpressedbythe
componentsofvector
z(z(1xp)゜(xαls'¨9xαp) (2.1)
whichisarealizationofap-dimensionalrandomvariable.
(2)Theobservedvaluesonnindividuals{cc=l.・・・,n)aresummarized
inthedatamatrix
Xlp
^np
「--'―一一
(2.2)
whoserespectiverowsareindependentlydistニributedaccordingtoa
p-dimensionalprobabilitydistribution.
Notethat゛CZandxg(a.6°19"・,n,a≠B)ai°eIndependentofeach
other.whilethecomponentsof・>^ax(xiandx(xj(恥j°19°・・.p.i≠j)
arenotindependentofeachother.ThevariatesXrtl,・・・9x(zpare
dependentamongthemselvessothatwecannotsplitoffoneormore
fromtheothersandconsideritbyitself.Thevariatesmustbe
consideredtogether.:
-18-
![Page 22: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/22.jpg)
Anotニherimportantcharacteristicofmultivariateanalysis
arisesfromadivergenceofinterestニbetweenmathematicianand
statistician.Thenaturalinclinationofthemathematicianis
towardsgeneralization.Givehimaresultforonevariateandhe
inquiresaftニertheresultfortwo;givehimthatandheinquires
aftertheresultforp.Thestatistician,ontheotherhand,is
continuallystrugglingtoreducethedimensionsofhisproblem.In
multivariateanalysisheusuallyhasembarrassingprofusionsof
variatesandhisobjectistomakepassmallashecan.(Hestill
prefersnaslargeaspossible.)Discriminantanalysis,forexample.
triestoreducetheproblemofdistinguishingbetweenmultivariate
populationstothescaleofasinglevariate.(2)
Now,thissortofdifferenceofstandpointisobligedtocause
adifferenceofevaluationofcertaintheories,themostintensively
influencedoneofwhichwasthemultivariateanalysisofvariance。
Thetheorywasestablishedbyextendingthestatisticalconcept
andmethodoftheunivariatecasetothemultivariatecasedirectly・
Unfortunately,remarkableapplicationofthemu:Ltivariateanalysis
ofvariancetopracticalcasehasnotbeenfounddespitethefact
thatitstheorywasoneofmagnificentresultsinmathematicS53)
TherewereevensomeauthoritieswhodenieditspracticaluSagj4)On
thecontニrary,inthisstudy,thismethodwasveryusefulfordiscover-
ingseveralnewfactsonspeechsound.Thisrevealsthattheprac-
ticalvalueofmultlvariateanalysisofvariancewasconfirmedsub-
stantlallyforthefirsttimehere.Asimilaritymaybepointedout
abouttheregressiontheoryinthecasewheredependentvariatesare
multivariates。
0ntheotherhand,multivariateanalysishasaweakpoint.
Namely,itisdifficulttoproceedwithsystematicanalysisexceptin
thecasewherewecanassumeanormaldistributiontobethefunda-
mentaldistribution.But,weareveryfortunateinthisstudybecause
1tispossibletoassumethenormaldistribution,aswecanunderstand
itinthesectiondescribinganalysisofresidual.Respective
-19-
![Page 23: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/23.jpg)
componentsofprobabilityvectorinthisstudydonotrepresent
numerizedvaluesofvarioustypesofconceptsasinpsycho:Logy,but
physicalvalueswhichhavethesamedimension,thatis,thefrequency
componentsofspectrumdistribution。
Wewillgiveanoutlineofmultivariatedistributionfirst,in
thischapter,followedbytheregressiontheory,analysisofvariance.
andprincipal-componentanalysis.Wewilltouchalittleontwo
proposalsformultivarlateanalysisofvarianceinatheoretical
aspect.
2。2MultivariateDistribution(5)(6)
2.2.1NormalMultivariateDistribution
Ifthedensityfunctionofap-dimensionalrandomvector.x;(1xp)
is
(27r")"^IAけ 収-μ)が(x-U)'} (2.4)
whereμisagivenp-dimensionalvector,andAisagivenpxp
symmetricpositivedefinitematrix,xissaidtobedistributed
normally,ordistributedaccordingtothep-dimensionalnormal
distributionN(μ,Λ).Then,μiscalledtheexpectedvalueofthe
vectorX,andΛiscalledthecovariancematriχ.
μ=6(.χ;)(2.5)
A=e(x-μ)'(x-μ).(2.6)
2.2.2ProbabilityEllipsoid
lf硝九分)isdistributedaccordingtoN(μ!A),(‘l;-μ.)rl(゛‾μ)
20-
![Page 24: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/24.jpg)
hasaχ^-distribution(Chi-squareddistribution)withpdegreesof
freedom.
Letχ;((x)bethenumbersuchthat
Pr(χ;≧χ7)(o,)}=(19
whereχ;hasaχ^-distributionwithpdegreesoffreedom.
Thus,
Pr{(jc-μ)Λ‾1(.x;-μ)'jχ^(a)}=1-a.
Inthep-dlmenslonalspaceofZ,
(z-μ)Λ‾1(z-μ)'£Xp(o・)
(2.7)
(2.8)
(2.9)
representsthesurfaceandtheinteriorofae1:Lipsoid■whosecenter
isμ.TheshapeoftheellipsoiddependsonΛ‾1,andthesizeon
χp(a)forgivenΛ-1Theinterioroftheellipsoid(2.9)is
consideredtocontain(:L-a)×100%ofthepopulationandisca:Lied
the"Probabilltニye:L:Lipsoid."
lfΛ‾1°Ip,forexample,(2.8)saysthattheprobabilityIs
1べxthatthedistancebetweenZandμislessthanぶ<?,(a),whereIp
isthep-dimenslonalunitmatrix.
2.2.3ConcentrationEllipsoid
LetμandΛ=e(Z-μ)゛(Z-μ)betheexpectedvalueofthep-
dimensionalprobabilityvectorj:(1Xp)andthecovariancematrix.
respectively・
Then,
(x-μ)Λ‾1(z-μ)'=P+2 (2.10)
1scalledtheconcentrationellipsoidcorrespondingtoX,whereA1S
-21-
![Page 25: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/25.jpg)
thepXpsymmetricpositiv!E!definitematrix.Thevolumeofthe
concentrationellipsoidisequa:Lto
pp
(p+2)T伊
Γ(号+1)
(2.11)
Nowweconsidernewp-dimensionalprobabilityvectory,whosedensity
is
g(μ)=
r(-|-p+i)
固{‘(r+2)号訪,
0,
(μ-μ)Λ‾1(μ-μ)゛≦p十2,
otherwise.(2.12)
Then,itcanbeprovedthate(y)=μ,e(μ-μ戸(ZI-μ)=A.
Therandomvectoryisuniformlydistributedintheinterior(andon
thesurface)oftheconcentrationellipsoid,whilethedensityofμI
isOattheoutsideoftheellipsoid。
Theconcentratニione1:Lipsoidisconsideredtobethegeometric
expressionofdistributionofzwithexpectedvectorμandcovariance
matrixA.
2.2.4SamplesfromtheMultinomialPopulation
Let^1.‥・,Xjjberealizationsofp-dimenslonalrandomvariable
distributedaccordingtothemultinormaldistributionN(μ,Λ).
Thesesamplevaluescanbesummarizedinthedatamatrix
ゴ~
「万
X11°゜゜Xlp
●●●●●●●
xnl°●●
xnp
]
(2.上3)
ThedensityfunctionofthejointdistributionofX,゜¨゛znis
22-
![Page 26: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/26.jpg)
f(ZI・¨゜・zn:μ・/1)゛
一
一
1
(2π)}叩|/1け1 呻緋j
l(゛i‾″)j‾1(゛i‾″)勺
(2.14)
whereM'(pxn)゜(μ≒¨',μ・)andtrindicatesthetraceofmatrix,.
becauseXj,・・・タxnaremutuallyindependent・
Thefunction(2.14)isalsoconsideredtobethelikelihoodof
observatニion(2.13),fromwhichwecanconcludethatthemaximum
likelihoodestimateofthemeanvectorμisμ=Z..
Wherex.゜圭£Ziisthesamplemeanvector,andthatthemaximijinni=IA1
likelihoodestimateofthecovariancematrixAisΛ=-Qwheren
n
q=Σ
1こ1
(zi-z.)゛(zi-z.)
isthematrixofthesumsofsquaresandcrossproducts
ThematrixQcanbewritteninthefollowingform;
where
Then,
Q=Y'Y
Y= [レン]
rank(Q)=rank(Y'Y)=rank(Y)=inin(n-l,p),
because
n
ふ(Xj-X゜)≒R‘`;i‾゜;'゜0°
Distributionofthesamplemeanvector
(2.15)
(2.16)
(2.17)
LetZ.=否iミyl°xibethesamplemeanvectorofnindependent:
observatヌionsZifromthemultinormalpopulationN(μ,Λ).
-23-
![Page 27: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/27.jpg)
(2.20)Q:(pXp)=X'AjX
Thenx.isalsoamultinormalrandomvariablewithparametersμ
and(l/n)A.
e shartdistribution
ThematrixQdefinedby(2.15)canbewrittenasthesumsofthe
productsofn-1independentp-dimensionalrandomvectorswiththe
conmondistributionN(0,Λ).Ingenera]L,anysymnietニricpositive
definitematrixQofquadraticandbilinearforms,whichcanbe
transformedtothesum
且‘'?“i (2.18)
whosep-dimensinalvectors"iareindependentlydistributedaccording
tothedistribut:ionN(0,Λ),issaidtohavetheWishartdistribution.
ThedensityfunctionofQis‥
W(Λ,P,r)=
2‾
1
‘
圈1(r-p-1)
{‾iトA‘IQ}・(Tり),(2'1?)
「nr[-|-(r+i-i)]
exp{緋tΓΛ
ifQisapositivedefinitematrix.
皿j‾p(p‾1)IAI
Cochran'sTheorem
Forthesamplematrix(2.15),r=n-1.Weshallrefertoras
thedegrees-of-freedomparameteroftheWishartdistribution;together
withAitspecifiestheformofthedensityfunction。
Ifp°1andΛ=1,theWishartdensitybecomesthatoftheChi-
squareddistributionwithrdegreesoffreedom.Wishart-dlstributed
matricespossessmanyofthepropertiesofChi-squaredvarlates.
SupposezαisdistributedaccordingtoN(O,A)independentlyof
゛6((゛邦)゛SupposethematrixAj(nxn)usedintheforming
-24-
(i°1,‥・,m)
![Page 28: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/28.jpg)
isofrank■■i.andsuppose
x゛x゛Q1十゜¨゛十%i (2.21)
wherex゛(pxn)゜(ぷll゛‘",x').(Note:工n°AI十'゜'十Am')
ThenanecessaryandsufficientconditiontニhatQi(i=l,・・・,in)is
distニributedas
n
ム "a '"a' (2.22)
whereuばisdistributedaccordingtoN(0,Λ)independentlyofaβ
((がB),andQiisdistributedindepe゛dentlyofQ-j(i≠j)isthat
Corollary
工fr
ri+r2+'‘゜+rm°n°(2.23)
ilp,thenQiisdistributedaccordingtotheWishartdistributionW(A,p,ri)
2。3llultivariateLinearRegression.(5)
Inunivariableleastsquaresweconsiderscalardependent
variablesxh¨'9xndrawnfrompopulationswithexpectedvaluesare
zlβ,*・・,znβ,「espectively,whereβisacolumnvectoroftheI
componentsandeachoftheZぼ1Sarowvectorofiknowncomponents.
Undertheassumptionthatthevarianceineachpopulationisthe
same.theleastsquaresestimatesofthecomponentsofβare
&(か(1)゛(£j(゛'≒)‾1(j
l句≒')゜
-25-
![Page 29: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/29.jpg)
Ifthepopulationsarenormal,thevectoristhemaximumlikeli-
hoodestimateofβ(β=6).Undertheassumptionofnormality,the
maximumlike:lihoodestimateof(y2is
a2゛`lぷ1(%‾゛(j)2'
Thealgebraofthemultivariatecaseisessentiallythesameas
thatoftheunivariatecase.
InthemultivariatecaseXn,1Savector.βisreplacedbya
matrixBand(J2replacedbyacovariancematrixΛ.
SupposeXl゛・‥タ*nareasetofnobservations,°x;Cχ(a°1ダ≒n)
beingdrawnfromN(z(zB゛Λ)゜Ordinar:i`:lythevectorszoc(with兌
components)areknownvectors,andthePXpmatrixΛlandthe
兌xpmatrixBareunknown.If
X'(pXn)゜(・ぢ,¨',ぷぶ),Z゛(兌xn)゜(ZI≒¨'9Zぶ)。
(2.24)
thenwecansummarizedtheaboverelationinthefollowingequation:
e(X)=ZB (2.25)
Assumingn2p十又。andtherankofZtobe£,themaximumlikelihood
estimateofBis
{B =(ZIZ)‾1
andthatofnAis
(Z'X)
nA=I(X-ZB)1(X-ZB).
(2.26)
(2.27)
IftherankofZislessthana,theinversematrixofz'zdoes
notexist.工nsuchacase,(2.26)isnotcorrect,butthemaximum
likelihoodestimateBsatisfies
-26-
![Page 30: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/30.jpg)
Z'ZB=z'χ (2.28)
and(2.27)holdsstilltrue・
Themultivariateanalysisofvariancestatedinthefollowing
sectioncanbeobtニainedfromthemodel0fthefixedcoefficientsofthe
multニivariatelinearregressiontheory.
2.4llultivariateAnalysisofVariance
2.4.1Introduction
Themethodofmultivariateanalysisofvariancewasobtainedby
generalizingthemethodofunivariateanalysisofvariancetothe
caseofvectorvariates.Therefore,asthebasicconceptisthe
sameasthecaseofunivariate,wearegoingtoexplaintheconcept
briefly.`
Factor residualandmodel
Analysisofvarianceisoneofthemeansfortheanalysisused
inthedesignofexperiments.Thegoal0fthedesignofexperiment
istoelucidatetheactionsofasetofvariouscauseswhichinfluence
thecharacteristicvaluesshowingtheresultsofexperiment.
(1)工ntheexperimentweartificiallysetuptheseveralconditions
oftreatmentsforeachcause-whichiscalledfactor-and
comparetheeffectsofthefactorsonthecharacteristicvalueswith
eachother.
(2)Thedataofthecharacteristicvaluesshowingtheresultofthe
experimentarenotconstanteveninanidenticalsituation,butare
accompaniedwithsomeaccidentalerrors.
(3)Toexplainthephenomenaoftheresultsofexperiment,amodel
1Spreparedaccordingtothefactorssetupat(1)andtheerrors
indicatedat(2).
-27-
![Page 31: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/31.jpg)
!:neveryexperiment,whethertheexperimentissuccessfulor
not1Sdeterminedbywhethertheconditionsoftreatmentspicked
upforcomparisonareadequate.Thatiswhy(1)isimportant.
Then,itiS十aprobl万゛万i71万howtochoosefactorsandlevels.
Factors:Thecauseswhicharepickedupandcomparedamonga
setofvariouscauseswhichareconsideredtoinfluence
thecharacteristicva:Luesoftheresults.
Levels:Theconditionswhicheachfactニortakes.(Theyarealso
calledcategories.)
Accordingtothenumberoffactニorspickedupinanexperiment.
theexperimentiscalled"one-factorexperiment","two-factorexperi-
ment","three-factorexperiment"andsoon.Taking"k-factorexperi-
ment"asanexample.ineachcaseofallthecombinationsofkfactors
weexperimentrtimes(r≧1)undera:LItheconditionsbeingkept
constant.Particularly,whenr>l,theexperimentiscalled"k-factニor
experimentwithrepeatedmeasurements"inwhichwecanobtaininfor-
mationontheinteractionaswellasonthemaineffect(described
later).Thetypeofthemode:レof(3)isdifferentaccordingtothe
numberoffactorsandaccordingastherepetitionexistsornot.
Anexperimentcontainsgenerallysomeunmanageableerrors,as
insistedat(2),evenwhenitisperformedinconsiderablyaccurate
conditニions.Unmanageableerrorsshouldbeessential:3-ydistinguished
frommistakeandfailure.Thecauseoftherandomerrorsisgeneral¬
1ycalled"Error-factor"thataccompanythedatahoweversuccessfully
anexperimentisaccomplishedwithoutanymistakeandfailure.The
error-factorisasetofinfiniteminutecauseswhichcouldnotbe
managedintheexperiment.
Besideserrors,thereisanotionof"Residual"mainlycaused
fromthewayofobservinganexperiment.Thedifferencebetween
theactualvalueofthephenomenonandthevaluewhichcanbeexplain-
edbythemodel(thatis,theresidualpartwhichwecannotexplain
-28-
![Page 32: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/32.jpg)
bythemodel)iscalled"Residual"whenweexperimentwiththefactors
chosenat(1)andthemodelsintroducedat(3)inordertoobservea
certainphenomenon.Aserrorsandresidua:LsappeartogetherWitニhin
themodelofcertaintype,botharegenerallycalledresidual.
StatisticalDecision
Nowletusconsidertodecidestatisticallywhethertheeffect
(influence)ofcertainfactorappearsintheresultsofexperiment
ornotニ.Then,wemustneveradmittheeffectonlybythefactthat
thedifferencebetweenthemeanvaluesofthedatawhichbelongto
therespectivelevels(forexample,AiandA2)ofthefactorseems
apparentlylarge,becauseexperimentsareinevitablyaccompaniedwithresidual.工ftheresiduaユislargerthanthedifferencebetweenmeanvaユues,it"becomesdifficulttodetectthedifferencesincetheyare
lostニinnoise,thatis,residual.1nthatcase,itisinsignificant
t:oconsiderthedifferenceofmeanvaluesitself.Wemustconclude
thatsuchfactorhasnoeffect.Meanv
valuesareconsiderablylargerthantheresidua:L,wecanconfirm
sufficientlythedifferenceofthemeanvalues,theeffectofthe
factor,andthedifferencebetweenAiandA2isstatisticallysignlfi-
cant。
Thus,statisticaldecision(so-namedstatisticaltest)is
characterizedbythewayofthatwearesuretニOcomparecertain
characterswiththelargenessofresidualwhenwecomfirmstatist!-
callythecharacters.Thisisaveryimportantpoint.Furthermore,
instatisticaltest,weprovidequantitatニivelyadegreeofcertainty
forjudgingthesignificancy。
ThesignifIcancyofeachfactorandthevalidityofmodelare
notrecognizeduntiltheseprocessesarecompleted.
2。4.2TheLikelihoodRatioTest(7)
WeWi1:Ldescribeaboutニthelikelihoodratiotest,whichwas
usedinthisstudy,amongvariousmethodsofstatisticaltest.
-29-
![Page 33: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/33.jpg)
LetΩbeam-dimensionalparameterspace,andωbethepartial
regionin・theΩ.Likelihoodratiocriterionfor・thetestofthe
statisticalhypothesisHo:∂ミωagainstthealternativehypothesis
H:∂eΩ-ω,isasfollows:△ll
λ=
max
∂、ぞω
-
max
∂4Ω
L
L
(∂)
(∂)
-
一
一
Lω
-LΩ
(2.29)
whereL(∂)=!f(X:∂)iSthelikelihoodfunction(Xisthedatamatrix
asinEq.(2.13),∂:m-dimensionalparameter).Inthelikelihoodratio
tニestgweobtainsamplingdistributionofλunderthehypothesisHO・
ThecriticalvalueλaischosensothatPr{入く入{ZIHO}゜a,whereais
thepreviouslygivensignificancelevel°{入:Oく入<λCz}iScalledthe
critica:Lregion.IfO(λ(λ(χ91:ejectthehypothesis。
工nactualtestweoftenperformatestequivalenttothelike-
llhoodratiotestbyuti:Lizingrathersimplestatistics,whichhas
amonotonicrelationtoλ,forthetestcriterionthanλitse:Lf.
2.4.3MethodofMultlvariateAnalysisofVariance(8)
Wewilltaketheanalysisofvariancefortwo-factordesignwith
repeatedmeasurementsinordertoshowtheprocedureofanalysisof
variancesimply.Considertwofactors
provided,
-30-
AandB,andassumeA1~
(2.30)
AaandBi'`'Bbtobetheirlevelsorcategories.(SeeTable2.1)
Whenlet*ijkbethek-thobservationunderthei-thtreatment(level)
offactorAandthej-thtreatmentoffactorB,wepresumethe
linearmodel
゛ijk(1)(p)゜μ十αi十ノ?j十‘rij十≪ijk・
(i°1,゜゜゜,a,j°1,゜゛゜,b,k°:L,"゜,r)
![Page 34: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/34.jpg)
and
abぶ
1“i乱気βj
bΣF=り
aΣだ
= '■ij一
一 0,
^ijk-NCO・A).
Table2.1Multivariateanalysisofvariancefor
two-factordesignwithrepeated
measurements.
FactorB
B,Bj Bb
≪
』
A.
χUI
Z11「
●
●
●
*Ibl
Xibr
A,
●
●
●
りi'
りj`
Zり「
●
●
●
Aa
Jfau
.χ;alr
●
●
●
Xabi
.χ;abr
(2.31)
(2.32)
Where,d^represents"Maineffect"ofthetreatmentkj_.βjalso
represents"Maineffect"ofBj,and^ijrepresentstheeffectspecific
tothecombinationoflevelAiandlevelBjタthatisthe"Interaction"
betweenfactorAandfactorB.Wedetermineμtobeindependently
distributedaccordingtothep-dimensionalnormaldistributionN(O,A).
Thetotalvariance,namely,thematrixofsumsofsquaresand
abrcrossproductswiththesamplemeanZ¨゜゜轟石呉ぶぶり.maybe
-31-
![Page 35: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/35.jpg)
representedas
abrQ(pXp)゜ぶ石ぶ(゛ijk゛…)'(゛ijk゛…)
where'denotesthetransposedmatriχ.
ThebreakdownofthetotalvarianceQbecomesas(2.34).
Where,
(2.33)
ab
Q°昌(゛i¨‾゛…y(゛i¨‾‾゛…)十jUI(゛゜j゛‾゛…)な゜j'‾孔¨)
ab
十且jFI(‘゛ij'‾゛i¨‾゛゛j゛十゛…y(゛ij'‾゛i・.‾゜゛;・j°十゛…)
abr
十iEjylぷ(゜`;ijk‾゛ij゛)でijk‾゛ij.)-Qi十Q2十Q3十R'(2‘34)
1br1arlr
゛i¨゜言石ぶ1゛ijk'゛'j'ニ盲且ぶ゛ijk'゛ij.゜7ぶ゛ijk-
Now,letusconsiderthelikelihoodratiotestofthehypothesis
thatHo:tfl=・・゜゜(Za°O'
ThemaximumofthelikelihoodfunctionLinSection2.4.2is
‰゜(211jl)11>n|頑今eぶ(“1pn)
(n=a・b・r).WhereAisthemaximumlikelihoodestimateofA:
andnA==R.
(2.35)
Also,themaximumofthelikelihoodfunctionLunderthehypo-
thesisthatHo:(Z1°‘゛゛゜゜゜(Za゛Oistrue,is
‰゜(2好ipnjl門nA]^・xp(弓pn) (2.36)
《WhereAisthemaximumlikelihoodestimateofΛunderthehypothesis
H,andnA°Qi十`R'∧
Weutilizethestatistics,whichisamonotonlcfunctionofthelike-
lihoodratiocriterionλ,
z・=(Lω
-LΩ)
11
一一
辿リ
-
-
IrI
|Qi+叶
-32-
(2.37)
![Page 36: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/36.jpg)
forthetestcriterion.Itisgenerallydifficulttoobtainthe
exactdistributionof7。.However,sinceQi~で耳“i'≪i(≪i~N(0,Λ))
holdsifthehypothesisistrue,andR~W(Λ,p,n-ab)holdsalways,
themomentof・lumaybeobtainedexactlysothatwecanlearnits
asymptotニicdistributionbyBox.(9)
N3°elyy
=-{n-jと2-(p十聡+1)/2}1og
J
(2.38)
トノヤ勺-(pサけ1)/サ1゜(巴斗斗`・
isdistributedasaChi-squaredvariateWitニhp兌1degreesoffreedom
3二二で二ぶ二こ二二ご乱
。,。(,。-。。。)
Table2.2Multivariateanalysisofvariancefor
two-factordesignwithrepeated
measurements.anddegreesoffree・
domofQi.Degre・offreedomof
Risn-ab(n=abr),瓦十島=ab.
Factor Effectvector Qijl:訟m欧fl゛e‘
A αiQi a-1
B βj Q2 b-1
AB Ti] Q3 ab-a-b+1
(Thedetailedprocedureofthedemonstration1Sunderstandableinthe
explanationofsimilarmodel0fAppendixAorG.)
2.5TwoproposalsforMultivarlateStatistical
AnalysisofVariance
Inthisreport,twomatteres,intheoryaswellasinapplica-
t:ion,werefoundtocontributetomultivarlateanalysisofvariance.
-33-
![Page 37: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/37.jpg)
Theseare:
(1)Notionofdirection,besidesnotionofamountindicatedgeneral-
1ybytheanalysisofvariance,wasnecessarytounderstandthe
effectsofrespectivefactorsofanalysisofvariance,andoneofits
expressingtechniqueswasdescribed.
(2)Amodel0f"Divided-type"analysisofvariancewasproposed.
Thisisanup-to-datemodelinwhichwedevelopananalysisof
variancefortwo-factordesignbydividingoneoftwofactorsintomore
thantwoportions。
ThelatterwasIntroducedtodoresearchonpatternsofsome
typesoftheco-articulationsandthenformulated。
Thedetaileddescriptionof(1)isinSection4.5,andthat
of(2)isinSection6.2.
2.6Principal-ComponentAnalysis(2)(10)
SupposewehavepvariablesXl,・・・・*・jXp,eachofwhichis
observedonnindividuals.工fwewritexαiforthea-thobservation
onthei-thvarlate,thea-thobservatニionvector.x;ぼisshownas
fo1:Lows;
゜X;α゜(xα111°゜IX(Zp)(2,40)
Theobjectofcomponentanalysisistoeconomizethenumberof
variates..Todothis,weshallseekforlineartransformationsof
type∧--づで.I`'●・・
p
Yai°jE“ijzαj i=l,2,...,p. (2.41)
]
theyα19¨゜≫yofp゛Wehavetheneffectedagenuinereductioninthe
-34-
![Page 38: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/38.jpg)
dimensionsoftheproblem,thewholecomplexofvariationsbeing
expressibleinmくpvariates.Butthisisexceptional.Whereitis'
notpossibleweshalltrytニocarryoutanapproximatereductionin
thissense:weshallchoosethecoefficients“ijsothatthefirst
ofournewvariatesyαlhasaslargeavarianceaspossible;weshall
tニhenchoosethesecondyα2soastobeuncorrelatedwiththefirst
andtohaveaslargeavarianceaspossible;andsoon.工nthisway
wetransformtニonewuncorrelatedvariateswhichaccountforasmuch
ofthevariationaspossibleindescendingorder.Itmaybefound
t:hatthefirsttwoorthreeofthesevarlatesaccountfor"nearly"
thewholeofthevariation,say850r90%,andthecontributionof
theotherp-2orp-3issmall.Wecanthensaythatthevariation
isrepresentedapproximatelybythefirsttwoorthreevariatesand
infavourablecircumstancesmaybeabletoneglecttheremainder.
Now,whenwepresumeZ(Z((Z゛19¨゜,n)tobeobservatニ10nvector
●asmentionedabovetoexplainmoreindetail,wedefinethatsample
meanvectorZ・,andsamplecovariancematrixSare,respectively,
z。(lxp)=
S(pxp)=
ごぷ
`
とj
l(゛i‾゛゜)'(゛i‾゛')
(2.42)
(2.43)
Here,letXi>λ2>'¨>入p>ObethepeigenvaluesofSande19e29‥・9
ep・ン.‘bethecorrespondingeigenvectors,wheree,e,'°1seje:'°O(
i≠j).Thei-thcomponentyαi°(Xfjrx.)ej'of
H(i=(yぼ19¨゜9y^p)=(・%-ぷ.)(ei',"≪,ep')
iscalledthe"i-thprincipal-component"ofthesamplevector・χα,the
varianceofwhichisequalto入i°(ycciandy(Zjareuncorrelatedwith
eachother,(i≠j))
Then,wecanmeasureby入i/t17(S)ho`ヽ7muchthei-thprincipal-
componentisimportantsince入1十゜¨+λp°tニr(S)‘Ifweassumethat
-35-
![Page 39: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/39.jpg)
itispossibletoneglectthe(m+1)-th,・・,thep-tニhprincipal-
components,
m且
ぐ λi/tr(S))x100% (2.A4)
ofthetotalvariationofZα((Z°!,・・・,n)maybeexplainedbythe
1-st,・・・,m-thprincipal-components。
Therefore,thecomponents(thenumberofwhichisp)ofthe
orgina:LsamplevectorzcZcanberepresentedwithaSma:Llernumber
ofprincipal-components(thenumberofwhichism)thanp,thatis,
thecomponents(thenumberofwhichism)ofthenewsamplevector
Va.・
-36-
![Page 40: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/40.jpg)
4
CHAPTER3
FacilitiesforSpeechAnalysisand
SpeechSpectrum
3.1Introduction
Ourideasontheequipmentforspeechanalysiswillbediscussed
intニhischapter.Thespeechsoundsutteredinasimplifiednon-
reverberantroomareana:Lyzedthroughthe20-channel:L/4-octave
filter-bank,andtheoutputsarerectifiedandthensmoothed.The
smoothed20-channelsignalsaresimultaneous:1-ysampledatintervals
of10msec,arrangedontニhelineofonechannelbythemultiplexer,
andguidedintotheAD-converter.ThedigitalphonemedataAD‘convert-
edarecompiledonmagnetictapesthroughthecomputer.(SeeFig.3.1.)
A1:Lthecontrolsmentionedaboveareconductedbyon-lineand
inrealtime.Itis,ofcourse,alsopossibletocompilethesignals
recordedbyt!letaperecorder.Thesoundsthusobtainedareatfirst
embossedontheline-printerinordertosegmentthestationaryor
theboundarypartsoftニhesyllables.Bycontrollingtheapparent
shadesofeachpointbychangingthelettersonthepaperforthe
line-printer,thespectraofutterancesarepresentedonitニ(the
expressioni&similartotheSonagram)。
Theresultsofthesegmentニationthroughthevisualobservation
areputintothecomputerbythetypewriter,andmemorizedwiththe
speechsoundspectra。
AnalysisofvarianceisperformedinFORTRANlanguage.Asa
considerablylargeamountofdataisneededforoneexperiment,the
diskpackfilesystemisusedonthecalculation.
-37-
![Page 41: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/41.jpg)
MIC:Microphone
TRこTaperecorder
P-E:Pre-emphasis
F-A:Filter-Analy多er
MPX:Multiplexer
AD:AD-converter
LP:Line-printer
MTごMagnetictape
DISK:Diskpack
20-channels
Computer
NEAC2200/200
ヶ∠
○○ DISK
Fjg・;3.1On-line,realtimedatainputsystem.
Typewriter
3.2NonreverberantRoom,Microphone,Taperecorderand
Pre-emphasisCircuit
Thenonreverberantroomofourlaboratorywillbedescribed.
0ntheinnersurfacesoftheauditory-examlnation-room(modelAT-5E),
whichism万j1万debyRionCo.Ltd.(2meterscube),wehavestickedpieces
ofglass-woolthethicknessofwhichis20cm.Itis,namely,a.
simplifiednonreverberantroom.Asaresultofroughmeasurement,
wefoundthesound-insulation-effectwasmorethan30dBabovelOOHz.
38-
![Page 42: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/42.jpg)
Λlthoughwedidn'tmeasuretheechoes,wecouldscarcelyhearthem.
ThemicrophoneusedhereismadebySonyC0.Ltd.(mode1ECM-52)
工tisthe"electret"condensermicrophone,thatisasimplified
condensermicrophone・
ThetaperecorderusedhereismadebyTEACC0.Ltニd.(ModelR-
341C),andusuallyusedatthebroadcastニingstation.工tscharacter-
isticsagreewithBroadcastingTechnicalStandardinJapan(BTS
5313).
Apre-emphasiscircuitwasincludedinordert0improveto
signal-tニ0-noiseratioathighfrequencies.Thefrequencycharacter-
isticofthepre-emphasiscircuitisshowninFig.3.2,andwemade
thecircuitasFig.3.3byusingtheoperationalamplifier(analog
integratedcircuit).
dB
0
-10
uicn
-20
二.ニーp
J〆y
////
〆///
//
//
-
/100 200 500 1K 2K 5K
Fig.3.2Frequencycharacteristicofthepre・emphasiscircuit
(breakpointfrequency=1.6kHz).
-39-
10K
Hz
![Page 43: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/43.jpg)
IN
A:TA7502M(Toshiba)C,=500PF
R,=200Kぶ?
50K300K
Fig.3.3Pre-emphasiscircuit
3.31/4-OctaveFilterAnalyzer
OUT
十
-
The1/4-octavefilteranalyzeriscomposedofthreeparts
frequenciesofthefiltersincrease:・byafactor
-40-
/linorder.
(1)1/4-octavebandpass-filter-bank(MadebyRionCo.Ltd.,Model
SA-11)
(2)rectifiers(homemade),and(3)smoothingcircuits(homemade).
Wewill,now,explainabouttheseparts.
(1):L/A-OctaveBandpass-Filter-Bank
1/4-octavefilter-bankisthefilter-bankin‘whichthecenter
Thefrequenciesusedinthisresearchare210,250,297,354,420,500,
595.707,841.1000,1190.1410.1680.2000.2380.2830.3360.4000.4760and
5660HZ,andthenumberofthefiltersis20.
Theattenuationcharacteristicsofthefi:Ltersareasfollows:
(1)Theattenuationvalueatthefrequencywhichis1/8-octavefar
fromthecenterfrequencyis3±ldB.
(2)Theattenuationvalueatthefrequency,1/4-octavefar,ismore
than12dB.
![Page 44: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/44.jpg)
UTU3
(3)Theattenuationvalueatthefrequency,one-octavefar,ismore
tニhan35dB。
Thefrequencycharacteristicsofthefiltersaredescribedin
Fig.3.4.ThecircuitcompositionisasFig.3.5.Thelinearityof
theinput-outputoftニhefilterismorethan40dB.
dB
0
-10
20
30
40
j万リ j 万 A 7 X
⑥仙j 坏 1
ノ/
ノ
リ ス y リ 1
V
ヅノ
坦ぶノリj A \
K
A//
ラ X ぐり∧
y
ぐぐ
言冷万万 y \
K
\
X
\
ノ⑤
万万ク y泣
ミ 万
言回1 ぐ ス く \
\
X
\
で
\
\
\
\レ1002005001K2K5K
Fig.3。4Frequencycharacteristicsofthe1/4-octavebandpassfilter-bank.
-41-
10KHz
![Page 45: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/45.jpg)
INCi
Li
Fi&3.5Filtercircuit
OUT
(2)Rectifiers
Theoutputwavesofeachfilterarefu:Ll-wave-rectified.To
guaranteethelinearityoftheinput-outputsufficiently(namely,in
ordertoobtaintheoutput-valuesinproportiontotheinput-amplitudes
evenwhentニheamplitudesbecomesmall),wemadethecircuitasFig・
3.6byusingtheoperationalamplifier(analogintegratedcircuit)・
Asaresult,thelinearityof40dBwasguaranteed。
,A:TA75Q2M(Toshiba)
20KeD:1S306(NEC)
Fig.3.6Fullwaverectifier.
OUT
(3)SmoothingCircuit
Asmoothingcircuitisnecessarytosmootheachfull-wave-
rectifiedsignal.Foreachchannel,weprovidedthesecond-order
-42-
![Page 46: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/46.jpg)
lowpass-filterwhosecut-offfrequencyis40Hz,asthesmoothing
circuit.Uemadetheactive-filterasFig.3.7usingtheoperational
amplifier(analogintegratedcircuit).
Thefrequencycharacteristicoftニhelowpass-filterisasin
Fig.3.8.
E,
Ri=11.5Ki?
R3=7Kj?
R4=17.2Kぶ?
C2=0.8μF
C5=0.16μF
Fig.3.7Smoothingcircuit.
Transferfunction;
E2-Hωo2-=一一-一一一一E,s2+ぼωos+ωo2
一
一
A:TA7502M(Toshiba)
| H=1.5
α=√2
fo=言=40Hz
OUT
+
-
-l/RiRa
s2C2C5十sC5(1/RI+1/R3+1/R4)+1/R3R4
-43-
![Page 47: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/47.jpg)
ureo
dB
0
-10
20
30
40
‾‾'べ
へへ犬
Xゝゝ
XXN
ぺ
`戈
\
、
\曳
\\
\\
X
10 40 100 400
Fig.3.8 Frequencycharacteristicofthelowpassfilter
(smoothingcircuit).
3.4On-line,RealTニimeData工nputSystem
Hz
Wehaveanon-linerealtimedatainputsystemwiththecomputer
NEAC2200/200showninFig.3.1.Thesignals(20channels)
Speechspectra
-44-
whichpassedthroughthe1/4-octavefilteranalyz-
eraresampledatIntervalsof10ms,andthenAD-converted,and
arrangedonthelineofonechannelbythemultiplexer..Theaccuracy
oftheconversionis10bits(andsignbit).Timerequiredfor
![Page 48: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/48.jpg)
AD-conversionis50ys.
TheprocessoffilingtheAD-convertedspeechdataonthe
magnetictapesofNEAC2200/200isperformedbyon-lineandin
realtime.
3.5SpeechSpectra
Iftheoutputsofa1/4-octavefilteranalyzerwith20channels
(theoutputsof(3)inSection3.3)thisisnamed"Theamp:Litude
outputニsofthe20-channel1/4-octニavefilters"
≪(t)゜b21(t)+b22(t)十゜¨十b2p(t)
-45-
arerepresentedby
(3.1)
b,(t)・b2(t)・‥゜・bp(t)(p°20)inorderofincreasingfreque゛cy,they
denotespeechspectrumattimet(bi(t)≧O).And・thele昭th‾
ofthevectorof(bi(t),b2(t),≪≪゜,bp(t))isconsideredtobe
theinstantaneousamplitudeofspeechsoundattimet.Schematic
representationofaVCV(vowel-consonant-vowel)syllablesequence,by
usinginstantaneousamp:Lltudea(t),isseeninFig.3.9.Asseenin
tニhefigure,theinstantaneousamplitudeofspeechsoundcorresponds
tニotheintensityofspeechsoundateverymoment,varyingapproximate-
lyover60dBrangeeveninausualconversation,anditissurelyone
oftheimportantpieceofinformationonspeechsound.Inthisstudy,
however,wearetogetridoftheinformationoftheinstantaneous
amplitudeinsteadofpayingattentionparticularlytotherelative
intニensitybetweenfrequencycomponentsofspeechsoundspectニrum.In
otニherwords,normalizethesquaresumofspectrumcomponentsat1。
Meanwhile,theauditorysensefortheIntensityofthespeech
soundissaidtobeproportionaltothelogarithmofphysical
intensityofsound(Weber-Fechners'Law)。
Takingaccountofthesetwopoints,inthispaperspectrumxCt)
![Page 49: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/49.jpg)
attimetisdefinedasfollows:
where,
dB
0
α(t)
-30
60
゛(t)゜(Xi(t),X2(t),.."Xp(t)),
xi(t)=1og」ヂ{ま=1ogフフウtアヅサケT
Vowel
0 50 100 150 200ms
Fig.3.9Schematicrepresentationofavcv(vowel・consonant-
vowe!)wordbyusinginstantaneousamplitude研い.
3.6SegmentationofConnectedSpeech.
byVisualObservation
(3.2)
250
t
Tosegmentaconnectedspeechutterenceintodiscreteparts
suchasphonemes-isknownastheproblemofsegmentationof
connectedspeech.Automaticsegmentationofspeechsoundisoneof
theresearchsubjectsoninformationprocessingforhumanspeech,
andIsalsoadifficultproblemasyetiinsolved.Sincehumanabilitニy
-46-
![Page 50: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/50.jpg)
isconsiderablyfavorabletotreatinformationastothissortof
problem,wearetosegmentvisually.Todothatニ,thespeechsounds,
recorded・onthemagnetictapesoflthecomputer,areoncetypedouton
theline-printer.Typeoutthreekindsofinformations,whichare
presentedbelow,inordertoacquireaccurateresults.Thevalue.
introducedfromtheoutputsofthefilteranalyzer
butnotx(t)ofEq.(3.2)
b(t)=
b1(t)゛゛¨bp(t)゛
(3.3)
(3.4)
,isusedhere.(SeeSection3.5.)
十‥・十(bp(t2)-b
(1)Theinstantaneousamplitudeofspeechsound:
α(t)=/bl(t)十‥.十bj(t)
(2)Themagnitudeoftimedifferentiationofspectrumvector
(b,(t:)・゛゜'・bp(t)):
(b1(t2)-bl(t
t2-tl
(3)Shaderepresentationofspectrum(bi(t),・・゜゛bp(t))゛
Although(1)or(2)hasbeenadoptedfortrial0fautomatic
segmentationofthespeechsoundbyutilizingthevariationofits
value,itisactuallyimpossibletoperformexactlythesegmentation
onlyonit.However,ontheoccasionofsegmenationbyvisualobser-
vation,theseareconsiderablyreferentニial.
(3)meanstoobservedirectlytheintensityofspectrum
distニrlbution.工tissimilartotheSonagram,andveryfavorablefor
analysis.Controltheapparentshadeofeachpointbychanging
lettersontheline-printer.Adoptapieceofapparentlydarktype
forthepointニcorrespondingtointensivespectrumcomponent,whilea
pieceofthintypeforweakcomponent.
Usingthesethreesortsofinformationdetニerminevisuallythe
stationaryortheboundarypartsofthephonemes;andputthetimet゛S,
correspondingtothem,intothecomputertythetypewriter.Theseare
¥●●●j●●filedwiththeoriginalspeechsounddata.
-47-
![Page 51: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/51.jpg)
X100
b(t
300
*\
*\
yt‥
\,
●●●●
/゛
●●●●●●♪。。
(ljb
●●●●●
*'*-*'*ぺ本`*j
●●●●●●●●●●●/
●●●●●●!*
/
。
y
。・。
\
●
300nns本一本-*一水´
*
d
1000
a(t)
≫―'
/゛
●●●●●●●●●500
150ms/
寒-χ一¥:-*-*一本一本
AA/AA//////。
/'////A/A*3A)f()xAA//A/
//l\^/^^^^。7)^^f…^^l\/a/
?'.AAAAAAr^ax≫r'K*A//A/
AAAAAAAhttp://www.fAA/A/
AAAAAA>f*'5f?xift")fifAAAA/
AAA*FAA!f*-f'>f!iX)ft5x≫AAAAA
ンンン‘'XXご'心心X"I"‘'ンンン\"
AAAi(xXH#rKJX≫cl)(A≫AAA*
y≫x*xx≪)f。i);jK>ff)n)A≫AAA*
AA≫*)*j<≫>
ヽ\AA≫≫≪>f≫f)f?i<Ax*≫*AAAx-
/AAA/A///AA////////ヽ\
////AA
///////AA
A//////AA
A/////AAA
/A//AAA*AA
////AifKAA/AAxX\)(・≫/
AAifAAAxtJe'ifAA≫xill;xxX/
#*■x■AAltf)S)(■AAA*x■・^^;^(。^(^(■/
水***^*Oi)*/\Af\*x--ni***/
xx)nf≪^tt>*f\f\'\^^9.fi'**/
水*****c。is≫≫AAAi<.x.r*t.v≫≫)f/
)(*HK)(X(^;e"<AAAi(K^f^y^f^(x/
x≫xxxtffcjJxAAAifxXi'^^Ak/
A*)(≪)(xc;yxAAAAAxx≫A)fA
A***^(*ui^≫AAAAA)f**A≫A
ンン*ンン**をンン\へ\`、'本水`'、。、'
ヽ\AA/A/AAA//////////ヽ'
\
へ
0
4KHz
2K
ヘヘ
1K
500\
\\ヘヘ
ヘ\>>>>>
\へ
\
\
250
t350300ms
00
1t-
1
'I
t.
09
10・*-'
CM200150
t7t5
100
1
t3
50
Fjg。3。10Shaderepresentationofthespeechspectra.a(t),and6(t)bytheline-printer.
andthes^mentationbyvisualobservatioaThewordis/deba/thatweactually
providedforanalysisofChapter5。
-48-
![Page 52: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/52.jpg)
Fig.3.10showsanexampleofhowtotypeoutontheline―printer
andthesegmentationbyvisualdecision
1≦・bi(t)
1≦bj(t)<10
10
10
くb
争
<bi(t)
102くb・(t)≦10‾251
10‾2‾くbi(t)≦103
≦1000,
t:hewordis/deba/that
/
V
*<SJ
(3.5)
(3.6)
-49-
weactuallyprovidedforanalysisofChapter5.
AdjusttheintニensitニyofspeechsignalswhenthesignalsareAD-
convertedsoastomakethemaximumvalueofspectrumcomponentbi(t)
below1000.1fthevalueobtainedbyAD-conversionislessthan1
b:(tニ)く19itisalteredthroughacomputニertobi(tニ)゜1.Hence,
andthedynamicrangeofAD-conversionbecomes60dB.
Theassignmentoftypeforshadeexpressionofspectrain
Fig.3.10isasfollows.
-----…・-------・----・--・--…・―blank
■At)<10ヘ.--..…...…….-........_...
- - I
![Page 53: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/53.jpg)
CHAPTER4
AnalysisofJapanesevcvUtterancesofFive
ニHaleSpeakers
4.1Introduction
Thereasonthatwemakesyllablesequences
vowel(CV),vowel-consonant-vowe:L(VCV),etc.
consonantandvowel
articu:Lation
-50-
consonant-
anobjectofthe
basicanalysisoftニhespeechsoundsisthatthesephonemes
arenotutteredindependently.Itis,of
course,basicallynecessarytoinvestigatethecharacteristicsof
thephonemesutteredindividually.However,itismoreactualto
investigatethecharacteristicsofeachphonemeinthesyllable
sequencesdescribedabove,since,particularlyinJapanese,consonants
areseldomutteredIndividuallyタbututteredrespectivelyinthe
formofthesyllablewhichaccompaniesavowel。
lieinvestigatedthecorrelationsofthesephonemesco-
andtheindividua:Ldifferencesbetweenvarious
speakers,makingthesyllablesequencesofvcvtypeanobjectof
ana:Lysis。
Althoughformantfrequencieshavebeenutニi:Lizedtraditionally
forthiskindofresearch,wehavebeenkeenlyfeelingrestrictions
onitsana:Lysis,asmentionedinChapter1,andinsufficientpoints
initswayofexplainingtheco-articulation.工nthisstudy.
however,ithasbecomepossibletodealdirectlywiththespectraof
consonantsaswellasthoseofvowelsbyintroducingamethodof
multivariateanalysissothatwecouldbereleasedfromtherestric-
tions.Furthermore,weinvestigated,thistニime,notonlytheCO-
articulationsbutalsotheInfluencesofspeakers.Thisfactmay
notbeseeninanyotherresearch.
![Page 54: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/54.jpg)
Now,inthischapter,v^econsiderdthecomponentsofspectrum
distributionsofavcvwordatvarioustimepointニsasthecomponents
ofmu:Lti-dimensionalvectorsatfirstニ;performedmultivariateanalysis
ofvariancewithfourfactors speaker,initialvowel,consonantニ
andfinalvowel;analyzedthechracteristicsofco-articulationsand
individualitニiesofspeakers;andtreatedthemquantitatively.'
Asaresu:Ltofcomparingthevaluesofthefactor-effects
(obtainedfromtheresultsofthisanalysisofvariance)withthe
discriminationscoresofeachfactor,notionof"Direction"aswell
as"Amount"wasfoundtobenecessaryforexplainingthesefactor-
effectsforthefirsttime.Wealsoclarifiedthedifferencebetween
tニhisanalysisandtheprincipal-componentanalysis。
Others,wemaderesearchesfor,arethedifferencebetween
informationincludedinspectraandthatincludedinformants,and
thecorrelationbetweeneachsectionofvcvwordandfinalvowelby
t:hemethodofthemultipleregressiontheory.Wechosenasalsounds
forC,inthischapter,inordertomaketheanalyseseasy.Thisis
becausenasalsoundpossessesthebeststationarinessamongconsonants.
4.2EstablishmentyofExperimentalObjectsandFactors
SupposeaViCV2(vowel-consonant-vowel)wordlike/ame/.
Choosingoneof/a,i,u,e,o/fortheinitialvowelViandthefinal
vowelV2,respectively,andoneof/in,n,o/fortheconsonantC,we
makeallthecombinationswiththem.Then,75kindsofwordsare
equipped.Wemade,further,375wordsasmatニerialsfortheanalysis
byasking5adultmalestニouttertニhese75kindsofwordsinthesimpli-
fiednonreverberantroom.Schematicrepresentatニionofv1CV2word
byamplitudeisseeninFig.4.1..9pointsofstationaryortransition
partニsofeverywordarechosenanddenotedt=tl,・・・,t9,inturn,upon
visualobservation.Presumethespectrumcomponentsateachtimeto
-51-
![Page 55: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/55.jpg)
bevectorcomponents,andvectorscorrespondingtothetimettobe
x(t).
apmiTduiY
-一単一二
t. t2t3t4tst6t7t8t9
Fig.4.1Schematicrepresentationofv,cv.wordand
definitionofti.
t=tl:stationarypartofv1,
t=t3:boundaryofV,-C,
t=t5:stationarvDartofC,
t=t7:boundaryofC-V,.
t=t9:stationarynartofv2.
Ifiiseven,ti=(ti+i十ti-i)/2.
t
Weperformedmultニivariateanalysisofvarianceforfour-factor(
speaker,Vi,CandV2)designvithsingleobservationasTable4.1upon
assortingthevectorswhichcorrespondtothesametimetfroma11
materials.
Table4.1Multivariate。alysisofvarianceforfour-factordesignwithsingle
・observatioa
Factor Level Maineffect No.oflevels
A:speaker Ai αi i=1~a(a=5)
B:V, Bj βj j=1~b(b=5)
C:C Ck rv k=1~c(c=3)
D:V2 Di ∂l I=1~d(d=5)
-52-
![Page 56: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/56.jpg)
ΛsdescribedinSection3.5ofChapter3,iftheoutputsofthe20-
channel1/4-octavefilter-bank(20filterswhosecenterfrequencies
cover210upt0。5660Hz)areassumedtobeb.(t)・b2(t)・‥・・bp(t)
(p°20),inorder,b.(t),・..,bp(tニ)representニthephonemespectニraat
timet.Aftニernormalizingthesquaresumofthesecomponentニsat1,
weestablishedp-dimensiona:LvectorZ(t)bytakingt:he:Logarithmof
itscomponents.Namely,wedefinedp-dimensionalvector
゛(t)゜(Xj(t),.."゛p(t))゛ith
Xj(t)=log戸=
よ
=L=
S
=
E2
=一一√匹(4.1)
whichwouldbeusedfortheanalysis.Theamplitudeoutputsofthe
filteranalyzerwereAD-convertedatevery10ms,thenputintothe
computerbyon-lineinrealtime.Thespeechspectニrumpatternsshaded
bychanging:Letterswereplottedontheline一printer,andthenmarked
eitherboundarypointsorstationarypartsuponvisualobservation,
asinSection3.6.
4.3LinearModelandMultivariateAnalysisofVariance
Supposealinearmodel,whichhasthefourfactorsmentionedin
Sectニion4.2,ateveryt(゜tニ19°・・jtg)forvectorsx(t)゜(xi(t),..゜9
Xp(t))thatrepresentthespectra(T゛able4.1)
Xijkl(t)゜μ(1)十a-(t)十βj(t)+rk(t)十∂I(1)+11jkl(1)
(1≦i≦a,1≦j≦b,1≦k≦c,1≦l≦d)一-
abc?げ
1βjマヤ
'ijkl‾N(o・A)9
-53-
dぶ δ1°0,
Λ=Λ(t).
(4.2)
(4.3)
(4.4)
![Page 57: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/57.jpg)
αi(t),βj(t),r^Ct)anddi(t)representthei-thmaineffect
offactorA,j-thoffactorB,k-thoffactorCand兌-thoffactorD,
respectively.Wedetermineμ(t)inordertosatニisfythecondition
givenbyEq‘(4°3),andassume゛ijkl(t)tobeindependentlydistributed
accordingtothep-dimensionalnorma:Ldistribution.
Letting"1"representtransposedmatrix,thebreakdownof
thetotalvariance0(pxp)(matrixofsumsofsquaresandcross
products)becomesasinEq.(4.5).
(Wedisregardtニaslongasthereseemstobenomisunderstanding.)
abcd
Q゛ぶ亙ぶ1F1(゛゛;ijkl‾゛‥‥)゛(゛ijkl"゛‥‥)
゜QI十Q2十Q3十(れ十R,
where
QI=
and
====
23J-
QQQR
aΣ貨乞公がΣ心丑一
・1・Jkl。1
X....こ
Z・j‥゜
(4.5)
(゛i‥.-z‥‥)1(zi・‥-z‥‥)
(゛・j・.‾゛‥‥)゛(゛・j・.-゛‥‥)
(z.・k.-z‥‥)゛(z‥k.-z‥‥)
(X...I-z‥‥)i(z‥・I-X....)
Cd
1=1k=iぷ(゛iikl‾゛J...゛.j..゛‥k.‾゛...1+3゛‥‥)゛
゛(゛iikl‾゛i‥.-゛・j‥-゛‥k.-X..・1+3x....),
膳縁・聶
俗士
■*^ijkl'zi'¨゜占
j
t
l
J
I
A
・^ijkl'
dlabdlabcぶ
1゛ijkl'゛¨k°゛涵Jぶ1ぶIF1゛ijkl'゛・・・'abcふぶkl?ijkl゛
PLQajQajQi*.representthevariances(Matricesofsumsof
squaresandcrossproducts)correspondingtofactorsA,B,CandD,
respectively,andR(Matrixofsumsofsquaresandcross
products)doestheresidual。
Now,wesha:LIdevelopatestofthehypotニhesisforfactorAat
timet(゜th・...tg)thata:LItheeffectsofAiareequal(thereisno
effectofA);
HA(t):α1(t)゛‥‥.゜ぼa(t)=0.
-54-
(4.6)
![Page 58: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/58.jpg)
V7ecantestthehypothesissinceitispossibletoprovethatthe
likelihoodratiocriterion
リ゜{n-£2-y(p+£^+1)}log!≒も旦L (4.7)
isdistributedasymptoticallyaccordingtoχ^-distributニionwithpjと1
degreesoffreedomundertheconditionn=a・b・c・d,又,1=a-1,
聡十几2°8十b十c十d-3,n-a-b-c-d+3≧p-`゛hennissufficiently
large.(SeeAppendixA.)WeobtainedtestニcriterionVforeach
factoratt=tl,...,tgbasedonthematerialsmentionedabove(Fig.
ム.1).Asthedegreesoffreedomofthefactorsaredifferentfrom
eachother(SeeTab:Le4.2)
Table4.2D^reesoffreedomofQi;
detrreesoffreedomofRisn-a一b-c-d+3.(n=abcd)
Factor A B C D
Qi Qi Q2 Q3 Q4
瓦 a-1 b-1 C-1 d-1
p瓦 P(a-l)lp(b-1) p(C-1) p(d-1)
瓦十瓦 a+b十c+d-3 a十b十c十d-3 a+b+c十d-3 a+b+c十d-3
andsoaretニhevalueofViosignificantlevelofx2testdifferentfromeach
other,wesignifiedthenormalizedcritニerion
v' -
Valueofり
significantlevelofχ2testcorre-;
spondingtothedegreesoffreedomof1リ)(4.8)
inFig.4.2.FromFig.4.2,itcanbesaidthat
(1)Theeffect(maineffect)ofthespeaker-factoristheぐLargest
amongfourfactorsatthestationarypartofnasalconsonant:
(2)Theeffectofthevowel-factoristhelargestatthestationary
partofvowelamongfom:ヽ'factors.
(3)Althougharesultdoesnotinsistthataneffectofv2(VI)
-55-
![Page 59: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/59.jpg)
is・observedatthestationarypartofVi(V2)(り'<!),itmaybe
necessarytoschemeamoreaccurateexperimentinordertoassert
it.
(4)AtthestationarypartofVi(V2),theeffectoffactorCis
observedbutitissmallerthanthatofspeaker-factor・
(5)TheeffectoffactorCismaximumatthepartofCamonga11
sections.TheeffectofCatV:,islargerthanthatofCatV;.
。nU0U3JU3pOZl[UUUO|^
30
20
10
1
tl t2
Fig.4.2
t3t4t5t6t7,‥t!L・t9
Multivariateanalysisofvariancefor
four-factordesignwithsingleobservation.
4.4Relationbetweennormalizedcriterionxヌ'anc!
DiscriminationScore
`ThelargerthenormaユizedcriterionV',forfactorAisthan:L,
themore∧(ziり.(i°1~a)differgreatlyfromeachother・-
Hence,whenacertain‘X(oneofzijkl(t))isgiven.itmaybeeasier
-56-
![Page 60: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/60.jpg)
todiscriminateinwhichcategory(tニhatcorrespondst0levelAiinthe
caseoffactorA)xbe:Longs.Wetニrieddiscriminationineachcaseof
allfactorsattニ(=tニ1,...,t9)byusingthedistancewhichis"basedon
quadraticforminordertoinvestニigatetherelatニionbetニweenV'and
discrimination.Forexample,wecalculatedthemeanvectorATi...(t)
andtニhesamplecovariancematrixS;(t)forthecategorywhichcorres-
pondstoeachlevelA:(i°1p・・,a)offactorAattimet,usingx's
whichbelonginthatcategory・
Xj...(t)=
Si(t)=1
-bed
1bcdbedぶぶEXijkiCt) (4.9)
bcd石ぶ
111F
1(‘゛ijkl(t)-Xi...(t))'(゛ijkl(t)-゛i‘..(t)).
(4●10)
Assume,now,thatwediscriminatethatgiven゛(t)(oneof°゛ijkl
belongsinthecategoryAiwhere"i"makes
(t))
(X(t)-Zi‥.(t))Sマ(t)(Z(t)-Zi‥.(t))1+10glSi(t)|
(4●11)
minimumamongi=l,°・・タa.
Ifeverycategoryhasthenormaldistributionandfrequenciesof
occurrenceofeachcategoryofAI~Aaarethesame,thenotionof
discriminationofEq.(4.11)correspondsthat"Thecategory,inwhich
theprobabilityoftheoccurrenceofgivenj;(t)isthelargest,is
Ai."(Bayes'theorem.)Thediscriminationscoresforeachtimeand
eachfactorobtainedbythisjudgementarepresentedinTable4.3.
Thediscriminationscoresofthefactors,whoseeffectswerethe
largestinanalysisofvarianceateachtime,arenearly100%,and
thatofthefactors,whoseeffectsarethesecondlargest,arenotニ
sounexpected,too.AlthoughC-effectsatthestationarypartof
(七゛ty)`ct-t^yふぐヽnasa:L^andatthebot!ndarybetweennasalandvowelareconsiderably
!"・rl'j八smallcomparingwiththelargesteffect,weshouldnoticethatthe
-57-
![Page 61: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/61.jpg)
0.5 301●2510
Normalizedcriterionμ/
Fig.4.3Discriminationscorpsvs.normalizedcriterionμ/
-58-
discriminationscoresofthenasalsare95.5%and89.1%,respectively
Table4.3Disariminationscoresineachfactor(%).
FactorNo.of
categoriestlt2
1
t3
1
t4 t5 t6 t7 t8 t9
A:speaker 5 100'99.8i99.811001100 99.899.8 99.5 98.7
B:Vi 5 10010096.5179.5'69.9 64.5 55.7l
52.853.1
C:C 3 69.6フ3.1j85.1 95.2 95.5 95.5 89.1180.377.91
D:v. 5 54.7 53.3 58.1 64.0 72.0 82.7 96.011001
100
Wevjillspeakfurtherofthefactthatthereistherelation
describedinFig.4.3betweenthediscriminationscoresobtainedhereand
~
thenormalizedcriterionり≒Thus,itisfoundthatv'hasclose
relationswiththediscriminationscorethoughV'meansoriginally
thestatisticsfortestingthehypothesis.
100
90
8070
(心)S8SUOIJBUimUOSIQ
60
50
![Page 62: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/62.jpg)
4.5GeometニricRepresentationof:iulti-FactorDistributニions
AssignifiedinSection4.4,itissupposedthatthedirections
ofthedistributニionsofvariancesofeachfactormaybedifferentニ
fromeachotherbythereasonthatthediscriminatニionscoresofthe
secondandthethirdlargestニfact・ors(whichpossessconsiderablysmall
valueofeffectascomparedwiththelargesteffect)donotbecome
worse.AsitdealswithratiolQi十判/|r|,thatis,theratioof
thevarianceofeachfactortotheresidualvariance,inanalysisof
varianceasdescribedbyEq.(A.7),wecanobserveonly,sotospeak,
therelativelargenessofthedistributionsofeachfactor。
Therefore,weschemegeometricinterpretatニionofthedistribution
asfollowsinordertoclarifytherelationbetweentニhedirectionsof
distributionsofeachfactor。
Atfirst,ifwegenerally:Letμ,Λbetheexpectedvectorand
tニhecovariancematriχoftheprobabilityvectニorofZ(1xp),respec-
tively,wemaythinkthat
(z-μ)Λ‾1(.x;-μ)゛=p+2 (4.12)
expressesgeometricallythepatternofthevarianceofX,whereEq
(4.12)representsaconcentrationellipsoidforZasmentionedin
section2.2.3.
Inthischapter,wearegoingtosignifythevarianceofzby
utilizingthefollowingellipsoid(4.13)whichissimilartothe
aboveellipsoid(4.12)(similarityratio1/√iこ<-2).
(z- いか1(z- μ)'=1. (4.13)
WhereぶandAarethemaximumlikelihoodestimatesofμandA,
respectively.(LetZI・'‥9Xnbesamplevectors.then
μ=X.=nRli
―一n ATj,andxぺyい乱か.゛i‾゛.)'(゛i-゛.).)
-59-
![Page 63: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/63.jpg)
Thereasonisthat(4.13)iseasytounderstandnumericallybecause
thedistancebetweencenterμandthepointofintersection(produced
bytheellipsoidrepresentedbyEquation(4.13)andtheprincipal
axis
八
eigenvectorcorrespondingtoeigenvalueofΛ)isjust
ヽ一
べifoneofeigenvaluesofxisa;(i=l~p).(SeeAppendixB).
Next,asshowninAppendixC,Qi十Rcanbethoughttoexpress
thevarianceoffactorA,andi-(Qi十R)canbealsothoughttoexpress
thecovariancematrixoffactorA.
V'e,further,projectthevarianceQI十Rontothenewvectorspace
obtainedbynormalizingtheoriginalvectorspacebytニheresidualR.
Supposethenonsingularlineartransformation
X→
_主
i=z(R/n)2, (4.14)
whereRistheresidualvariance,andn(=abcd)issamplesize・
(Zandiarevectorsintheoriginalspaceandthenewspace,
respectively.)
Then,thecovariancematrix士(QI十R)istransformedto
i-(Ql+R)=r"を(QI+n)f‾1
(SeeAppendixD.)
SothatRitselfbecomes
尨
n゜R‾
1RR‾
士
゛Ip
(4.15)
(4.16)
(Notethat・(QI十R)R‾lisalsoconsideredtobeanothernormalization,
butitisnotalwayssymmetricmatrix,sothatitisunsuitablefor
geometricexpression.)
WewillcontinuetodiscussEq.(4.15).AsresidualmatrixR
issymmetricandpositニivedifinite(provided,n-a-b-c-d+3≧p.See
60-
タ .
![Page 64: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/64.jpg)
1-l-・-1・rq.(A.7)ofΛppendi芦Λ),P.2'exists.WhereR2R2°RandRistニhe
inversematrixofrび.(SeeGLOSSARYOFSYMBOLSastothedefinition1、
ofR^).^'atrixQiissynunetricanditsrankisa-1(provideda-1≦p.See
Section2°2-4),andR2isalsosymmetricandrealnonsingular...1_.L
Accordinglyn^Oi?.2becomessymmetニricanditsrankisa-1.Hence,
.1tfollowsthattheeigenvaluesof
-1-_-Lz゛
areOi>O2>゜゛゜>(:ya-1>O(withprobability1)and(ya°‥.゜op°0
Thentheeigenvaluesof
ぎ1'(QI十R)R‾1al=λa' (4.17)
become入1>λ2>・‥>λa-l>l,入a°...゜入p=:L.Because,入=0+1isshown
fromthefact
Also,
λa'=[R
斗特りけ
p]α゛゜R
、_1
((h十R)R‾7
‾礼IR‾
y
α゛十lpα1°oa'十a'=(a+l)a'
=λ1・入2・・●●入a-1(4.18)
Whenjlisnamedthemaximumeigenvalue,andtニheeigenvectorOi
correspondingtoitisprovisionallynamedthefirstprincipalaxis
oftheellipsoidwhichisexpressedby
心i (61ナli)]-lil=i[r"2"(Qi+R)R2fl?=1 (4.19)
andwhichrepresentsthevarianceoffactorA,y/χ‘iandOiare"consider-
edtobetheamountandthedirectionofthesubstantialproportion
ofthevarianceoffactorA,respectニively・
-61-
![Page 65: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/65.jpg)
Attq:フ(attheC-V2boundary),themaχimumeigenvaluesandthe
firstprincipalaxesofspeaker-factor(Qi十R),C-factor(Q3+R)
~~andv2-factor(C八十R)・werecomputed,thentheresultsbecame(2.42,
a1)タ(:L.429Cj)and(2.5^,di),respective:1-y.Theseareillustratedin
Fig.4.4。
a1_
C1
di
Vz-factc〕r
Fig.4.4Geometricrepresentationofmulti-factor
distributions(f=t7).
TheIntersections,madebythreeellipsoids(thatarerepresentedby
~~i[言(Qi十R)]‾lX'=1(i=l,3,4))andthreeplanestニhataredeter-
minedbya,&Ci,Ci&dianddj&Ojarealsoillustratedinthefigure.
(HowtoobtaintheintersectionsisinAppendixB.)
TheangleebetweenvectorsXandUisdefinedby6ふcos'x‘が・
Theanglebetveen,「Olandc.is93,andsoon.Besides,
匹に
甲
i(-4?R)‾1i'゜ilj'4F°11
-62-
![Page 66: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/66.jpg)
fortheresidualH.Fromthisdescriptニion,itcanberealizedthatthe
principalaxesoft二hevariancesofthesethreefactorsmeetnearly
atrightangleswitheachotニher:Similarly,thedistributニIonalrela-
tionsbetweenthespeaker-factor(Q*,十R)andC-factor(Q,十R)at
t°ts(thatisthestationarypointofC),andthatbetweenspeaker-
factor(Qi十R)andVz-factor(Q'4十R)att=t9(thatisthestationary
pointofv2)areasFig.4.5.Fromthisfigure,wecanunderstandthat
thediscriminationscoredoesnotdecreasesincethedirectionsof
thevariancesaredifferentfromeachothereven・iftheamountof
varianceisslight。
Fig.4.5Geometricrepresentationofmulti-factor
distributions(t=t5,t=t9).
4.6ComparisonWitニhPrincipal-ComponentAnalysis
Klein,PlompandPolsexpressvowelsbyusingthefirstfour
principalcomponentsofaprincipal-componentanalysisandarguethe
distributionofvowelsandspeaker'sindividualities,regardingthe
-63-
![Page 67: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/67.jpg)
amplitudeoutputsof18-channel1/3-octavefiltersasthecomponents
of18-dimensionalvector.(Theydealtwith600utterances
-2十Q
1'aコ匹Fふ
日U=
J=lk=II=1
1abcd
abcd,?i^ふふ(゛ijkl"‘X..")゜0
-64-
12
kindsofvowelsbypronouncedby50malespeakers,)However,their
explanationsarenotdirectbecausetheyconsiderprojeCtニionof
vowelsontheplanedeterminedbytheprincipalaxesofaprincipal-
componentanalysiswhichmakeneitherthevowel-factornorthe
speaker-factormaximum.
Weobservedthefollowinginordertoclarifylhoweach
factorisexpressedbyaprincipal-componentana:Lysis.
TotalvarianceQofEq.(4.5)isexactlythesameasthesamp:Le
covariancematrixSwhichisusedinaprincipal-componentanalysis.
Namely,Q°nS,wherenisthenumberofx's.Astheeigenvectorsof
Sarearrangedinorderoflargenessoftheeigenvaluescorresponding
tニothem,andarenamede.,e2,‥・9respectively,theinnerproductby
i(゜゛ijk1(t)-゛・...(t))andCmmakesthem-thprincipalcompone吐of
X(eiei'=1,eiej・=0.i≠j).Ifthe2-dimensionalvectormadeby
thefirstandthesecondprincipalcomponentsisrepresentedbyAT,X
denotestheorthogonalprojectionofionei-egplane.(Thevariance
explainedbythefirstandsecondprincipa:Lcompenentinthiscaseis
83Zofthetotalvariance.)ThebreakdownofthetotalvarianceQby
usingisimilarlytoEq.(4.5)isas
(4.20)
whereQj,Rhavethesreletn:nFasQiandRinEq.(4.5)do,andp=2
工nthiscase,x....=-瓦石Jぶljl1J1占iijkl゛0'because
-Similartotheprecedingsection,Rrepresentstheresidualvariance・
--VarianceoffactorA,forexample,canberegardedasQi十R.And
letiぞlil=1,i【サ?;-(可1十瓦)]‾lil=1betheellipsoidswhichdenote
![Page 68: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/68.jpg)
.~な・
thesevariances,respectively・
TheillustrationsoftheellipsoidsaredrawninFip。4.6.
eiSpeaker
-factor
Fig.4.6Principal-componentanalysis
(t=t5,t=t,).
e2
Oneofthatdenotesspeaker-factor(り'1十R),factorC(Q3十R),and
residualRattニ=t5,andtheotherdenotesspeaker-factor(可3十i)。
一--factorv2(Q4十R)andresidualRatt=t9.ComparingFig.4.6with
Fig.4.5,itisunderstoodthatthefactorswhichhavethesecondand
lesslargestニeffectsarelostinthevarianceofthefactorwhich
hastニhelargesteffect,andthattheratiosofthesevariancesto
theresidualvariancearenotsofavorable。
Tobebriefastocharacteristicsoftheprincipal-component
analysisinthisstudy,whatニismainlyexp:Lainedisaboutthesuperior
factor,butitisdifficultfortheinferiorfactorstobeexplained
sufficiently.Itニisjustlike,"Theweakbecomethevictimofthe
strong.μ
-65-
-・・-
![Page 69: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/69.jpg)
4.7DiscussionaboutResidual
Bythemodel0fEq.(4.2),itispresumedthatEq.(4.4)
Eijkl°zijkl-μ-(zi-βi-7k‾δl(4.21)
hasthep-dlmensionalnormaldistributionN(0・A).lfμ・町・戸j・rk・
∂larereplacedbymaximumlikelihoodestimatesj:・・・・9Zi・・・Xt....・・・>
and・゛...{-x(Eq.(A.2)ofAppendixA),then^iiklreplacing^ijkl
WJ-.●●7-一匹.I●/'‘゛″‾'‾:'I‘"~-----・.-し●l………77'`‾・・・‥
ゾフJ……"'‘してこいj………、し……….・1-
■'ijkl°'系jkl‾゛i…‾゛・j‥‾゛‥k.‾゛…I+3乱…ト(4.22)
isdesiredtobedistributedaccordingtoN(0,Λ).Now,wearegoing
toinvestigatethedistributionalpatternofeachcoordinatecomponent
upontransformingcoordinateinorderthatΛ=(1/n)Rbecomesa
diagona:Lmatrix.WeutilizedA,maximum:likelihoodestimateofΛ,
insteadofΛwhichisunknown.
Arrangetheeigenvalues'ofΛinorderof:Largeness:denotethem
byai^,...,a^p;and:Letrl""゛'■prepresenttheeigenvectorswhich
correspondtotニhem.
Ifthem-thcomponentof
恥jkl°(JZijkl:'・l'"'゛・'^ijklV)
issupposedtobe^m,iikl■Thesecomponentsareuncorrelatedwith
Λeachothe:°.If^ijkl‾N(o・Λ)・
Ym,ijkl/(Jm~N(0,:L):Singledimensional
standardnormaldistribution.
isobtained.since
八
,
'
ym.ijkl-'ijkl^m'~N(0,‰Λlml)=N(0,(y2m).
Theobservedcumulativefrequencydisti°ibutionSN(X)ofym,ijkl./(Jni・
plottedOlbnormalprobabilitycoordinates,becomes,forexamp:Le,as
Fig.4.7,whent°t7andm°1.ThedistributionshowninFig.4.7is
-66-
匈
![Page 70: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/70.jpg)
iswell-approxir.atedbytニhenormaldistribution.
99
9 0
7 0
5 0
3 0
1 0
{%}uoijnqujsipXouanbaaiaAHBinmnopoAjasqQ
1
● ●
/
//
ソ
ダ
KS-test
.-'
5亙
g
-
1
/
,////ダ
//●/ ●
ヅ//
J//
ツ/
/
/X´
/
うぐ
ノシ/
/
~//
/
/
∧-3 2
Fig.4.7
-1012
ylか1
Observedcumulativefrequencydistribution
ofthefirstcomponentoftheresidual.(t=t7)
3
Toexaminewhethertheseobservedcimiulativefrequencywereobtainedfrom
thepopiilationhavingthenormaldistributionornot.wewillutilized
KS-test(Kolmogorov-Smirnovone-sampletest).(SeeAppendixE.)
LetFo(X)denotethecumulativefrequencydistributionfunction
ofthestandardnormaldistributionN(O,1),thatis,thestraight
lineofFig.4.7,andthebandofFo(X)±D(z(fromthemathematical
tableofKS-test,thecriticalvalueDOC°7ZwhenN°375,level0f
-67-
![Page 71: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/71.jpg)
significanceCX°5%)isbetweentwodottedlinesinthesamefigure.
Thefactthat‘the:LinedenotingSN(X)doeSnotcrossbeyondthe
dottedlinesrevealsthat^m,ijk1/Cmhasthestandardnomaldistri-
butionN(O,1).Fig.4.8showstheresultofobtainingDforeachymタ
ijkl/びm(゛゜1~p)att°t7andofcomparingthiswithsignificant-
levelvalueDa'EachvalueofDissmallerthanDsothatallof
theseseemstohaveunivariatenormaldistributions.Although
^ijklisnotalwaysdistributedaccordingtop-dimensionalnormal
distニributionevenifeverycomponentニof痢jklhastheunivaritenormal
distribution,theassumptiononresidualZijklmaybealmostニappro-
priate,consideringfromtheseresults.
Q
%
10
U0HBIA9P
mnuiixem
CriticalvalueofKS-test
1 5
Da=7%
10
Them-thcomponent
(α=0.05)
15
一 一
20
Fig.4。8MaximumdeviationDofeachcomponentandthecriticalvalue
ofKS-test
4。8AnalysisbyFormaりtFIでequency
m
Ueanalyzedthesamespeechmaterialsinawaysimilarto
Section‘A.3,considering‘thefirst,thesecondandthethirdformant
frequencies(represented・by‘・Fi,F2andF3,respectively)asthree-
dimensionalvectors(P=3).''FormantFrequencyExtractionbyainverse
FilterandMomentCalcu:Lation';(12)reportedbyNakatsuiandSuzuki
wasusedforextractingtheformantsfromthespectra,obtained
throughthe1/4-octavefi:Ltersmentionedbefore.Todescribethe
accuracyofthismeasurement,theerrorofeχtractingthefrequencies
f -
-68-
![Page 72: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/72.jpg)
inthecaseofthesyntheticspeechsoundislessthan3.6Z.(13)
However,wesometimesreliedonvisualinspectationsinceitisnot
alwayseasytニoextractintニheactualcasewhichcontニainsnasaユizedvowels。
Formantfrequenciesobtニainedintニhesamedatamentionedaboveat
t(=ti,t3,t7,t9),andwecarriedoutニmultivariateanalysisatthese
tines.(Only/a,u,o/areusedasobjectsforv2inthecaseof
formant.V1&Carethesameasbefore.)Theillustrationofthe
ellipsoid,tニhatニrepresentsvarianceoffactニorinthemeaningof
Sectニion4.5,isasFig.4.9whent°t7,t9・
V2-£ictor
1aA。‐
~~Q4+R
t-=:t7
Fig.4.9
Speaker
-factor
還
-69-
di
Speaker-factor
Geometricrepresentationofmulti-factordistributionsin
thecaseoftheformantfrequencies(t=t7,t=t9).
Keanwhi:Le,Table4.4givesthediscriminationscoresobtainedby
formantinthemeaningofSection4.4.BycomparingFig.4.9with
Fig.4.4andFig.4.5,itニispossibletoexplainthereasonthatthe
discriminationscoreforeachfactorInthecaseofusingformants
becomesworsethaninthecaseofusingspectra。
Generally,itcanbesaidthatformantニisrathera"vocalic
![Page 73: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/73.jpg)
factor.'^Theinformatニionsonanyfactorotニherthanvowel-factor
arebeingdecreasedascomparedwiththecaseofspectrumdistribu-
tion.
Teble4.4Discriminationscoresineachfactor(%):
byformantfrequencies.
FactorNo.of
categoriestl t3
1
t7 t9
A:speaker 5 43.1 45.3 56.0 46.2
B:Vx 5 96.0 84.0 30.7 24.0
C:C 3 37.3 49.3 64.4 49.3・
D:V2 3 37.3 39.6 77.3 99.6
4.9RegressionEstimatebyUsingFinalVowelsand
AnalysisofVariance
Intheanalysisuptothesectionabove,thespectrumsections
atdifferenttimepoints(t=ti~19)havebeentreatedasindependent
sectionsofeachother.Here,oneaspectニofthecorrelationbetween
thespectrumsectionsatvarionstimepointsinvcvwordsisexamined。
Supposetheequationsofmultivariatelinearregressionateach
timetinordertoanalyzesimilarlytoSection4.3aftere:Liminating
theinfluenceoffinalvowelfromeachspectrumsection。
LetZ(X(t9)(astationarypartoffina:Lvowel)representaknown
vector,andXn(t)observationvector;
presume
'X°a (t)(1xp)=17(t)+・を(t9)・B(t)十e^(t)(t=ti,...,t8)
ε
(X (t)~N(O,A),B(t):p゛pmatrix;Λ゜Λ(t);
-70-
(4.23)
![Page 74: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/74.jpg)
andestimateV(t)andB(t),where°raisoneofぷiikl^anda°1゛"9n゛
Onlv/a,u,o/areusedasV2inthissection.(Vi,Carethesameas
inSection4.3.).Pence,n=a・b・c・d=5×5×3×3=225.Supposingx゛(pxn)
゜(xUt),...,xn'it)),Z'(p+lxn)=(zi',22≒‥.,ぢ),andZ¢(1×p+l)=(:L,*a
(t9)),vdt)andB(t),maximumlikelihoodestimatesofD(t)andB(t),
respective:Ly,areobtainedfromEq.(U.23).
t)
t)
=(Z゛Z)‾1(Z゛X) (4.24)
(SeeEq.(2.26)ofSection2.3.)
Byusingthese令(t)andB(t)・esti°ate゛(t)from・゛゜(X(t9)illeach゛)゛d・
oneofthesameVCVwordsusedinEq.(U.23),andlet£(t:)bethe
estimateof厦t).
Theresultsofthemultivariateanalysisofvarianceforthediffer-
&
enceZ
(X(t);
lo,(l)yx'ojt二) 礼収)゜゛Cz(tニ)一令(1ニ)-゛a(tニ9)冷(t:)
(similarlvtoSection4.3)isdrawnasFig.4.10.(Dottedlinesof
ミthesamefigureindicatetheanalysisfor・`゜(t)itself.)
Asaresult,V2andthespeaker-effectdecreased
Therfore,
●@
4(1)theeffectofCbecamethelargestニintheregionbetweenCand
V2amongalleffects:
(2)tニherelativevalueofeffectVitoothereffectsincreasedat
partVi.
Thisrevealsthatintensivecorrelationisseenbetweenv2and
speaker-effect.
(Thecorrelation[canonica]Lcorrelation)betweenvectorsZand
yhascloserelationswiththeregressiontheorybetweenぷandy。)
-71-
(14)
![Page 75: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/75.jpg)
15
1 05
。nU0U911J0pazitEuuoM
1
Vi---
V一十-C→V
゛≒
へ`ヽ
、
-〃〃--
仁ど
"""
1
1
/
/
11
1111
/
/
1
/
1
"
/
-・
/
/
/
/
/
/
/
- 一 一
/
C四--
i-
/
/
「≒
『゙`
I
//
`χχ
'
ドト
ーー
レド
'
/-―'
χ1?
^I"*"/
4//
ゝ
ゝχ
kゝ---
・ 四
一一- 一 一 一 -
t
tl t2 t3t4t5t6t7 t8t9
Fis4.10Comparisonbetweentheanalysisof
varianceofXaandthatofXa。‘
ldisthedifferencebetweenZα
anditsregressionestimateby。χ:a(t9)
4.10Conclusion
Consideringthespectrumcomponentsasthecomponentニsofmulti-
dimensionalvector,weperformedmultivariateanalysisofvariance
insectionsatvarioustimepointoftheViCV,utterancewithfour
factorsVi,C,V2andspeaker;andcomparedtheamountsofthe
effectsofeachfactorwithinthemselves.Thenweinspectedthe
relationofthevarianceellipsoidsofeachfactoralongtheir
principalaxes;signifiedthatnotionofdirectionaswe:LIasamount
isnecessaryforexplaningtheeffectsofeachfactor;andcompared
theseanalyseswiththeprincipal-componentanalysis.Anotherthing
weinvestigatedbythemethodofregressionestimatewastherelation
72-
![Page 76: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/76.jpg)
betweenfinalvovrelsandeachsectionofwords.Furthermore,weper-
formedsimilaranalysisonthebasisofthree-dimensionalvectors
whichconsistoftheformantfrequenciesextractedfromthesame
materialsasabove,andcomparedtニhesewiththecaseofspectra.
Theresultsconcerningspeechsoundsareasfollows:
(1)Speaker-effectニisconsiderab:Lylarge,whileconsonant-effectis
notsolarge.However,thedirectニionsofthreedistributionsof
tニhesetwoeffectsandvowel-effectmeetatnearlyrightangleswith
eachother:
(2)Intensivecorrelationisseenbetweenvowelandspeaker-factor:
(3)Inthecaseofformantfrequency,theinformationsonanyfactor
otニhertニhanvowel-factニorarebeingdecreasedascomparedwiththe
caseofspectrumdistribution.
-73-
![Page 77: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/77.jpg)
CFAPTER5
AnalysisofJapanesecvcvUtterances
5.1工ntroduction
Oneofthesubjectsofthisthesisistoconsiderquantitatively
theco-articulationsofutterance.工ntheprecedingchapter,v/e
experimentedalonpthesubjectabovefromvariouspointsofhypothesis
basedonvcvsyllables.'。
Howevertheconsiderationintheprecedingsectionwasinsome
aspectsinsufficientbecause:
(1)ItwasdifficulttoinvestigatethecharacteristicsofCO-
articulationitselfbecausetheinfluenceofspeakerwasnixedas
weinvestigatedsimultaneouslytニhedifferencecausedbychanging;
speakers:
(2)Becauseweadoptedthemodelwithoutrepetitionsfortheanalysis.
wecouldnotclarifytheinteractionbetweenprecedingandfollowing
phonemes:
(3)Thenumberofthephonemesequencewastoosmall.'Andthekind
ofconsonantsweretoolim:ited.
:Therefore,weplannedanexperimentparticularlyemphasizing
muchimportancetoconsiderationofco-articulations:
(1)7echpseonlyonespeaker.Thereasonwasthatifwetreated
simultaneouslyvoicesofvariousspeakersinordertoinvestigate
tニhedifferencebetweenspeakersalltogether,itbecametroublesome
tofindoutthecharacteristicsthatwerepurelydependentニonc0.-
articulationsincethecharacteristicswerelostinthefluctuation
causedb'ythealterationofspeaker,thatwas,noise.Instead,as
a!consequenceoftheforegoing,theresultofthatanalysisinclined
-74-
![Page 78: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/78.jpg)
tocontaincertaincharacteristicsdependentonthearticulational
habitニSofthesoeaker:
(2)・Inorder・toclarifytheinteractionsbetweenprecedingand
followingphonemes,weappliedamodel0fanalysisofvariancewith
repeatedmeasurements.りtlaineffect"indicatesthedegreeofthe
effectofphonemeitselfwhichisindependentofcontext,and
'‘Interaction"doeshowmuchtheeffectisdependentontheconteχt.
Interaction:Effectspecifictothecombinationsoftwo(ormore
thanthat)ofprecedingandfollowingphonemes・
Precedingandfollowingphoneme:Notonlyonesclosetoeach
otニherbutalsoseveraloftニhoseskippingoverafewphonemes.
(3)1・7emadeuseofwords0fCVCV-sequence.Thereasonisthatthe
scaleoftheexperimentistoolargeifweusemorethanfivephonemes
inconsiderationofinteraction,andthatthenumberoftwo-syllable
wordisfrequentinJapanese.(Cvrepresentsonesyllable.)Nine
consonantswereprovidedfor.C.I
Inthebeginningofthischapter,wedescribedthemultivariate
analysisofvarianceforfour-factordesignwithrepeatedmeasurements.
Althoughitニwasconsiderablyhardtoproveraathematicallvthemodel,
Mr.Kubo,astudentofKyotoUniversity,co-operatedinit.
Fromnowon,wewi].1describetheexperimentindetail.Consider-
ationoftheresultsoftheexperimentincludesexaminationof
residualsdistributionandevaluationofinteractions.
Ast二heexperimentofthischapterwasprojectedwiththatof
Chapter7,suchresultsofSection7.4shouldbeshownhere.
5.2llultivariateAnalysisofVarianceforFour-Factor
DesignwithRepeatedMeasurements
-75-
![Page 79: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/79.jpg)
Themodel・oftheexDerimentaldesipnofthischapteris':Multi-.
variateanalysisofvarianceforfour-factordesignwithrepeatedme-
asurements."]:tisimpossibletoinvestigatetニheintニeractionsbetween
eachfactorbymultivariateanalysisofvariancewitニhsingleobser-
vation,whileitニbecomespossibletodoifweusetheanalysiswith
repeatedmeasurements.However,thelaborforprovingmatニhematically
themodelincreasesconsiderab:lyascomparedwiththecaseofthe
previouschapter.工nthemodeloftheprecedingchapter,thelongest
expressionconsistsof16terms:whi:Leinthischapter,thelongest
consistsof256terms。
Mr.MasatoshiKubo,astudentofKyotoUniversityco-operatedin
accomplishmentoftニhemathematica:Lproofofthemodelofthischapter.
Sothatonlytheresultofitisdescribedhere,referingtohis
graduationthesisfort1!ebachelardegreeofKyotoUniversitysoas
torealizetheproofyL5)
designwithrepeatニedmeasurements.
Themodelisdefinedasfollows(tisomitted):
゛ijklm(1)(p)゜g
十(zi十βj+rk十∂1
generallevel
maineffect
+Sij+ζik十ηil+∂jk十λjl'1μkltwo-factorinteraction
十μijk十pijl十"ikl十^jklthree-factorinteraction
+91jklfour-factorinteraction
十■^ijklmresidual
(5.1)
where・1≦i≦a,1≦j≦b,1≦k≦c,1≦l≦d3゛d1≦!11≦e.(SeeTable5.1.)
Theconstantsa,b,canddarethenumbersoflevelsofthefactor
A,B,CandD,respectively.Theconstanteisthenumberof
-76-
![Page 80: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/80.jpg)
repetニitions.
Table5.1Mu】tivarjateanalysisofvariancefor
four-factordesignwithrepeated
measurements.
Factor Effectvector Qi
I
J
A (Zi Q.
B βj Q2
C rk Q3
D δ1 Q4
U
AB りj Q5
AC ぐik Q6
AD フ7il Q7
BC ∂jk Q8
BD λjl Q9
CD μkl Qio
oh
ABC りjk Q11
ABD ρ甲 Q12
ACD OiU Q13
BCD りd Q14
|| ABCD Vijkl Q15
Thegenerallevelvectoric(lxp)isdeterminedinorderthat
effectvectorsoC;(lxp)-‘?ijkl・(lxp)satisfythefollow!昭conditions:
a
Σ
1二1
a
Σいb胃
a-=.
^ij-
∂jk°
b
Cβ戸O'ぶ
1rk°O'
l)ajWjiro'ぶ
1ぐik
Cb_CDJ
I
∂jk゛o'胃1
λ
。Σ㎞
jl°
d且 *i=o
aΣ出0。
一に
ら '?il°0
Cd
I゛O'ぶ1μkl°IRμkl゛0
-77-
![Page 81: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/81.jpg)
ab
iEμijk°z;
aC
ぶ(7iklニkミFI
μijkcab
語りjk°o'yijlブPu\=
d
"ikrぶjy
d
Σ
1=1
^ijl=o
bcd
ikl-o'jlJjjkFぷ1'J'^'=l?.'■jkl』0
abcdぶ
191jkにjlJlfijkげぷlfijkげぶ1?ijkにO‘
Besides,assumethat
Zijklm‾N(o・A)・
(5.2)
(5.3)
thatiSタ^ijklmisassumedtobeindependentlydistributedaccording
tothep-dimensionalnormaldistributionN(0,Λ).
Thebreakdownof・totalvariance.
Let"1''denotethetransposedmatrix,thenthebreakdownof
totalvarianceQ(pxp)(matrix6fsumsofsquaresandcross
products)isasfollows:(SeeTable5.1.)
.Q=Qi十Q2十Q3十Q4十Q5十Q6十Q7十Q8十Q9十(卜o
十Qll十Q12十Ql3十Q14十Q15十R,
where上
abcde。,.
Q°ぷ
1
石ぶ
I
F
lr
E
I
(゛ijkhll‾y7¨‾j.)'(゛7ijkllllう゛¨゜゜')リ
蜀y㎜-'-か'--a●・-aW¶a
Q1°bcde
i=i(゛i…'‾゛…‥y(゛i…'‾゛¨…)'
bQ2=acde2
JJ
Q3=abdeΣ
Q4°abcedΣμa
1(゛'j…‾゛…‥)べ゛'j…‾゛…‥)
1
(Z‥k‥-Z…..)'(x..k..-J:…‥)
1
1(゛¨小‾゛…‥y(゜゛;…1°∇‘゛…‥)'
b
Q5°cde且仏(Xij…‾゛i….‾゛・j…十゛…‥)"
(゛ij…‾゛i….‾X.j‥.十゛…‥)・
aCQ6°bdei=ik=i゛i'k¨‾zi‥‥‾ぷ‥k‥十z…‥y
'(zi.k‥一zi….-z‥k‥十z…‥),
ad
Q7°bce且ぶ1(゛i‥I°‾zi・.・‥‾z・‥・I.十X)'
-78-
(5.4)
![Page 82: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/82.jpg)
Here,
Q8=ade
Q9°ace
●(
bcΣΣ(
j二ik=i
●(
ぐdふ
b石
Xi.・l.‾・Xj―X・‥l.十ぷ).
゛・jk‥‾゛・j…‾゛・・k‥十x…‥)'
゛・jk.--゛・j…‾゛‥k‥十z…‥),
゛・j小‾゛・j…-J:...1.十z…‥)'
(゛・j・1・‾゛・j…-゛…1,十z…..).
CdQio=8beJ
I
ぶ
1(ぷ‥kl‘‾z‥k‥‾ぷ‥・I.十z…‥)
'(χ‥kl.-z..k‥-z‥.I.十a・…..)
abc
Qii=de
Q12°ce
i
y
l石ぶ
1(゛ijk"゛ij…‾゛i-k-.~゛'jk・.十゛i….十゛・j・‥十゛‥k‥‾゛…‥y
'(゛ijk・.‾゛り…―X:・k・.‾゛・jk・.十゛i….十゛・j…十゛・・]r..-X…・.)
abdぶふF
I(耳j'1°‾゛ij・..~X:.,‘‾゛'j小十゛i….十゛'j…十゛…1'‾゛…・.)'
・(゛ij・l.‾゛ij…‾゛i・・1.‾゛・j小十゛i….十ぶ・j…十゛…I.‾゛‥…)・
acd
Q13°beぶ1ぶ1ぶ1(゛i゛kl°‾゛i-k"゛i‥I°‾゛¨kl°十゛i…'十゛¨k¨十゛‥小‾ぷ¨…)'
Qi5=e
'(zi
十ぷ・jk‥十ぷ・j・l
X=
z…..y
bcd
Q14°8e胃ぶ1ぶ1(゛゛゛'jkl.‾ぷ・jk‥-ぷ
i・・I.‾゛・・kl.十゛i….十゛・・k・.十z…l.-x…..).
・j小‾゛・・kl-十゛・j…十゛‥k‥十゛…I.-z…‥)'
'(゛・jkl・‾゛・jk・.‾゛・j小‾゛‥kl丿゛・j…十゛‥k‥十゛…1.-*・・…)
abcdふ
j=ik=i
ぶ
1(゛ijkl'‾゛ijk‥‾りj小‾゛i-kl°‾孔jkl°十゛ij…‾トr;.],..十゜r;...
'(゛ijk!・‾゛iik""^ii・I.‾゛i・kl.‾゛・jkl.十゛ij…十゛i・k・.十゛i・・I.十゛・jk・・
十゛・j小十■r..ki.-゛i….‾゛・j…‾゛‥k‥‾゛…I.十゛……)'
abcdeR≒万万;
1
石ぶ
I
E
I
ぶ
1(゛ijkh‾゛ijkl°y(゛ijklm゛ijkl‘)'
1bcde.labce
゛i°‘¨゜i召7ふぶEIぶ1゛ijklm*゜゜゜'゛"小abcei=iJUIぶ1「ぶ1゛りkh'
-79-
![Page 83: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/83.jpg)
1cde12ce
町…cdek=iぶ1Jjijkh'゜゛゜'゛゜j゛に゜77ぶぶJ1゛ijklm'゜"'
゛ijk‥゜士
jよ
゛ijklm'T'
lbeyi・kl-゜百胃Jごijklm'゛"
^ijkl-゜‾シjl・'^ijklm°
Thenaximumlikelihoodestimatesofmaineffectsandinterac
心4;゜X,
(Zi°Xi....―X°・‥‥
八βj°ぷ・\...-x°…‥'
八rk゛Z‥k‥‾Z‥‥.・
I
〈δ
八*ij
八ぐik
八ηil
〈θ
一
一
こ
一
一
X...I.-X..一一●1
'x'ij…‾.x;i….‾z・j…十ぷ…‥'
zi・k‥‾zi‥●●‾z‥k‥+ぷ>
Xi・.l.-Xi・‥.‾Z‥●I.+.X7‥・‥9
jk゛X、jk・.‾ぷ・j・・.‾z・・k・.十z…・.・
2j
八
I゛z・j小‾z・j…‾z‥小十ぶ…・.'
μkl一
一
=k‥~
〈μ
z‥kl.‾z‥k‥‾z…1.+x…‥・
゜x;ijk‥‾zij…‾zi・k‥‾ぷ・jk‥十zi…汁ぷ・j…十ぷ・・k‥‾ぷ…・.・
^ijl°xij°1.‾zij…‾xi・小‾‘x;・j・1.十zi….十z・j‥.十ぶ…I.‾‘x;……・
"ikl°‘x;i・kl.‾'x;i・k・.‾zi‥I.‾z‥kl.+zi・‥.+ぷ‥k・.+x..・i.-x..・‥タ
rjkl°'r・jkl.‾χ・jk・.‾z・j小‾z・・kl.十ぷ・j…十ぶ‥k・.十ぷ‥・1.‾ぷ……・
Pijkrzijkl.‾.x;ijk・.‾zij・l.-Xi・kl.‾x・jkl・
十・^ij..ヽ十Jri・k‥ナzi・・1丿ぷ・jk・.十ぷ・j・!.十.x;・・kl・
‾zi….‾z・j…‾z・・k・.~x..・I.十ぶ・●…
-80-
(5.5)
![Page 84: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/84.jpg)
Thelikelihoodratiotestfornull otニhesis.
Letusconsideratestofthehypotニhesisforthemaineffectof
factorAthatalltheeffectsofAiareequal(thereisnoeffectof
A):
HA(t):α1°゜"゜αa・・f0, (5.6)
1・7ecantestthehypothesissinceitispossibletoprovethat
thelikelihoodratiocriterion
リ={n一又,2一一y(p十几1十:1)}log聯亙(5.7)
isdistributedasymptoticallyaccordingtoX^-distributlonwithpjと1
degreesoffreedomundertheconditions-n=a・b・C・d・e,
聡゜a-1,又・1十又・2°a°b゛c°d,andn-abCd≧p-whennissufficiently
]Large.ThedegreesoffreedomsoftherespectiveQiareshownIn
Table5.2.
Table5.2DegreesoffreedomofQi;乙十島=abed,and
degreesoffreedomofRisn-abed.(n=abcde)
Factor Qi jl
A Qi a-1
B Q2 b-1
C Q3 C-1
D Q4 d-1
AB Q5 ab-a-b+1
AC Q6 ac-a-c+1
AD Q7 ad-a-d十1
BC Q8 bc-b-c十1
BD Q9 bd-b-d+1
CD Qio cd-c-d+1
ABC Q11 abc-ab-ac-bc+a十b十c-1
ABD Q12 abd-ab-ad-bd+a+b十d-1
ACD Q13 acd-ac-ad-cd十a十c十d-1
BCD Q14 bcd-bc-bd-cd十b+c+d-1
ABCD Q15 abed-abc-abd-acd-bcd+ab+ac十ad+bc十bd十cd-a-b-c-d十1
-81-
![Page 85: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/85.jpg)
5.3Establishmentof・ExperimentalObjectsandFactors
Schemedthefollowingexperimentalproiectonthematニerialof
wordspronouncedintheformoftwo-syllablewords-CiViCoV,
(consonant-vowel-consonant-vowe:L) byamaleadu:Lt.
Namely,assumingfourphonemes-Ci,Vi,Co,V2tobe
factors,letthemcorrespondtofourfactors-A,B,C,D-for
analysisofvariancementionedinSection5.2.Nineconsonants/
p,t,k,b,d,g,m,n,o/wereusedforCiandCo,andthreevowels/a,e,o/
forViandv2.ThesenineconsonantsareclassifiedasTable7.10f
Chapter7consideringfromtheviewpointofmannerandplaceof
articulation.The・‘reasonforchoosing:/a,e,o/forvowelswasthat
wecanhaveobservationinallposslb:Lecombinationsoflevelsfrom
eachfactorbyeliminating/i,u/since/tl.tu.di.du/donotoccurin
modernJapanese.゛
Amaleadultpronounced,threetimes,eachwordofthetotal
combinations-729kinds-ofCi,C2=/p,t,k,b,d,g,m,n,o/and
Vi,Vz=/a,e,o/atrandom.Thosewereutteredinthesimp:Lifiednon-
reverberantroomwithinoneday,andthetotalnumberwas2187.(
a°c°9,b=d=3.e=3)
WeaccentedalltheCiVimoradespitethefactthattherearea
10tofmeaninglesswordsInthecombinationsabove.
Furthermore,thesespeechdataarethesameasthoseinChapter
7.
X・7ewillgiveadefinitionofspeechspectraandprocedureof
segmentationbe:low(detaileddescriptionisinChapter3.)
VThenletbi(t)'゛‥‥'bp(t)(p°20)represent,inorder,a『』1it゛de
outputsof20-channelK-octavefilters(bankof20filterswhose
centerfrequenciescover210upt05660Hz)attimet,theyare
consideredtorepresentspeechspectraatthatlt:ime.Afternormali-
zingthesquaresumofthesecomponentsat1,weestablishedp-
dimensionalvectorx(t)bytakinglogarithm・ofitscomponents..・・.j.・●
Thatis.一一‥
-82-
![Page 86: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/86.jpg)
Xi(t)=log{bi(t)ソ
ズ‾Uyで‾こ
} (5.8)
Wedefinedp-dimensio°1vectorx(t)゜(x,(t)・‥・・り(t))withEq.(5.8)
whichwouldbeusedfortheanalysis(p°20).Amplitudeoutputsof
thefilteranalyzerareA-Dconvertニedatintervalsof10ms,then
putintothecomputerinrealtime.0nthepaperoftheline-printer
werepresentニthespeechspectrumpatternsoftheinputwithwhichwe
observedtodeterminethestationarypartsortransitionones.
Thisstudydefinestimestl.t2,,tl3asFig.5.1,correspond-
ingtothestationaryortransitionpartsineachword.
apnjiidiuv
tl
CI
ル
V,
ル
C2
→
V2
t2 t3t4t5t6t7t8t9tiotilt12t13
Fig.5.1c,v,cNiwordanddefinitionofti.
t=tl:stationarypartofc,,
t=t3:boundaryofCi-V..
t=t5:stationarypartofV,,
t=tT:boundaryofM-C3,
t=t9:stationarypartofC2,
t=tn:boundaryofCj-Va.
t=tl3:stationarypartofv2.
Ifiiseven.ti=(ti+1十ti-i)/2.
-83-
t
![Page 87: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/87.jpg)
5.4Resu:LtofAnalysisofVarianceforFour-FactorDesign
Veperformedanalysisofvarianceateacht(=t3,tit,t5,t6,t7,t9.
t11,ti2,ti3)accordingtotheprocedureexplainedinSection5.2.
(t=t,isomittedTaecausethestationarypartofc,ishandto"bedefined‘)臨φ√ぶぷlomputedrespecti々elythetestcriterionfoy?he?死竺芒'sisthattherearenoeffectofeachfactorornointeraction‘between
eachfactor.Wenormalizedりbythevalueofsignificant:Levelas
thefollowingequation,becausethedegreesoffreedomcorresponding
tomaineffectsandInteractionsaredifferentfromeachotherand
soarethevaluesof\1osignificantlevel0fχ^testdifferentfrom・each
other.(SeeTable5.2)
X)1=
(vaユueof\%significaよitlevelofχ testcorre-
spondingtothedegreesoffreedomoiヽり)(5.9)
Thei1:Lustrationofvaluefortestcriterionりlnormalizedaboveis
asFig.5.2.工nthefigure,Ci,Vi,C2,V2representmaineffects,and
VlC2,C2V2thetwo-factorinteractions,andV1C2V2thethree-factor
interactions,andsoon.
Thefollowingisrevealedfromthisfigure:しI'
(1)工nthestationarypartsofeachphoneme,themaineffectofthe
phonemeismaximum:
(2)Theinteractionbetweentwocontiguousphonemesis,ofcourse,
smallerthantheeachmaineffect,butlargerthanthemaineffect
ofanyphonemeotherthanthetwocontiguousphonemes:
(3)Thetwo-factorinteractionbetweenthetwophonemes,thatare
notjustcontiguoustoeachother。isconsiderablysmallerthanthe
interactionoftwocontiguousphonemes.Therefore,theinfluence
(onco-articulation)specifictothecombinationsofthephonemes
whichputafewotherphonemesbetweenthemisSma]。1:
(4)Thereishardlyanyinteractioninthecaseofmorethanthree
-84-
![Page 88: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/88.jpg)
こゝ
200
100
50
2 0
1 0
UOUOJUJ(X)ZI[tUUi()Z
5
2
t3t5t7tot11
Fig.5.2Multivariateanalysisofvarianceforfour-factordesign
withrepeatedmeasurements.
-85-
t13
t
![Page 89: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/89.jpg)
factors.i・7edidnotillustrateanythree-factorinteractionwhose
り゛iSlessthan1,butonlytheeffectsofthree-factorinteractions
ofthree―phonemeV1C2V2andCiViC2.(whicharecontiguoustoeach
other)aresignificantニ:
(5)(1)and(2)aboveshowthesubstantialproportionofthetota:L
variance.Accordingly,inthecaseofadoptinganroughmodelupon
omittingconsideratニionofinteraction,wehaveonlytotakeaccount
oftheeffectsofthelustprecedingandfollowingphonemes.V'emay
notthinkof'theeffectofthephonemesapartfartherthanthem.
(6)Influencesoftheprecedingandfollowingvov;elsofViandv2
onC・2arealmostthesame.(SimilarinFig.4.2and7.4.)Thisdoes
notcorrespondwiththeconclusionofareportobtニainedbyanexDeri-
mentofauditoryperceptiononthematerialofsyntheticspeech
SOUndS.(16)Thereportinsistsontheimportanceofinfluenceofthe
followingvowelV2upontheC2。
Theremaybeadifferencebetweenphysicalandauditorycharac-
teristicsofco-articulation.
5.5ExaminationofResidual
Inthemodel0fthefour-factorexDerimentofthischapter,
■*ijklm~‘N(O・Λ) (5.10)
isassumedfortheresidualtermX..asinEc.(5.3).Now,ifwe-1!klm'
makeuseofthemaximumlikelihoodestimateofeffect-vector
ofEq..(5.5)astheeffect-vectorofEq.(5.l),theactualresidualis
ぷijklm・^ijkl(5.11)
whichisdesiredtobedistributedaccordingtoN(0,Λ).Eyexactly
-86-
![Page 90: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/90.jpg)
(j)iioiincjujsinXouonbojjOAnuiniun。7)己。j()
99.9
99
mo
99
70
50
OOOin
3C^J―I
1
0.1
ノ
1
/
j
ノ
言/
z/KS' est
5§
ズ/
/ケ
が差ダ
//
ブ/
ゾ汐
.ケ´〃〃//
/
ノ //
/ブ1/
/
1
1・
ノ|
3 -2-1-012
y./o1
Fig5.3observedcumulativef。quencydistribution
ofthefirstcomponentoftheresidual(t=t,,)
-87-
3
![Page 91: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/91.jpg)
thesameprocedureasinSection4.7、weillustratethepatterns
ofdistributionandKS-Testoftニheresidualz‥-ぷ‥as1、1]klmijkl.
Fig.5.3andFig.5.4.FiR.5.3denotesthedistributionofthefirst
principal-componentoftheresidualatt°13、andFig,5.4doesthe
maximumdeviationofthefirstuptothe20thprincipal-componentニat
t°13.Judgingfromtheseresults、theassumptニionoftheresidual
Zijklmisnearlyproper.
%
321
nuoquiAap
UiniUIXBTAJ
CriticalvalueofKS-testn-=2.9%foc=5%)
15101520
Them-thcomponent
Fig.5.4Maχimum.deviationDofeachcomponentandcriticalvalueofKS-test.
-88・-
4
![Page 92: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/92.jpg)
●
5.6Evaluatニionof!nteraction
AsitwasascertainedbytheanalysisinSection5.4thatthe
amountofthetwo-factorinteractiondoesnotニeχceedthatofthe
mainfactor,V7eare,now,goingtoconsiderhowtheresultofthe
analysisiswhenwedisregardtheinteraction.Usingtニhesamedata
asusedinSection5.4,provideamodelforanalysisofvariancefor
four-factordesignwitニhrepeatニedmeasurementsinwhichalltheinter-
actionisdisregarded.(SeeEa.(5.1)andTable5.3.)
Table5.3Multivariateanalysisofvarianceforfour-factordesignwithrepeatedmeasurements
inwhichaUinteractionsa・disregarded,anddegreesoffreedomofQi・
DegreesoffreedomofRisn-a-b-c-d+3(n=abcde),and
乙十島=a+b十c十d-3.
Factor A B C D
Maineffectvector 哨 βj ric δl
Qi Q. Qj Q3 Q4
瓦:訟ご認(,, a-1 b-1 C-1 d-1
where,
゛ijklm゛μ十(Zi十βj十rk十∂I十^Uklm・
ab
Σ(Zi°IΣβj゛i二1j=1
cd
°0,
k=ll=1
(5.12)
(5.:L3)
:p-dimensionalnormal
!ijklm‾N(O・Λ)distribution.(5.14)
NowwehaveabreakdownofthetotalvarianceQforthismodel.
abcdeq°ぷ
1且ぶ
1ぶlnΞ
1(゛iiklm゛…¨)'(゛ijklm゛…¨)
=Qi+Q2十Q3十Q4十旦,
QI°bcdeぷ1(゛i…゛‾゛….J'(Xi…゜‾゛…・.)
b'
Q2°acdejE(゛゜jよ‾゛゜゜¨゜)'(゛.j...゛).
Q3=abde JI(゛..]c..-X…¨)I(゛¨k‥‾゛¨…)'
-89-
(5.15)
![Page 93: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/93.jpg)
where
Q4°abce
aΣ『
R「
dX(x...,‘‾‘x゛…..)'(x..゜1‘‾z¨…)'
bcdeふ
k
Rぶ
l
j
l
(Zりklm""^l・・"‾り'‥‾‘X;¨k‥-Z・・・..+:iぷ・….)
゜(ぶりklm‾Zi….-X.j7‾Z・・k・.‾X・・・1f3Z…・・)
X=
1-abc
abc
砿jlΣ
Q1,‥.・,Q4areexactlythesameasthoseofEq.(5.1↓),andRisexpres-
sedas
R=Qs 十Q6十‥.十Q15十R (5.16)
byQ5,Q6,‥・,Ql5andRofEq.(5.1↓).・Namely,makinguseofnewresidual
Rwhichisthesumofalltheintニeractionsandtheoriginalresidua:LR,
analyzethemaineffectsofQ1~ら.Forexample,thetestof
hypothesis‾HA:α1°‥゜゜(な゜0canbeconductedbyusingthe
factthat
り=ln一見2一手(p十見1+1)}1og拍およjkL (5.17)
isdistributedasymptoticallyaccordingtoχ^-distribution,`゛hose
degreesoffreedomisp見lundertheconditions一見i=a-l,見1十几2=
a十b十c十d-3,n-a-b-c-d+3≧p,n=abcde.(SeeAppendixF。andTable5.3)
Theresultofcalculatingforeachfactorof'C1,Vi,C2,V2ateach
timetisasthedashedcurvesinFig.5.5.Herein,
り↑゜(5.18)り
(Valueof\%significantlevelofxtestcorre-
spondingtothedegreesoffreedomofV)
Besides,themaineffectsofCi,Vi,C2,V2inthecaseof・considering
interactionsaredrawnforcomparison.(Solidcurves.)
Thisrevealsthatthemaineffectsdonotvaryoutstandinglyeven
wheninteractioniS・supposednottotakeplace.AsseeninFig.5.2。
despitethefactthattheinteractionswerenotsosmallascompared
-90-
![Page 94: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/94.jpg)
200
100
50
2 0
1 0
。nUOUOJUOIJOZIIUUIJONJ
5
2
1
..、し、.」。|し1''‾ ``ゝI' -ヽ,,二ヽ4
/~∽~-ぺ
/ノ´V,ゝゝ
/ゝ
/ゝχ
χ
L. /I
・ へ゛゛/.″
,1`\..゛`ヽ、/ぺ`、χ/'`V,″
`、`y7χ
k
ズノX///tゝX
\■/'W7fいぺ
\c/ 'X\//ヽ、へ
\/でシ/″iNへ、Vゝ
ゝ
ゝゝ
ソ//
ノレIN\ /.'`、 -
/,`\ /,ヽ へ
/!ヽX //χ へ
/・`ヽ\ //`ヽ 、へ/1゛`ゝうぐ/// ----
W
nj Xゝ
ゝゝ
ノ
1
7ノ
/べ\
u//■・■■ぐ〉心
○ノゝべ
ゝ
ゝゝゝ
ゝ
W -
J
〆ノ
●-Sが-〆〆〆
f/
/
/●,●
/
t3 t5
1
t7 t9 t11
Fig.5.5Multivariateanalysisofvarianceforfour-factordesign
withrepeatedmeasurementsinwhichallinteractions
aredisregarded(dashedcurves一一一一).Solid
curves()indicatethemaineffectsofFig.5。2・
-91
t13
![Page 95: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/95.jpg)
withthemaineffects,thesemaineffectsdidnotchanpeso
greatlyevenwhenalltheinteractionswerepooledtotheresidual.
Accordingly,itispredictedthatthedirectionsofspatialdistri-
butionofthemaineffectsaredifferentfromthatoftheintニeractions.
Toclarifytherelationbetveenthedirectionsofdistニributionsofthe
maineffectsandinteractionsbythesameprocedureasinSectionl
4.5,wetrytoexplainthedistributionsgeometrically・
Q3+R:VariancebvthemaineffectoffactorC2.
(!latrixofsumsof・squaresandcrossproducts)
Q4十R:Variancebythemaineffectoffactorv2・
(MatrixofSt!msofsquaresandcrossproducts)
Q10十R:VariancebyC2V2-interaction.
(llatrixofsumsofsquaresandcrossproducts)
R:ResidualVariance.
(Ilatrixofsumsofsquaresandcrossproducts)
Threee1:Lipsoidsfori(=3,4,10)expressedbyanequation
i[÷(Qi+R)]‾lil=i(ft(Qi+11)11‾もX=1
andtheresidualellipsoidby釧士R']~'x'=xx'=l
(5.19)
areillustratedinFig.5.6.Inthefigure,thecrosssections
producedbythethreeplanes,whichareproducedbythefirstmain
axesofthethreeellipsoidsexpressedby
i[まー(6i+11)]‾lil°1(i°3,4,10),-
aredrawn,too.
AccordingtoFig,5.6,thedirectionofthemainpartofdistri-
butionofC2V2-interactioni13nearlyatrightangleswiththatofC2-
main-effeetニandV2-inain-effect.犬
Performanceofanalysisofvarianceuponthenewresidualto
-92-
![Page 96: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/96.jpg)
whichtheinteractionarepooled(asconductedinthissection)cor-
respondstospatialshrinkoftotalvarianceQinthedirectionof
distributionoftheinteraction,buttheoperation-shrinking
doesnotinfluencetotallymuchondistニributionofthemaineffect
sincethedirectionofdistributニionoftニheinteractionisnearlyat
rightangleswiththatofthemaineffects.Hereby,agreatchange
maynotbeseeninリ゛ofFig.5.5。
c,v.
.VZ\
Fig.5.6
c.
Geometricrepresentationofdistributionsofv2-effect,
C2-effectandtheirinteractionCaV^(t=tu).
5.7Conclusion
V'echoseonlyonespeakerbecauseitwastroublesometofind
outthecharacteristicswhichwerepurelydependentonco-articula-
tionこandwhichwerelostintheincreasingfluctuation-noise-
causedfromchangingspeakerifwetreatedvoicesofvariousspeakers
atonce。
Performedmu:Ltニivariateanalysisofvariancewithrepeated
measurementsonfourfactorsCi,Vi,C2,V2,rakingtニwo-syllable
words/CiViC2V2/anobjectofanalysis.!‘7einvestigatedtheinterac-
tionvjhichindicatestheextentoftheeffectspecifictothecombi-
nationsofmorethantwoofC1,Vi,C2,V2,besidesthemaineffect
-93-
![Page 97: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/97.jpg)
whichindicatestheextentoftheeffectofCl,Vl,C2,V2themselves。
Theresultofanalysis:
(1)Themaineffectofthephonemewhichweremarkedwas
thelargestamongalleffectsatitsstationarypart:'・
(2)Theinteractionbetweentherema万rkedphonemeandthephoneme
whichwasjustcontiguoustoitwaslargerthanthemaineffectof゛
othernon-contiguousphonemesthoughitwas,ofcourse,smallerthan
maineffectsoftwocontiguousphonemesthemselves.
(3)These(1)and(2)signifythemainpartsofthetotalvariance・
Therefore,whenwetakeacertainphoneme,wemayhavetotake
intニoaccountoftheinfluences―maineffects―justprecedingand
followingphonemes,butdonothavetoconsidermaineffectsof
phonemesfartherthanthem.'Ifwe:Intendtoconsiderthelattereffects,
wehavetoconsidertheinfluencetwo-factorinteraction一一specificto
thecombinationoftheremarkedphonemeandjustprecedingone(O「
゜followingone」priortoconsideringthemaineffectsoffarther
phonemes。
Thetwo-factorinteractionisnotsosmal:L,buttoneg:Lect
itinanalysisofco-articulation.doesnotyieldasignificant
differenceintheresultsofthemaineffect,becausethedirection
ofdistributionofthetwo-facterinteractionisdifferentfrom
thatofthemaineffects.(ThisfactStニrengthenstheanalysis
model0fChapter4.)
-94-
![Page 98: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/98.jpg)
CHAPTER6
AnalysisofJapanesecvandvvuttニerances
6.1工ntroduction
Nopaperseemstoverifythehypothesisthattheinfluenceof
consonantonvowelinc-vseauenceislargebuttheinfluenceof
precedingvowelonfollowingvowelinv-vsequenceisfurtニherlarger。
Theeχperimentofthischapterwasschemedtoobtainoneanswer
tothisproblem.Inthescheme,weprovidedsyllab:Le-sequencesof
t:ypeCV2(consonant-vowel)andV1V2(vowel-vowel),investigatedthe
influenceofConv2andthatofVionV2,andcomparedonewiththe
other.Thespeakerislimitedtoonemaleadultニ。
Nowwecannotuti:Lizetheusualmodelofanalysisofvariance
inthiscase.工ntheusualmodel,forexample,ink-factormodel.
t:hetreatmentswhichinhereineachofallkfactorsaresimulta-
neouslycarriedoutsurelyineachexperiment.工nthischapter,C,
Vl,V2standforthefactors.However,evenifweschemeanalysisof
varianceforthree-factordesignontheassiffliptionthatthesetニhree
arethefactors,theydonotactニalltopethertowardspeechspectraZ・
Itisimpossibletoschemeanalysisofvarianceforthree-factor
design,because/cv./and/ViV,/areuttニeredasdifferentwords。
Meanwhile,itispossibletoschemeatwo-factorexperimentfor
CandV2,andanothertwo-factorexperimentforViandv2.Asthere
is,however,noconmunityintheresidualsofthecoupleofexperi-
mentsintニhiscase,anditisimpossibletocomparethevaluesobtain-
edbythecoupleofanalysisofvariance,wecannotcomparetheamount
oftニheinfluenceofConv2andthatofv10nv2。
y・'etriedtoeχtentthemodel0fanalysisofvariancefortwo-
factordesigntooneofthemethodswhichmustbeusefultoresolve
-95-
![Page 99: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/99.jpg)
theproblemofthischapter,standingonthismatter.Wewilltempo-
ra:Llynametニhemodelsuggestedhereanalysisofvariancefor"Divided-
Type"two-factordesign.
6.2 MultivariateAnalysisofVarianceforDivided-TypeTwo-FactorDesign
AsmentionedinSeとtion2.4.3,theusualanalysisofvariancefor
two-factordesignwithrepeatedmeasurementsisasfollows(Table6.1).
Considertwofac‘torsA,EandpresumetheirlevelstobeA1‾"a>and
Bi゛Bi^,respectively°VJhenwelet^iji゛‥"゜x;ijr
observations(thenumberofwhich・isr)obtainedfrom
combinationsof(Ai,Ej),thelinearmodelisassumed
thecondition
“゛1≦i≦a,
1≦j≦b,
や■■■
1≦k<r.ヶ
where,
arjjkClxp)=μ十鴫十βj゛''ij+゛ijk
aΣ[
(Zi=0 ら。Σ戸
^jk
a
デ0'.Σnj=
1=1
bΣ釦
rij=09
~.N(0,A):p-dlmensiona:Lnormal
distニribution.
bethe
the
tobeunder
(6。1)
(6.2)
(6.3)
Dividethefactc?rA(whose1叩91sareAI~Aa)illtoAandj2)如
thismodel(Table6.2).Namely,assumeやheleve1Sofjl)tobe
-96-
![Page 100: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/100.jpg)
Table6.1Multivariateanalysisofvariance
fortwo-factordesignwith
repftatedmeasurements.
Table6.2Multivariateanalysisofvariance
fordivided-typetwo-factordesign
withrepeatedmeasurements.
-97-
![Page 101: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/101.jpg)
(6.9)
AI‾"ai9andthatofAtobeAal.1‾Aa(1≦ai<a).
Accordingtothisdivision,weintroduceanewmodelhereinafter.
*iik=μ十μ(t)十αi(tリβj十..(t)十γり(t)+61jk
o巾:1ケ≒レ
Undertheconditiona2°a‾a19(t°1,2)
atμ(1)+32μ(2)こo'
C3Σ一「
印y"
aΣ心
bΣり
」一
~j
6Σ坤
aΣこ
i=ai+i脂
=0
bΣ
Jニ
・1べ
1
βに0'
)十a2r(F=0
シトか卜o・
りk ~N(o,A):p-dimensionalnorma:Ldistri‘bution。
(6.4)
(6.6)
(6.7)
TherelationonparametersbetweenEq.(6.1)and・Eq.(6.4)isas
(1し1″‾元tl
即ΣE
川1町二石
引Σ図
μ(2)_1a{z?}=αi‾μ(1)(1≦i≦31)'(z12)ニαi‾μ(2)(1り‾1≦i≦3)'
a2iニal+1'
r・.謬し上£r
IJ.Ja2i=al+1
ra)=γij‾「(2J
IJ
ij.rWニ'uJ
(al+1≦i≦a)
(1≦i≦aべ
(6.8)
and8stoμ'βjor゛ijk'itisthesame8SEq'(6.1).
HereinタthemeaningoftheparametニersofEq.(、6.4)isasfollows:
(1)μ(1) ee
(2)α(1)i:
(3)(Z(2)i:
(4)β丿
(5)ざ1)js
(6)r(1)ij
(7)r(2)ij
11aineffectcorrespondingtodi゛xsionAorfactorA
llaineffectofleve:LAiwhichbelongstojl).
MaineffectoflevelAiwhichbelongstoぷ).
MaineffectofBj.l
工nteractionofぶ)andBj・
Interaction‘of・B・・and:LeveヒLAiwhichbelongsto即).
InteractionofB
l
andlevelAiwhichbelongstoA.
-98-
1)
![Page 102: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/102.jpg)
.TJgwill,then,trytobreakdownthetot:alvariance.Inthe
modelofEq.(6.1),thetotニalvarianceisbrokendowntニotニhethree
metricesofsumsofsquaresandcrossproductsformaineffectsαi,βj
andinteractionT'iandoneresidualmatニrixasindicatedinEq.(2.34),
whileinthenodelofEq.(6.4),thebreakdownbecomes
abr
Q(p勺)へEI
ふ
k
E
I(゛りk‾゛√゜゜)'(゛りk‾゛“‘)
=Q1十Q2(1)十Q2(2)十Q3十〇4十Q5(1)十Q5(7仙R.
両㈲
1223J-
0c、QO、Q
=br
rrrr
bba
====
Q5(1)=「
Q5(2)゜r
i
R
where.
2
Σり引ΣJaΣ・JbΣJbΣJ
3t(x?!.-X...)へ(X!t!.-X...)
(Xi..-ぷ宍.丿(x-^..-X...),
1゛1
(X・j.‾ぷ・¨丿(X°卜-X...)
1
27?一
tl
・j
≪"ΣTi
ac-c
゜ヤ1
It(X\‘;.-z`ご‥-x・j'十ぷ‥'y
・(刈.∠χ(jt.ら;.j.←トぷ‥.)
ぷ
Z
いJfjj.X\・.一鶏.+x呪.)'
・(xり.-X;・.一刈.+jt・)
j
シ゛ij°‾゛i‥‾鶏
ニx?..y
rΣ淵
bΣJ
・j
aΣゞ=
:L
--゜゜‘abr
(1)
●●●
x. ●●
岬
りj
一
一
一一
一一
1
aΣが
'II
ai
Σ
l
b7t一b[-J
j=1
l
rマム一一r?乙
bΣコりΣ
りjk
1
rΣ一一
k
rΣ一一
k
^ijk"
りjk・
耳jk
Xi jk
(?》‥=12‘ぷ.・・a2bΓμal+
Z・j
(6.10)
bC
1j=1k=1りk
a「
=
ユΣΣzりk・
ari二1k=1
fjpH -
a2「
a「
i=ai+1k
E
I゛X7りk
-99-
aibriΞ1
1
ごーニbrに
L
1
rΣ一一
k
aμ
-
![Page 103: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/103.jpg)
Thesematricesofsumsofsauaresandcrossproducts(representedby
Qjcorrespondtothep.eaninps(1)~(7)ofeacheffectofF(!.(6.9)
asfolloV7S:
Qi●
●
●一
切%
●一
j0
9`
0.一
″
5-ぐ
(7)
(2)・sQ2(か:(3)・Q3:(4)・q4:(5)'Q5(1):(6)
(6.11)
V7hereRrepresentstheresidualmatrix.
Assuming,forinstance,thatH:μ(1)゜μ(2)'゜0(DivisionofA(1)
andA(2)iSnotSignifiCant),WeCantestRsinceitispossibleto
provethat
V={n一見2--}(p十見l+1)}1ogJjhに.^J
(6.:L2)
isasymptoticallydistributedaccordingtox2-distributionwhose
degreesoffreedomisp£i,where£1°1,ll十兌2°ab,n-ab≧p,n°abc.
VJecanalsotesthypothesesconcerningothereffectsinthe
sameformofthecriterionasEq.(6.12)byusingQ:andtheresi
R.(SeeTable6.3,andAppendixG.)
Table6.3Multivariateanalysisofvariancefordivided-typetwo-factordesignwithrepeated
measurements,anddegreesoffreedomofQy).DegrleesoffreedomofRis
n-ab(n=abr),andぷ1十ぷ2=ab.a1十a2=a.
(i)ofEq.(6.9) Factor Effectvector 卵) 瓦=訟つ(も片≒
(1)A(1)-λ2) μ(l) Q. 1
(2)I A(1) (Zi(1) Q?) ai-1
(3) A(2ド (Zi(2) 好 a2-1
(4) B βj Q3 b-1
(5) (A(l)-j2))B rj(`) Q4 b-1
(6) A(1)B 7'ij(1) 必) aib-ai-b十1
(7) A(2)B r..(2) /破) a2b-a2-b+1
-100-
![Page 104: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/104.jpg)
6.3Establishi"entofObjectsandFactors
forExDerinent
17edrewmaterialsfromthewordsofCV?(consonant-vowel)and
VlV2(vowel-vowel)syllabieutニteredbyamaleadult.
Aswechosearbitraryoneof13kindsofconsonants/p,t,k,s,h,
b,d,g,m,n,g,r,z/,andarbitraryoneof5kindsofvowels/a,i,u,e,o/
witニhwhichweprovidedasmanycombinatニionsaspossible,therewere
90kindsofwords(90=13×5+5×5).Thetotalnumberofmaterials,
providedbyutterinseachwordfivetimes,was450.
Although/ti,tu,di,du,zi,si/amongCV2-combinationsarepro-
nouncedbyphoneticsymbol[t/i,tsu,3i,zu,3i,/i]essentiallyin
Japanese,respectively,theywerepronouncedaccordingtothephonet-
icsymbols[ti,tu,di,du,zi,si]inorderthatwecanhaveobservations
inallpossiblecombinationsof:Levelsfromeachfactor.
Vedeterminedthreepointscorrespondingtoboundary,transition
partandstationarypartニofeachwordbyvisualobservation,andnamed
themt=ti,t2,t3(SeeFip.6.1.).Analysisofvariancefordivided-type
twofactordesign,mentionedinSection6.2,wasperformedforeacht,
consideringtheSDectrumdistributionsoftニ1>t2,t3asvectors.The
factorswereassignedasA(1)゛VI,A(2)゜C,B=V2.Consequently,ai°5,
a2°13,b=5,c=5.Define
x.(t)=log(
bjCt)-
沁aΣ四
ソ
), (6.13)
^^;hichisthei-tニhcomponentofp-dimensionalvectorX(t)=(XI(t).
,x(t))attimetニonthebasisoftheamDlitudeoutputs
bi(t)・‥・・b*.(t)of20-channel1/4-octavefiltersatthattime
(p=20).
-101-
![Page 105: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/105.jpg)
が
apiijiidLuv
aprnndxuv
V,
tl t2 t3
V2
tlt2t3
Fia6.1cv.andv,v.words。ddefinitionofti.
t=t,:boundaryofC-V2orVi-V2,
t=t3:stationaypartofV2・
t2=(t3十tl)/2.
6.4ResultofAnalysisofVarianceforDivided-TypeTwo-FactorDesign
Analysisofvariancewasperformedateacht(=ti,t2,t3)toward
thefactorassignmentdescribedinthesectionaboveaccordingto
procedureeχplainedinSection6.2.T・7ecomputedthecriterionリ
forthenullhypothesesconcernedwithseveneffectssignifiedinEI。
(6.9),respectively.Asthedegreesoffreedomsforeacheffectare
notthesame(seeTable6.3),thecriterionVwasnormalizedbythe
-102-
![Page 106: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/106.jpg)
correspondingvalueof1SsignificantlevelofX^
equatニion.
りl=り
test,asthefollowing
test
(6.14)
correspondingtothedegreesoffreedomv)
Thenormalizedcriterionリ゛iSillustratedinFig.6.2.Symbols,in
thefigure,correspondtoeacheffectofEq.(6.9)asfollows.
(C-V1):(1),Vi:(2),C:(3),V2:(4),(C-VI)V2:(5)
VlV2:(6),CV2:(7).
50
000
4CO(N
ヽnUOUDJUDpOZIIUlUJOM
10
jC;>
ぐV2
V,
ミ;
(C-V0V2・
"〆〆〆
(6.15)
〆
tlt2t3
Fig.6.2Multivariateanalysisofvariance
forthedivided-typemodeL
V,=/a・i・u,e,o/l
C=/p,t,k,s,h,b,d,g,m,n,o,
r.z/.
V2=/a,i,u,e,o/.
-103-
t
![Page 107: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/107.jpg)
Theillustrationreveals:'
(1)Thewayofinfluenceoftheprecedingphonemeisconsiderably
and。significantlydifferentduetowhethertheprecedingphonemeis
avowel0ra・consonant。
(2)Theeffectofprecedingvowelintheregionoffollowingvowel
islargerthanthatofprecedingconsonant,butmakesnogreat
difference.1.。.Iト
6.5Re:lationbetweenNormalizedCriterionゾand
DiscriminationScore
Accordingtotheresult(1)oftheSection6.4,thereisa
possibilitytodiscriminatewhethertheprecedingphonemeisvowelor
consonantonlybythespectruminformationofthestationarypartof
followingvowelsincetheinfluenceoftheprecedingphonemeon
followingvowelbecomesconsiderablydifferentbetweenthecases
onecase;precedingphonemeisvowel,andtheother;consonant.
-104-
Therefore,wetrieddiscriminationadoptingtニhedistancebasedon
quadraticform.(SeeSection4.4forthemeansofthediscrimination
whichisexactlythesameasthemeanshere.)
Thediscriminationstriedareasto:
(1)Vrhichistheprecedingphoneme,consonantorvowel?,
(Numberofcategorles=2)丿.`
(2)T'Jhichisthefollowingvowel,/a‘/,・/i/,/u/,/e/or/o/?,
(Numberofcategories=5)イ・
using'onlyspectruminformationxCtj)atthetimet°ti・
These(1)and(2)arecorrespondingto(1)and(4)ofEq.(6.9),
respectively・
Abovewasconductedrespectivelyateacht(゜tl,t2,t3).t°tl
indicatestheboundarybetweenprecedingandfollowingphonemes,and
t°tろthestationarypartoffollowingvowel.Fromtheresu:Ltofthe
![Page 108: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/108.jpg)
discriminationpresentedinTable6.4,thefollowingweresignified:
(1)Thetendencyonthediscriminationscoreissimilartothatof
V゛ofFig.6.2:
(2)1。7ecoulddiscriminatニewith99.8%correctnesswhetherthepreceding
phonemeisvowel0rconsonantonlybythespectruminformation
inthestationatypartoffo:Llowingvowel.
Table6.4Discriminationscores(%)。
FactorNo.of
c°!tegonesh t2 t3
V,―C 2 99.1196.9
|
99.8
V2 5 96.2 99.8 100
6.6GeometricRepresentationofAnalysisofVariance
forDivided-TypeModel
\'etriedaspatialexpressionofvarianceellipsoidindivided-
typeanalysisofvariance.Thisdiffersalittlefromthespatial
expressionoftheusualanalysisofvarianceIntroducedinSection
4・5。
ThevarianceoffactorAf1)isregardedasQ2(1)十R,andthatofjj
as02(2)十R,buttheseareconsideredinthespacenormalizedby
residualR.FromthediscussionofAppendixD,thevarianceofA
andA(2)isasfollows(n=abr):
1
-n
・1
-n
2
(Cぐ
(O'
ぐ
(1)十ii)=ぐ士(Q2(1)十R)R‘士,
2(2) +R)=r"士(Q2(2)-トR)R°士
Sincethecentersofthevarianceellipsoids,which
-105-
(6.16)
(6.17)
![Page 109: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/109.jpg)
and士(Q24)十R)signify,areontheendsofthevectorsofmaximum
like:Lihoodestimates
;;(1)=x(1!‥-z...3
;誤)=z(2!‥-z….
(6.18)
(6.19)
obtainedfromvectorsμ(1)μ(2)ofEq.(6.A),respectively(SeeAppendix
1ぶ(l),nG),thecentersofthevarianceellipsoidssignifiedby一手(Qj
and÷(02R十貫)arerespectivelyontheendsof
j(1)=(z(IL.-z‥.)(R/n)‘士,
ぴ)=(x(21‥-z‥.)(R/n)'士.
(6.20)
(6.21)
FromEquations(6.20)and(6.21),theconcentrationellipsoids
offactorA(1)andA(2)are,reSpecrively,
(i-i(1))[十(a2(1)+ii)]-1(i-i(1))・=p+2
(i一夕)[十(62°)+ii)]-1(
wherep=20.(SeeSection2.2.3.)
i-;y))l=p+2
(6
(6.
22
23
Assume!:heeigenvectorscorrespondingtothemaximumeigenvalue
ofthematrices・ofEq.(6.16)andEq.(6.17)tobe61,62>respectively・
Also,as
~(1)+aj(り=o (6.24)
;;(1)and;i(2)areononestraightline.Presumethatthevectorwhich
representsthislineisa。
レTheillustrationoftheintersectionsproducedbythetwoplanes
-106-
![Page 110: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/110.jpg)
(determinedbya,ろ1andbva,62)andtheconcentrationellipsoids
ofEq.(6.22)andEa.(6.23)(inthissection,A(1)=VI,A(2)=C,B=V2)iS
asI^ip.6.3(onlyt=t3).
c:上(司2)
:十(郁1)
4) ノ恥
Fig.6.3Geometricrepresentationof
distributionsforthedivided
-typemodel(t=t3).
V,=/a,i,u,e,o/,
C=/p,t,k,s,h,b,d,g,m.
n,i),r,z/.
α
4)
・S.
ItisrecognizedthattheinfluencesofprecedingconsonantC
andprecedingvowelViwereseparatedatthestationarypartof
followingvowe:LV2.
-107-
![Page 111: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/111.jpg)
6.7Discussion
Toinspecttheresultsoftheanalysisuptothesectionabove,
wechangedthedivisionoffactorA,andcompareditwiththedivision
inthesectionabove.
LettingPrepresentニthegroupofvoicelessconsonantニS/p,t,k,s,h/
(31°5),andTvoicedsounds/b.d.g.m.n.o.r.z.a.i.u.e.o/(a2=13),
theanalysiswasperformedundertheconditニionj1)=P,か)=T,B=V2,
providedthatAI)and瓦2)arediviSionSoffactorA.Theresultsof。9
performingsimi:LaranalysisonSection6,4,6.5and6.6areasFig・
6.4,Table6.5andFig.6.5,respectively・
Itisunderstoodbyconparingtニheseresultswitheachother
thatthedivisionofC-Viissignificant.
Table6.5Discriminationscores(%).
FactorNo.ofcat^ones
tl t2 t3
P-T 2 97.1 87.1 79.6
V2 5 96.2 99.8 100
-108-
![Page 112: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/112.jpg)
ゝ
こゝ
UOUDJUODDZITUUIJOI^
50
40
30
20
10
PTぐ
-
2
ヽ、、_ど認否1.
TV2PV,
tlt2t3
Fig.6.4Multivariateanalysisofvariance
forthedivided・typemodel.
P=/p,t.k.s,h/.
T=/b,d,g,m,n,n,r,z,a,i,u,
e,o/,
Vj=/a.i,u,e,o/・
-109-
![Page 113: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/113.jpg)
T:古(辺2)fi)
ノ
如し
a
Fig.6.5Geometricrepresentationofdistributionsforthe
divided・typemodel(t=t3)・
P=/p,t,k,s.h/,
T=/b,d,g,m,n,g,r,z,a,i,u,e,o/.
-110-
&2
![Page 114: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/114.jpg)
6.8Conclusion
Theresultsofanalyzing450words(whosetypesarecvandvv)
uttニeredbyamaleadultareasfo:Llows:
(1)Thewayofinfluenceoftheprecedingphoner!eonthefollow-
inpvowelisconsiderablyandsignificantlydifferentduetowhether
theprecedingDhonemeisvowel0rconsonant.
(2)Themaineffectofprecedingvowelintheregionoffollowing
vowel‘islargerthanthatofprecedingconsonant,butmakesnogreat
difference.
(3)Vecoulddiscriminatewith99.8%correctnesswhetherthepreceding
phonemeisvowel0rconsonantonlybyspectruminformationinthe
stationarypartoffollowingvowel。
Theusualmodelforanalysisofvariancecouldnotbe
utilizedintheana:Lysisofthischapter.Instead,weintroduced
newlya'Divided-type':modelforanalysisofvariancefortwo-factor
design.\-!etニriedtoexpressspatiallytheresultofanalysisof
thismodelinaparallelwithit,andconfirmedtheresultabove.
-Ill-
![Page 115: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/115.jpg)
り2項欠
![Page 116: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/116.jpg)
CFAPTER7
Three-DimensionalRepresentationofJapanesePhonemes
7.1工ntroduction
!・7ehavesofartriedtoexpressspeechsoundonacoordinate
planeorinthree-dimensionalspaceontニhebasisofparametersobtain-
edbyanalyzingspeechsound,andtoobservetニheirmutualrelations.
F1-F2formantplane,inwhichthefirstandthesecondformant
frequenciesareregardedasparameters,isamostpopularexampleof
them.Klein,PlompandP01S(:L1)havetriedtoplot12kindsofvowels
of50speakersinspacebyspectralinformation.Consideringspectニrum
representedbytheamDlitudeoutputsof18-channel:L/3-octavefilters
as18-diinensionalvector,tニheyperformedtheprincipal-component
analysis.They'theninsistニedthatifpresentvowelsontheplane
determinedbythefirstニtニwoprincipalcomponents,itsconfiguration
issimilartotheconfigurationinF1-F2forinantplane。
Thusbyrepresentingphonemesoundbyrelationofspatialdis-
positニion,itispossibletogivesupport,socalledinphonetics.
toac:Lassifiedtableofphonemesandtoacardinalvowelfigurefrom
aphysicalpointofviewiftheresultscoincide.Moreover,there
wouldbemuchmoreprofitableaspectsincetニhemeasureofdistance
wasimported。
I^enitcomestoconsonants,however,expressingexamplesin
thiswaywouldberare.Themainreasonistニhatitisnoteasierto
analyzeconsonantsthanvowels,fortherearesomedifficultieson
catニchingspectraofconsonantsattheirinstantofexplosioninsuch
caseasplosiveconsonants。
Thisstudytookninephonemes/p,t,k,b,d,g,m,n,andD/asa
preliminarytothistrial0nspatialexpressionofphonemesincluding
-113-
![Page 117: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/117.jpg)
consonants.Veexpressedthesephonemesinthethree-dimensional
spaceafterdefiningtheirspectra.Itisspontaneouslyindicated
thatthisexpressionhasaprofoundstatisticalmeaning。
Furthermore,thischapterusedthesamedataasChapter5and
partiallytheresultsofChapter5becausebothexperimentswere
designedatthesametime.
7。2cvcvUtterancesandtheirSpectra.
Itisactuallyrarethateachphonemeisindividuallyuttered.
Par!二icularlyinJapanese,itisconnnonforconsonantstobeuttered
withvowelslike/ka/.liepayattention,here,toC2oftwo-syllable
CiViC2V2(consonant-vowel-consonant-vowel)inordertoconsider
consonantswhichareplacedinmuchmorecomplicatedcircumstニances
thaninamonosyllableCV。
Using/p,t,k,b,d,g,m,n,0/forCi,C2and/a.e.o/forVl,V2,
anadultmanspokeeachwordofallthecombinations―729kinds
threetimesatrandom.HeaccentedalltheCiVi-moradispitニe
-114-
thefactthattherearea10tofmeaninglesswordsinthecombinations
above.Thesoundswereutteredinthesimlifiednonreverberantroomp
withinonedayandthetotalnumberofwordswas2187.Thereason
forchoosing/a.e.o/asvowelswasthatwecanhaveobservationsin
a1:Lpossiblecombinationsoflevelsfromeachfactorbyeliminating
/iandu/since/tl,tu,di,anddu/donotoccurinmodernJapanese・
Furthermore,thesespeechdataarethesameasthoseinChapter5.
Wewillgiveadefinitionofspeechspectraandprocedureof
segmentation,below(detaileddescriptionisinChapter3).
Whenletbi(t),.....,b(t)(p°20)represent.inordeらamplitude
outputsof20-channel1/4-octavefi:Lters(bankof20filterswhose
centerfrequenciescover210upt05660Ez)attimet,theyare
consideredtorepresentspeechspectraatthattime.Afternormal-
![Page 118: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/118.jpg)
izingtニhesquaresumofthesecomponentsat1,weestニablishedp-
dimensionalvectorx(t)bytaking:Log;arithir.ofitscomponents・
Thatis,
Xi(t)゜:Log{bi(t)//1♂(t)}J=lj
(7.1)
!'7edefinedp-dir゛ensio°1゛ectoi°x(t)゜(xi(t)・‥・・Xp(t))゛ithEq・
(7.1)whichwouldbeusedfortheanalysis(p=20).Amplitudeoutputs
ofthefilteranalyzerareA-Dconvertedatintervalsof10ms,tニhen
putintothecomputerinrealtime.0nthepaperfortheline-printer
werepresenttニhespeechspectrumpatternsoftheinputwithwhichwe
observedtodeterminethestationarypartsortransitionones.
ThisStニudydefinestimest1,ぢ2≫・・・≫tl3asFig.フ.1,correspond-
inptothestationaryortransitionpartsineachword.
oprmidiuY
tl t2 t3t4 t5 t6t7t8t9trotilt12t13
Fig.7.1c,v,c,v.wordanddefinitionofti-
t=t,:stationarypartofc..
t=t3:boundaryofc.-V.,
t=t5:stationarypartofV.,
t=t7:boundaryofV,-C2,
t=t9:.stationarypartofC2,
t=tn:boundaryofC2-V2,
t=t13:stationarypartofv2・
Ifiiseven.ti=(ti+l十ti-i)/2.
-115-
t
![Page 119: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/119.jpg)
7.3ConsonantSpectニra
Itisaimportantproblemwhetherconsonantswithindifferent
wordsshowcharacteristicsdeterminedindependentlyoftheirphonemic
environmentornot.工fthecharacteristicsofconsonantsdifferin
accordancewiththeircircumstances,wecannotphysicallymeasure
thembutonlybyauditorypsychologicalmeans.Analvsisofvariance
・●|●●ilforfour-factニordesignwithrepeatedmeasurementsonCiViC2V2Word,
inChapter5,answersthisquestion.LookingatFig.5.2inSection
5.4,itispossibletopersuadeourse:LvesthattheeffectofC2is
sufficientlysignificantascomparedwiththeresidualatt=to,t,,,
etC・,andthatthereexistdefinitephysicalcharactニeristicswhich
areindependentofthespeechconditions.
Todealwiththeargumentabove,wewillpresentagaintニhe
expressionofthelinearmodelEq.(5.1)usedinChapter5.namely,
wepresumealinearmodelrepresentedbythefollowingexpression
forspectrumvectorattimet=ti.
U
j(=1~b)thelevel0fthesecondfactorVi,k('=:L~c)theleve:Lof
thethirdfactorC2,1(=1~d)thelevel0fthefourthfactorv2,
respectively,andm(゜1~e)representstherepetition(a=d=9,b=c=e
=3).Omittingt,
xijklm(1xp)゜a;
4‘“けβけ7i,゛∂l
generallevel
maineffect
キsij+ぐik+ηil+∂jk+λj1ヰμ■kltwo-factorinteraction
+ジijk+pijl+・ikl+'■jkl
+91jk1
+xijklm
-116-
three-factorinteraction
four-factorinteraction
residual
(7.2)
![Page 120: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/120.jpg)
Determine
Σ
i二1
Σ
i=1
(Zi
^ij
--
a
Σ
J二
b
Σ
j=1
一一
J
a
Σ一一I
I
b
==Σ
j二l
b
£
β
j
=Σε
i=1
so astosatisfytheconditionofthefollowingequations.
dl;.=Σ・J.=0.
.a
り゜ぶ1ぐilf―2ぐikづミョ1ηa=l1'i>
λj11が1λj1ヘギラA'd
‾ぷ/ki‾0
ソijk°
b
2/jki
Pijkl
bC゛ぶjk°ぶ
1
∂jk
abda
μijk°ぶ/ijl°ぷ/ii.1ニFIpり1二Σミjikl
Cd°Jjjk11?F7jkl゛0
bC(1
°j弓ざりkにぶ?りklニぶ/ijkl°0
dΣ
'^ikl
(7.3)
Zりklm~N(0,A):p-dimensionalnormaldistニribution.(7.4)
AsweseeinFig.5.20ftheanalyzedresu:LtinSection5.4,
aspectrumofanysectioninautterancedoesnotrepresentone
phonemeonly.Influenceofotherarticulationally-comblnedphonemes
alwaysintervenes.Therefore,it1Sinsufficienttoconcludethat
spectraぶりklm(t)measuredwithintheregionsofconsonantsrepresent
spectraofconsonants.
Att°tilwhenspectraofconsonants/P,tニ,k,b,d,g,m,n,o/are
definedasmentionedlater,tニheeffectofthevowelV2issoover-
poweringthattニhecharacteristicsofconsonantmaybelostinthe
vowelifxりklmitselfisusedasspectrumofconsonant.
Wewilladoptmaineffectvectニorr.(k=i~c)ofC2inthelinear
model0fEa.(フ.2)asspectrumofconsonant.γkisavectorwhich1S
determinedonlybythek-thconsonantofC2,and1Savalueindepend-
entofitsphonemiccircumstances.工notherwords,7krepresentsa
spectニrumwhichisregardedasapurecomponentofconsonantC2after
removinganycomponentconcernedwithotherphonemesexceptC2from
Zりkim゛Althoughtheproblemiswhetherobtainedrl`4mustbe
sufficientlyandsignificantlydifferentfromeachotheraswellas
tニheyareseeminglydifferent,Chapter5clarifiesthispoint.
Wewillnowconsideratwhattimepointtwehadbetteradopt
へ(t)aSaconsonantvector.Fig.5.2saysthattheeffectofC2
-117-
![Page 121: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/121.jpg)
ismaximumatニ∧themiddleoftheprecedingvowelandthefollowingone,
thatist°t9.Itseemsreallymaximumonthewholeofthenine
consonants,butwhatwecanonlyclassifyintニhistimepointare
mannerofarticulation(!・!1,112,M3ofTable7.1)andnasalsounds(
/m,n,o/):wecanseldomdovoicelessstops(/p,t,k/),(It‘willbe
clarifiedinthefollowingsection.)
Accordinglywewanttoselecta.timepointatwhichthesenine
phonemeshaveequallybalanceddistributionevenifthemaineffect
becomessmall.WepresumedmaineffectvectorsofC2-factoratthe
boundaryofC2andv2(t=t11)tobespectraoftherespectiveconso-
nants.
Actually,maximumlikelihoodestimateof7k
7k°`・k●.-j;‥‥
isused,(SeeEq.(5.5))
(7.5)
7.4ClassificationfromtheViewpointofllannerand
PlaceofArticulation
Consonants/p,t,k,b,d,g,m,n,D/areclassifiedfromtheviewpoint
ofmannerofarticu:Lationorplaceofarticu:Lationasdescribedin
Table7.1.(17)Weeχaminedwhetherthiskindofclassificationwas
alsoref:Lectedinthespectrumdistribution.C2isinterpretedin
twowaysmannerofarticulation(M-factor)andplaceofarticu-
lation(Pこfactor) asTable7.1.
-118-
AssumeMI~M3levelsforll-factorandPI~P3levelsforP-
factor,respectively.TheseM-factorandP-factorareassignedto
thefirstfactorandthesecondone,respectively,oftheanalysisof
varianceoffour-factordesignofEq.(7.2)orEq.(5.1).AndViandv2are
assignedtothethirdfactorandthefouth,respective:Ly,asbefore.
![Page 122: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/122.jpg)
Table7.1Classificationfromtheviewpoint
ofmannerandplaceofarticulation.
p.Labial
P2 Alveolar
P3 Palatal/Velar
Ml Voicelessstop
M2 Voicedstop
M3 Nasal
liedidananalysisofvarianceontrio-sequence~V1C2V2(disregarding
CifromC1V1C2V2sequence)whichhasrepetitionscorrespondingtothe
numberofconsonantsCi(nineinthistime),regardingM,P,Vi,V2as
fourfactors。
yeconsider,foreχample,utterancesof/pame,tame,kame,bame.
dame,game,mame,name,Oame/tobeutterances/ame/repeatednine
times.工nthiscaseweusedthefirstutteranceoutoftニhree-repeated
utterancesofeachcombinationinCiViCaV,words(2187/3=729data)。
Normalizedcriterionリ'.inthesamewayasinSection5.4,is
showninFig.7.2.TheeffectニSofmannerandplaceofarticulation
aresufficient:1-yseparated.EspeciallyamountニSofthebotheffects
atpartv2arealmostthesame,andinteractionsbetweenbothare
smal:Latthatpart(lioreindetail,interactionsatv2arelargerin
placeofarticu:lationthanmannerofit)。
ToverifytheadequatenessoftheclassificationinTable7.1,
performanalysisofvarianceforfour-factordesignofχ,Y,Vi,V2based
uponclassificationmixingelementsofmannerandplaceofarticulation
intentionally.TheresultsareshowninFig.7.3。
Thus,wecanseethattheclassificationaccordingtoTable7.2
1Snotgood,becauseinteractionofx,Yislargefarbeyondthemain
effectsofthem.
-119-
![Page 123: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/123.jpg)
為
70
50
3 0
uouajuDpgzijBUUoj^
10
t5t7t9til
Fig.7.2Multivariateanalysisofvarianceforfour-
fectordesignwithrepeatEdmea万s万urements.
VI,V2=/a,e,0/,
M=(MI,M2,M3):manner.
'-P=(PI,P2,P3):place.
-120-
t13
t
![Page 124: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/124.jpg)
ゝ
以
70
50
3 0
U0U3JUDpOZIJBUUOJsJ
10
t5t7t9tilt13
Fi&7.3Multivariateanalysisofvarianceforfour-
factordesignwithrepeatedmeasurements.
VI.V2=/a,e,o/,
χ・=(χhχ2,χ3),
Y=(Y,.Y,,Y3).
-121-
t
![Page 125: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/125.jpg)
Furthermore,whythemain
Table7.2Somesortofclassification.effectofViisSma:Llerthanv2is
thatweassumedC1V1C2V2wordtobe
-ViCoV,.工ndespiteofthefact
thattheinfluenceofC1=/p,t,k,b.
d,g,in,n,0/onViisconsiderably
large,wedisregardedit.This
increasednoiseatpartVi,which
accordinglymadetheratiotothe
termofresidualworse。
Itwouldberelativelyeasyto
classifyinmannerofarticu:Lation
sincetheeffectofitwasenoughlargeattimetg>as"clearinFig・
7.2.So,weexaminedtheeffectsofconsonantニSbe:Longingtoeach
mannerofarticulation(Mi,M2,M3),respectively.Schemeanalysisof
varianceforthree-factordesignwithrepetitioncorrespondingtothe
numberofCi(nine),assigningthetニhreefactorstoVi,C2,V2intrio-
sequence~V1C2V2disregardingCifromC1V1C2V2.Modelsforthree-fac-
tordesigncanbedefinedinthesamemeansasEq.(5.1)~Eq.(5.3)。
Theresultsofperforminganalysisofvarianceforthree-factニor
designrespectivelyonthreecasesofC2-C2°/p,tニ,k/,C2°/b.d,g/.
andC2=/m,n,0/'-arepresentedinFig。7。4。
ItisunderstandablefromFig.7.4thatwecandistinguishnasals
/m.n.O/att=t9,butabsolutelynotvoicelessstops.Meanwhile,all
thethreecasesarealmostequallydistinguishableatt=tll。
Aboveisexplainedwhyweadoptedγkatt=tllasconsonant
spectra.
7.5ExpressioninThree-DimensionalSpace
Accordingtotheprecedingsectニion,wewillexpressthenine
-122
![Page 126: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/126.jpg)
。nU0IJ31U3|)OZ!|BIIUO;\I
ヽ4U()U.>}U3pOZiri-'tUJOM
。/\UOUOimpOZIIBlUJOM
ts
(1)C2
t5
(2)
t5
=/
t7
t9 tu t13
t
p.t.k/;Voicelessstop.
t
9t11 t13
t
C2=/b,d,g/;Voicedstop.
t7t9 t11 t13
t
(3)Cz=/m,n,o/;NasaL
Fig.7.4Multivariateanalysisofvarianceforthree-factordesignwithrepeated
measurements.VI,V2=/a.e,o/.
-123-
![Page 127: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/127.jpg)
consonants/p,t,k,b,d,g,m,n,o/withvectors
AA∧A∧AAA∧A∧7'1'`'り=rp゛rt゛rk'rb゛rdJg'rmJ7n・h
(maximumlikelihoodestimatesby2189wordsatt=tll).
Nowwewouldliketoprovidealessthanthree-dimensionalspace
inwhichwecanwellunderstandthebehaviorofmannerandp:Laceof
articu:Lation.工fwecanhavethespacelikethis,wehaveonlyto
projectorthogonal:1-yrespectivevectorsinthe20-dimensionalspace
onitssubspacewhosedimensionislessthanthree.
Let≪ini≪m2representeigenvectors(puttingtニheminorderof
largenessofthecorrespondingeigenvalues)ofcovariancematrix:
Sm゛
十('■ptk-'■ptk十''bdg'''bdg十rl°9゛rIIII11)
whichconsistsofthreevectorsof(7.7),thatis.
'■ptk°zp十r§十rk
タ
''bdg,゜jjゼ:‘?Pj!≒'^mnn°jLΞビjrニjL!・
(Note:''ptk十リbdg十mno°O)
(7.6)
(7.7)
Theextentofseparationconcernedwithmannerofarticulationmust
berelativelyfavorableinthesedirections.
Likewise,supposethat°P1,°P2areeigenvectorsofcovariance
matrixSpwhichconsistsofthreevectorsof・
^pbm°り)十rM十rm
9''tdn°3十7'nタ''kgO°
rk十
3/
(Note:りbln十''tdn十''kgo°0), (7.8)
thentheextentofseparationconcernedwithplaceofarticulation
mustbealsofavorable.
Projectorthogonal:Lythe20-dimensionalvectors''p~'"oon
itsthree-dimensinalsubspacewhichisgeneratedbythreevectors
≪mi,≪piand・≪P2.Anarbitraryvector(throughtheorigin)that
-124-
![Page 128: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/128.jpg)
belongstotニhesubspaceWIヽgeneratedby{amb°P1.≪p,}.is
representedas
xlaml十x2apl+x3ap2=iA(7.9)
provided,A'(px3)=(Om,≒apl≒ap2゛),X=(xi.χ2,X3).
LetF,)AtobetheorthogonalprojectionofrponW1.Sincerp'-ぢAisat
rightanglestoaarbitraryvectorjcAbelongingtoU1,
iis
r
p
O°(rp‾ デPA・iA)゜(7pA'-rpAA'・i)
arbitrary:then
A'°rp AA',thatis.〉r~
r
p
A'(AA゛j'1
Table7.3r=(7'1,rj.T3)which
istheorthogonalprojection
ofr(1xp)onWi.(p=20)
V「 γ1 r2 r3
p 4.53 -4.20 -3.2?
t 483 3.48 -0.85
k 3.70 -2.75 1.07
b 1.16 -1.11レ1.92
d 1.00 6.75 1.68
g 0.78 -0.83 4.48
m -5.69 -1.76 -1.55
n -5.83 3.95 -0.81
D -4.48 -3.53 1.17
(7.10)
(7.11)
WemadeTable7.3byobtain-
ing'■p-'"o
-125-
orthogonal
projectionsofrp゛roonWi
throughaboveprocess.The
resultofexpressing'■p-'-o
intニhree-dimensionalspace,whose
coordinateaxesare"mi.≪P1
and°P2becomesasFig.7.5.
A11thevectorsinFig.7.5are
movedparallelalongtheaxis
°mi,inordertomakeiteasier
tounderstand.Sothatactual
originisrepresentedbyG.
Thisrevealsthatニnine
phonemescomposenearlyatriangularprism,clarifyingrelation
betweenmannerandplaceofartieulation.
![Page 129: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/129.jpg)
/k/
/g/
llI
―・―
/p/
\
\
/D/
am,
OP2
3.65
11-
\
/1
/
/
1
卵
3。85
/
/
/
/
--
--
75°
-皿-〃"
ぺ
(y
/""
へ
S
90°
//
\
\
\
Residual
/n‘`
ゝゝゝ
ゝχ\
/
/
/
X
ゝ
ゝ
/
/
//
/
/
/
Fig.7.5Three-dimensionalrepresentationofJapanese
consonantsal!4!30%-probabaityellipsoidof
-126-
/d/
3.80
/n/
aPi
![Page 130: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/130.jpg)
64
00
0.2
0
0.2
64
00
0.2
0
-0.2
-0.4
1
20channel
250
5
←
500 1K 2K4KHz
Fie.7.6Directioncosinesofam*,am2,Op,andOp,
-127-
![Page 131: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/131.jpg)
工nFig.7.5wealsoshowthe90%-probabilityellipsoid
(whereAp
れi1/n)‘1N-I
Z= X2p (a) (7.12)
(a)isthenumbersuchthatEq。(2.7)ho:Ldswhenp=3,
(加0.1)correspondingtotheresidual
R(3×3)=(AI(λA・)‾1)・R(A・(AA写1) (7.13)
byiijklm(1)(3)゛hichisorthogonalproiectionof20-dir!ensional
vector■^iiklmontheabovethree-dimensionalsubspace,whereR(20×20)
istheresidualvarianceasinEq.(5.4)andn=abcde.
(SeeAppendixHandSection2.2.2.)
Wecanseeverticesofthisprismseparatedsignificantlyfrom
eachotherascomparedwiththesizeoftheresidualellipsoidR/n.
°P1,
Vesignifieddirectioncosinesofeigenvectors "mi,"m,.
°P2-inFig.7.6.Thepatternsarenevercontrarytothe
traditionalspectralknowledgeonmannerandplaceofarticulation.
7.6Conclusion
Weanalyzednineconsonants,andinvestigatedrelationsoftheir
relativedistributニioninthree-dニimensionalspace.Theseconsonants
wereplacedatphonemeC20fCiViC2V2Words(a:L:Lofthetotalcombi-
nationsof.Ci,C2=/p,t,k,b,d,g,ra,n,o/,andVl,V2=/a,e,o/).To
beginwithweextracted,fromspeechspectra(20-dimensionalspectra),
componentswhichrepresentindependentlyofco-articulation,then
selectedtheboundaryofC2-V2asasectionatwhichtheseconsonants
areevenlyseparated.VJhenprojectedorthogonallythenineconsonants
onthree-dl皿ensionalsubspacewhichisgeneratedbythreedirections
andpromotesseparationofconsonantsin・viewpointofmannerandplace
-128-
![Page 132: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/132.jpg)
ofarticulation,theytooktheshapenearlyofatriangularprismin
thespace.Itisguaranteedbymultivariatestatisticalanalysisthat
theseverticesaresignificantlydifferentfromeachother。
Spatialexpressionofconsonantsfromphysicalandanalytical
point二sofviewbasedonactualspeechseems,forthefirsttime,to
betriedinthisstudy,thoughspatialexpressionofvowelshasbeen
investigatedfromtニheearlystage.Spatialexpressionofother
phonemesisabouttobeinvestigated.
-129-
![Page 133: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/133.jpg)
ト即項欠
![Page 134: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/134.jpg)
CHAPTER8
Conclusion
Weconsideredthefrequencycomponentsofspeechspectrumas
componentsofmulti-dimensionalvector;performedmultivariate
statisticalanalysis;andunderstoodvariouscharacteristicsof
speechsoundsmainlyco-articulationsanddifferencesofspeakers
quantitatively・
-131-
Startニingwiththeconclusion・,wewilldescribethecharacteris-
ticsofthemethodsofanalysisintroducedinthisstudy,andnew
proposalsonthemethodsofanalysis.
(1)Asithasbecomepossibletoanalyzedirectlyspeechspectraby
introducingmu:Ltivariatestatisticalanalysis,analysisofconsonants
hasalsobecomeaseasyasthatofvowels.Onlyvowelscanbeanalyz-
edbyusingthetraditionalformantfrequencies.Themethodadopted
inthisstudyisalsosuperiorInthefactthatspectrumisfavorable
forasimp:Lerexpressionofphysicalinformationofspeechsoundto
themethodusingformantfrequencies.
(2)Particularlymultivariateanalysisofvarianceamongmultivariate
statisticalanalyseshastendedtobeundervaluedaboutpractical
applicationsindespiteofthemathematicalresultsofitstheory.
InthisStニudy,however,thismethodofanalysiswasextremelyuseful・
工tmeansthatthepracticalvalueofthemethodofmultivariate
analysisofvariancewas,forthefirsttime,confirmedhere.
(3)Twomatters,intheoryaswellasinapplication,werefoundto
contニributetomultivariateanalysisofvariance;
(i)Notionofdirection,besidesnotionofamountindicated
generallybytheanalysisofvariance,wasnecessarytounderstand
theeffectsofrespectivefactorsofanalysisofvariance,andone
![Page 135: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/135.jpg)
ofitstechniquesofexpressionwasdescribed。
(ii)Amodelofthe'Divided-type'"analysisofvariancewas
proposed.Thisisanup-to-datemodelinwhichananalysisof
variancefortwo-factordesignwasdevelopedbydividingoneoftwo
factorsintomorethantwoportニions.Thismodelwasintroducedto
doresearchonpatternsofsometypesofco-articulations,andformu-
lated。Itwasfurthertriedtoexpressspatial:1-ytheresultofthe
analysis.
(4)Wefoundsometypesof]Limitinthemethodoftheprincipal-
componentanalysis.
Wewilldescribetheresultsonspeechscienceacquiredinthis
study.Themostfavorablepointthroughouttニheseresu:Ltニsisthat
theywereinvestigatedquantitativelyonsufficientlystatistical
grounds.。
TheresultsofanalyzingVCV-typewordsutteredbyfivemale
adultsareasf0110WS:
(1)Atthestationarypartofnasalconsonants,theeffectof
speaker-factoristhelargestamongfourfactニorS:
(2)Atニtニhestationarypartofthevowel,theeffectofvowel-factor
isthelargestamongfourfactors.Theinfluenceofthenasalし
consonantニcontiguoustothevowelissmallerthanthatofspeaker:
(3)Theeffectofspeakeris,generally,considerab:Lyinfluentニial,
v/hilethatofconsonantisnotニsoInfluentia:L.Putthedirections
ofthemainaxes,seeninthedistributionsofthethreefactニorsof
vowel-effect,consonant-effectandspeaker-effectニ,meetnear:1-yat
rightangleswitheachother:く
(4)Intensivecorrelationisfound・betweenvowelandspeaker-factor:
(5)Formantinclinesto"vocalicfactor,"andhasnotsufficient
informationonanyfactorotherthanvowel-factorascomparedwiththe
caseofspectrumやI・・,●・●。。●
-132-
![Page 136: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/136.jpg)
Theresults,obtainedfromanalysisofCVCV-syllablewordson
co-articulation,arethen;
(6)Themaineffectofthephonemewhichweremarked1S,
0fcourse,thelargestamongalleffectsatitニSstationarypart。
Theinteractionbetweentheremarkedphonemeandthephoneme
whichwasjustcontiguoustoitwaslargerthanthemaineffectof
othernon-contiguousphonemesthoughitwas,ofcourse,smallerthan
maineffectsoftwocontiguousphonemesthemselves:
(7)Theabove(6)signifiesthemainpartsofthetotalvariance・
Therefore,whenwetakeacertainphoneme,wemayhave
totakeintoaccountoftニheinfluencesmaineffectsofjust
precedingandf01:Lowingphonemes,butnothavetoconsidermaineffects
ofphonemesfartherthanthem.IfweIntendtoconsiderl・thelattereffects,
wehavetoconsidertheinfluencetwo-factorinteraction一specificto
thecombinationoftheremarkedphonemeandjustprecedingone(o「
followingone」priortoconsideringthemaineffectsoffarther
phonemes.
(8)l・leanwhile,thetwo-factorintニeractionisnotsosmall,but,
toneglectitinanalysisofco-articulationdoesnotyielda
significantdifferenceintheresultsofthemaineffect,because
thedirectionofdlstributニionofthetwo-factorinteractニion1S
differentfromthatofthemaineffect.
FromtheresultofanalyzingcvandVV-syllablewords;
(9)Thewayofinfluenceofprecedingphonemeonfollowingvowelis
considerab:lyandsignificantlydifferentduetowhetherpreceding
phonemeisavoweloraconsonant.
(10)Themaineffectofprecedingvowelintheregionoffollowing
vowelislargerthanthatofprecedingconsonant,butmakesnogreat
difference.
Astotrial0fspatialrepresentationofconsonants:
(11)Thespectra,whichrepresentnineconsonants/p,t,k,b,d,g,m,n,o
-133-
![Page 137: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/137.jpg)
/jweredefinedattheboundarybetweenfollowingvowel.Nine
consonantsof20-dimensionalvectors,whichareprojectedorthogonal-
1yonthree-dimensiona1subspace(thatiscreatedbythethreedirec-
tionswhichpromoteseparationinmannerandplaceofarticulation),
producenearlyatriangularprism.Althoughspatialrepresentation
ofvowelshasearlybeentried,physicalandanalyticexpressions
ofconsonantsbasedonactualspeecharesupposedtobe,forthefirst
time,triedhere.へ
Abovearetheprincipalresultsobtainedinthisstudy.
Problems
speakers
-134-
(1)Evaluationoftheinteractionsbetweenspeechsoundsand
(19).
(2)Spatialrepresentationofa11Japanesephonemes:
,stニi11lefthereafter,areinvestigatedbythepresentwriter.
Wehopeinconclusionthattheresultsofthisstニudywillbe
helpfultothedevelopingstudyof"speechscience"asfundamental
data。
Wealsohopethatthemethodsofmultivariatestatistical
ana:Lysisconfirmedanddevelopedinthisstudy,willbefully
utilizedInmanyotherresearchfieldsuponthisopportunity・
![Page 138: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/138.jpg)
ACKNOI-TLEDGMENTS
IwishtoexpressmyheartygratitudetoProfessorToshiyuki
Sakaiforhisguidance,constantニsupportandencouragementduringthe
courseofthisstudy。
lalsowishtothankAssistantProfessorShujiDoshita(Tokyo
InstituteofTechonology),AssistantProfessorYasuhisaNiimi(Kyoto
Univ.of工ndustrialArtsandTextileFibers)forgivingmemuchneeded
andinformativehelp。
IamindebtedtoAssistantProfessorMakotoNagao,Assistant
ProfessorHidenosukeNishio,AssistantProfessorShigeharuSugita,
Dr.KeniiOtani,Mr.TakeoKanade,Ilr.SiniiTomita,Mr.Shigeyoshi
Kitazawa,Mr.IlasatoshiKuboandmembersofProfessorSakai'sResearch
Groupforhelpfuldiscussion,co-operationandvariousconveniences.
-135-
![Page 139: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/139.jpg)
1罫項欠
![Page 140: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/140.jpg)
REFERENCES
(1)D,J.ProadandR.H.Fertig,"Formant-FrequencyTrajectories
inSelectedCVC-SyllableNuclei'″,J.Acoust.Soc.Amer.,47,
6,p.:L572(1970)・
(2)M.G.Kendall,"ACourseinMultivariateAnalysis",Charles
Griffin,London(1957)・
(3)H.L.Seal,"MultivariateStatisticalAnalysisforBiologists
(Reprintedwithminorcorrections)",p.97,Methuen,London
(1964and1968).
(4)T.Okuno,H.Kume,T.FagaandT.Yoshizawa,''Multivariate
StatisticalIlethods",p.A,Nikka-giren,Tokyo(1971).
(5)T,U.Anderson,"AnI:ntroductlontoMultivariateStatistical
Analysis≒JohnWiley&Sons,NewYork(:L958).
(6)D.F.Morrison,"llultivariateStatisticall・!ethods",
IIcGraw-Hill,NewYork(1967)・
(7)II.SiotaniandT.Asano,"MultivariateAnalysisTheory",
p.55,Kyoritsu,Tokyo(1967).
(8)II.SiotaniandT.Asano,"MultニivariateAnalysisTheory",
p.104,Kyoritsu,Tokyo(1967).
(9)M.SiotaniandT.Asano,"llultivariateAnalysisTheory",
p.60,Kyoritsu,Tokyo(1967)・
(10)K.Ito,':TheoryofMultivarlateStatisticalAnalysis",
p.162,Baihu-kan,Tokyo(1969)・
(11)W.Klein,R.PlompandL.C.X・7.Pols,"VowelSpectra,Vowel
Spaces,andVowelIdentification",J.Acoustニ.Soc.Amer.,
48,4,p.999(1970)・
(12)M.NakatsuiandJ.Suzuki,"FormantFrequencyExtractionby
InverseFilteringandMomentCalculationandItsEvaluation
usingSyntheticSpeechSound",J.Acoust.Soc.Japan,26,5,
p・.211(Jun.1970).
(13)S.Kltazawa,"AQuantitativeApproachtoCo-Articulationby
FormantニFrequencies'",Graduationthesisforthebachelar
-137-
![Page 141: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/141.jpg)
.degreeofKyotoUniversity(1971).
(14)M.SiotaniandT.Asano,"MultivariateAnalysisTheory",
p.21,Kyoritsu,Tokyo(1967)・
(15)II.Kubo,"IlultivariateAnalysisofcvcvsyllables',
GraduationthesisforthebachelardegreeofKyotニoUniversity
(1972).づ
(16)Y.Kato,K.OchiaiandS.Azami,"SpeechSynthesisbyRule
SupplementarilyUsingNatualSpeechSegments'゜,The6th
工nternationalCongressonAcoustics,Tokyo(1968).
(17)J.L.Flanagan,"SpeechAnalysisSynthesisandPerception(
Second,expandededition)",p.19,Springer-Verlag,Berlin
(1965and1972)・
(18)S.Siegel・,"NonparametricStatisticsfortheBehavioral
Sciences'',p・,47・,McGraw-Hill・,NewYork(1956)・
(19)K.TabataandT.Sakai,"EvaluationofSpeaker-Factorinvcv
Utterances",RecordofJointConventionoftニheAcous.Soc.of
Japan(May1973),(Tobelectured).
OTHERREFERENCES
- 一 一
- 一 一
- 一 一
- - -
S.S.Vilks,"l・lathematicalStニatistics'JohnUiley&Sons,
NewYork(1962).
W.J.DixonandF.J.llassey,Jr・,''IntroductiontoStatistical
Analysis'・,I'icGraw-Hill,NewYork(1951,1957and1969).
T.OkunoandT.Haga,"DesignofExperiment≒Eaihu-kan,
Tokyo(1969)・
K.TakeuchiandH.Yanai,"Foundationofl・:ultivariateAnalysis",
Tpyo-keizai-shlnpoh,Tokyo(1972).
-138-
![Page 142: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/142.jpg)
- ● - 四
- - -
- 一 一
- 一 一
- 一 一
- 一 一
- 一 一
一 一 -
THEAUTHOR゛SPAPERSRELATEDTOTHISTHESIS
K.TabataandT.Sakai,"MultivariateAnalysisofvcv
Syllables',RecordofJointConventionoftheAcous.Soc.of
Japan(Nov.1971).
T.SakaiandK.Tabata,"MultivariatestatisticalAnalysisof
vcvSy1:Lables',TechnicalReportoftheProfessionalGroupon
AutニomatonandInformationTheoryofthe工nstituteofElectronics
andCommunicationEngineersofJapan(Jan.1972).
T.SakaiandK.Tabata,"MultivariateStニatニiStニicalAnalysisof
cvcvSyllables':,RecordofJointConventionoftheAcous.Soc.
0fJapan(May1972)・
T.SakaiandK.Tabata,'"Three-DimensionalRepresentation
ofJapanesePhonemes:!,RecordofJointConventionofthe
Acous.Soc。0fJapan(May1972).
T.SakaiandK,Tabata,"Three-DimensionalRepresentation
ofJapanesePhonemes'‘,TechnicalReportoftheProfessional
GrouponSpeechofAcous.Soc.ofJapan(July1972).
T.SakaiandK.Tabata,"llultivariateStatisticalAnalysisof
cvandvvUtterances'・,RecordofJointConventionofthe
Acous.Soc.0fJapan(Oct.:L972).
T.SakaiandK.Tabata,"MultivariateStatisticalAnalysisof
vcvSyllables'",Jour,oftheInstituteofElectronics
andCommunicationEngineersofJapan56-D,p.63(Jan.1973)
K.TabataandT.Sakai,"EvaluationofSpeaker-Factorinvcv
Uttニerances",RecordofJointConventionoftheAcous.Soc.of
Japan,(May1973)べ(To‘belectured).
-139-
![Page 143: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/143.jpg)
14a項欠
![Page 144: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/144.jpg)
AppendixA
TheAsymptoticDistニributionoftheLikelihood
RatニioTestCriterioninMultivariatニeAnalysis
ofVarianceforFour-FactorDesignWitニhSingle
Observation
Letusconsidertheasymptoticdistributionofthelikelihoodratio
testcriterionindicatedbyEq°(4°7).Arrange゛ijkl(the11°iberof
whichisn°a・b・c・d)ofEq,(4,2)inacertainorder,andassumea-th
,一一IIvectortobe。%(a=l,..・,n),andx゛(pxn)=(.X;1,‥・,・χn).Thenlet
F'(゛l゛○゜[fic.]・G゛(bXn)゜[Sjoc'i・H'(cXn)゜[hk(Z)・andM'(dXn)
゜[1111α]9i-jhereタ‥
fμ
一
一
1,
0,
if心e{゛りklIi=1.・・・,b
otherwise.
kI}i
一
一 1,…,c;1=1,…,d}
andKjot>・^kQ!'・“liαaresimilarlydefined‘
1!eanwhile,ifwedefinethatl;(1xn)=(1,‥・,1),BI゛(Pxa)=
((ぢ,‥・,0!'),B2'(pXb)゜(βし‥・・凡')・Bj'CpXc)゜(r;・‥・・
ぺ)andB'(pXd)=(≪,',‥・,≪d),itispossiblet.0alterFq・
(4.2)to
e(x)=!,μ十FB1十GB2十HB3十MBu
=(>nFGHM)(μ'Bi'E2'B3'E^')″≡ZB
whereCdenotes"expectedvalue";Z°(lnFGroi)thedesignmatrixof
theexperiment;Bunknownparametermatrix.FromEq.(4.4),the゛'
densityfunctニionofxis
」享..一讐'.1-1(2TT)21Aト2e°ip{--trΛ(X-ZB)'(X-ZB)卜
wheretrindicatesthetraceofmatrix.Fromthisdensity,wewill
findRandAthemaχimumlikelihoodestimatesofBandΛ,respec-
-141-
![Page 145: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/145.jpg)
tively.Bsatisfies
Z'ZB=Z'X,
(SeeEq.(2.28))
(A.I)
butsinceZ゛Zhasnotitsinversematニrix,weobtainBbycomparing
thebothsidesof(A,l)undertheconditionofEq.(4.3),thatis,
Jn'B,°09・`?9ln'B^°O.
八μ
'χ.●゛●9
§
1・
゜(尽' 3°'゜3
a
al)゜(
゜X;1'゜'1‾'X"'゜'
1
9
°゜'yza°`‘;‾‘x'゜‘'゜1).“゜yB4
/
:‘.・
Therefore,themaximumlikelihoodestimateofΛis
(A.2)
J=(χ-ZB)・(χ、-ZB)=(χ-lj_F礼一G92-臨3い瓜)・
・(X-1丿-FEi-GB,-HB3-mu)
abcd=Σ
i=1
ΣΣ
j二1k=1
Σ
l=1
(Zり.)
(A.4)
(゛ijkl‾゛i…‾゛:.:..-X..k.-X゜¨「+ろ゛¨¨」(A°3)
whichcoincideswiththeresidualRitselfofEq.(4.5)・
Subsequently,considerthehypothesisthatthereisnoeffect
offactorA
Ha:(zlこ・・.こ(za=0,thatis,Bi=0
《公《《《Letμ,B2,B3,Bit,andΛbethemaximum:Likelihoodestimatesニofμ
B2,B3,BいandΛundert:hehypothesisK^,
then,
;
=゛・‥・,Bz'=(x.i.
:
-x…
'
.,・・',X・b.
:
-z…
:)
《。,。R心1余http://www...
nA=(X-l
n
μ-GBo-HB5-MB4)(χ-lnμ-GBo-HB3
…,§4
而4)
abcd
°ぶ1j=ik=1ぶ1(゜x°ijkl―X.j..~JC..k°‾ぷ...1+2x....)'(A:ijki-AT.j,,-ぷ・・k-~ぷ"`1
aabcd
-t-2*....)=bcdぶ1(凪'゛゜-JC....)'(‘゛i¨゜-X)+ぷlj?FIぶ1ぷ](jfijklXi°¨
-142-
![Page 146: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/146.jpg)
areallidempotentmatrices,
‾z・j・.‾z・・k.‾x‥・1+ろx・・‥)・(zijkl‾xi‥.‾‘x;・j・●‾z‥k.‾z‥・1キろz・・・.)
=Q1-4-R
Thus,thelikelihoodratiocritニerionforthehypothesisHais
|謡1㈹
"|nA|1マh+p-l
(SeeF-q.(2.37))
(A.5)
(A.6)
Thedistributionofl,・underthehypothesis(A.4)isobtainedfrom
Cochran'stニheorem.
LetY=χ-ZB=χ-1,1μ二FBi-GB,-FB3-IIBii,andbreakdownY'Y
asfollowsこ
Y゛Y=Y゛(ヰ)Y十Y'(百一斗)Y十Y'(万一脂Y十Y'
(Jy一子)Y十YI(響一寧)Y十Y'dn一昔-ご一首
)Y
一
一
-
-
一
一
Y'WoY十Y'WiY十Y'W2Y+Y'W3Y十Y'W4Y十Y'WsY
VO十V1十V2十V3+V4十V5.
(ln:n-dimensionalunitmatrix.)
I<7herenxnmatrices
forexample,as
!・T09・゜・9W5
-143-
'.bed
こ‾牛
゜Wi
Theseareeasilyverifiedifwepayattentiontofollowingrelations.
![Page 147: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/147.jpg)
F゛F°bcdlg,I'F°bedla゛;ヽ3Fla'゜IngF'G°cdla'lb″'
Therefore,takingthetraceofmatricesinordertoobtainthe
ranksofeachmatrix,
aΣJ
。Σ心
晨 n
一一n
-
- -bed
tr(W2)=b-1,trCW,)=c-1.tr(W4)=d-:L',
tr(W5)=n-a-b-c-d+3
-1=a-1
(A.7)
areobtained.Andsincen=(1)十(a-1)十(b-1)十(c-1)十(d-1)十(n-a-b-c-
d+3),Vo,・・・>v5‘areindependentofeachother,andcanbeexpressed
tyCochran'stheoremasfollows:fi,・I・
Vi°j=1
"j≪j “j(lxp)~N(O,A),(i=0,1,...,5)
VJhere,りistherankofW;showninEq.(A.7),and"jareassumedto
bedistributedindependentlyofeachother.(X但enて戸p,V.is
distributedaccordingtoWishartdistributionU(Λ,P,てi).)
Now,keepingattentiont01a゛BI°'b゛B2°Ic'B3°Id・b.°09
substituteX-ZBforYinVj,andexpressViasthefunctionof
・^ijkl.Then
Vi゛bedjl(Xi...-x....-cCi)'(゜Xi..゜-x....-rfi)
ParticularlywhenHa:1?1°Oj;strue!Vi°Qi
Ontheotherhand,
abc
V5°ΣΣΣi=1j=1k=1
dΣμ
(゛りkl‾.゛゜i...-x・j・.-゛‥k.-.x°‥・1ギ叙….)
'(xijkl‾xi・.I.‾Z・j・,X..Ir.X…l+ろX….)゜R(A.8)
V5=Rhold・inomatterwhatBI~B41!laybe,andR~W(Λ,p,T5)when
●●・.ミ●.・●;・I・r■・
T5°n-a-b-C-d+31p.
-144-
![Page 148: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/148.jpg)
Therefore,themomentof11JofEq.(A.6)isaccuratelyobtained
sothatニwecanlearnitsasymptoticdistributionbyBoχ.(9)
Namely,リ=-{nぺ2-(p十ill+l)/2}loRwisdistributedasaChi-
squaredvariatニ;ewithP兌1degreesoffreedomトaS十thesampleSizentends
toinfinity.VJhereli=a-l.£2=b十c十d-2,n-a-b-C-d+3≧p.
-145-
![Page 149: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/149.jpg)
Appendix・B
TheIntersectionProducedbytheEllipsoidandthe
StraightlineorthePlane
Thepointofintersection,producedbytheellipsoidZA‾IZ'=1
andthestra・ightlinekCthroughtheorigin(krepresentニSarbitrary
realnumber),isexpressedby±c//ca"''c'.(Itisobtainedbysub-
StニitutingkCfor.X°inZA‾IZI°1.)
Asthedistancebetweentheoriginandthepointofintersection
is
(B.:L)
thisexpressioncoincideswithV^whenCistheeigenvectorof
AC'=λC'.(c=入A‾IC゛;CC゛=λCA‾IC″;λ=ccVCA'Ic.)
Theintersection,drawnbytheellipsoidandtheplanedeter-
minedbyCiandC2,willbeobtainedbyconnectingthepointsof
intersectionswhichareproducedbytheellipsoidandarbitrary
straightlinesontheplanewhichpassthroughtニheorigin.
SinceanarbitrarystraightlineisexpressedbykiCi十k2C2upon
choosingkiandk2arbitrarily.thedistancebetweenthepointof
intersection(madebythestraightline)andtheoriginisobtained
bysubstitutingklCI十koCpforCinEq.(B.l).
-146-
![Page 150: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/150.jpg)
AppendixC
VarianceofFactorA
LeteA(゛りkl)betheexpectedvalueof゛りkl(゜μ十αj十βj+71
十δI+61jkl)underthehypothesistニhatHA:α1°¨'゛αa°Oistrue.
Then,fromEq.(A.5)inAppendixA,themaximumlikelihoodestimate
ofeA(刷jkl)is
a
A(zijkl),゜j゛βけ;けs,
=z….十(X.j..-X....)十(x..k.一心‥.)
十(x...1-ふ‥●)
人余《余Sinceμ‾μ-0,βj‾βjきO゛rk‾rkきoandδ1-≪lso,
(C.1)
thedifferencebetweensa°pievectoi:・^ijkland4(゛jjkl)isapproxi-
matelyequaltoαi十りiklasfollows:
《。《
刷jkl‾eA(゛ijkl)゛(μ‾μ)十(zi十(βj‾βj)
十(rk-Fk)十(δl-21)十りjkl
~a.十陥kl
(C.2)
Therefore,theabovedifferencemaybeconsideredtobethedeviation
onlyduetofactorA(Ai,...,Aa).
(Ofcourse,ifhypothesisI--Aisactuallytrue(itmeansofj=0),
刷jkl‾ら.(りikl)゛njklandthisdifferenceistheresidualitself.)
Hencethematriχofsumsofsquaresandcrossproductsonthebasis
ofthedeviation(C.2)
abcd会4乱見ぶ1叫kl呵【りjk】))(りjkl‾eA(りjkl))
ヘダ
1
ムム,t<りjkl-゛¨¨‾(゛lj°゜‾゛....)-(゛¨k‘‾゛¨¨)
‾(゛…l‾゛….))・(りjkl-゛・...-(゛・j・.‾゛….)-(X・・k.‾゛….)‾(z…1‾x・・‥).)
-147-
![Page 151: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/151.jpg)
=bcdvO£:i‥.-x‥‥)゛(xi‥.-x‥‥)
十
万
j
ダムA(xijkl‾xi‘..-X.j°.-X゜゜k°¨づ¨'けろx¨゜゜)
ニQI・十R
'(J:ijkl‾xi"'‾:xj‘j¨‾X°‘k'‾゜X;●●●1+5*・・●●)
canbeconsideredtorepresentthevarianceoffactorA
(C.3)
Moreover,letAbethemaximumlikelihoodestimateofthe
covariancematrixAunderthehypothesis\
fromEq.(A.5)inAppend工XA
《nA°Qi十R,
whoseexpressioniscoincidentwithEq.(C.3)
-148-
:α1°‥.゜αa.Then,
![Page 152: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/152.jpg)
ApoendiχD
TheVectorSpaceNormalizedbytheResidualR
Supposethefollowingnonsingulartransformation(notalways
orthogonaltransformation).
.x・→i=χ(R/n51 (D.:L)
!・ThereRistheresidualgivenbyEq.(4.5),nisthesamplesize
(n°abed),and.randiarevectorsintheoriginalandnewspaces,
respective:Ly・
Then,十Qに'き1(zi‥.-i‥‥)1(,x;i‥.-z‥‥)(QlgivenbyEq.(4.5))
istransformedto
こそj゛‾
とタ](Xj...-X….)'(ij…‾i….)
a-!-
゜‾yE:
1(R/11)2(゛j...゛`'`゜)゛(゛i'∵‾゛゜...)(R/n)
lal
°R‾7{EI
(ら‥‾゛‥‥)'(゛i…-χ….;}R"2
=R‾九1R唾
and
!R(pxp)4
n‥、
Similarly,:.
悩ノRj'仙・
and
∠―(Qi+R)→十(61+ii)∠μ
(QI+l)す,
i[ま[Qi+R)]X=X[Iけ(QI+n)Iドを]‾li'・
=i(R/n)2 t'÷(Qi+R)]‾1
=z(十(Q1十R))'1z'
・
(R/n)23c'
-149-
■
2
(D.2)
(D.3)
(D.4)
(D.5)
![Page 153: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/153.jpg)
AppendixE
TheKolmogorov-SmirnovOne-SampleTest(18)
TheKolmogoでov-Smirnovone-sampletestisatestof:goodnessof
fit.Thatis,itisconcernedwiththedegreeofagreementbetween
thedistributionofasetofsamplevalues(observedscores)andsome
specifiedtheoreticaldistribution.Itdetermineswhetherthescores
inthesamp1ecanreasonablybethoughttohavecomefromapopu:Lation
havingthet:heoreticaldistニribution.。
Briefly,thetestinvolvesspecifyingthecumulativefrequency
distributionwhichwou:Ldoccurunderthetheoreticaldistニribution
andcomnaringthatニwiththeobservedcumulativefrequencydistribution.
Thetheoreticaldistributionrepresentswhatwouldbeexpectedunder
H0。
LetFO(X)=acompletニelyspecifiedcumulatニivefrequencydistribu-
tionfunction,thetheoreticalcumulativedistributionunderKo.
Thatis,foranyvalueofX,thevalueofFo(X)istheproportionof
casesexpectedtohavescoresequaltoorlessthanχ。
AndletSN(X)゜theobservedcumulativefrequencydistribution
ofarandomsampleofNobservations.1・Tiere,χisanypossiblescore,
SN(X)=k/N,wherek=thenumberofobservationsequaltoorless
thanχ。
Nowunderthenu:LIhypothesisthatthesamplehasbeendrawn
fromthespecifiedtheoreticaldistribution,itisexpectedtニhatfor
every゛8:Lueofx・SN(X)Sho゛1dbefairlyclosetoFO(X).Thatis,
underHowewouldexpectthedifferencesbetweenSvj(X)andFo(X)to
besmallandwithinthelimitsofrandomerrors.TheKolmogorov-
Smirnovtestfocusesonthelargestofthedeviations.Thelargest
valueofFO(X)-SN(X)iScalledthemaximumdeviation,D;
D°maximum1Fo(X)‾SN(X)|
-150-
(E.I)
![Page 154: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/154.jpg)
ThesamplingdistributionofDunderHoisknown.TheTableE.1
givesDa゛tニhecritニicalvalueofDwhenNisover35.1fNisover
35,onedeterminesDa゛thecriticalvaluesofDbythedivisions
indicatedinTableE.I.白Foreχample,supposearesearcheruses
N=43cases・andsetsa=・.05.Table(E.l)showsthatニanyDequalto
orgreatertニhanDa°ミFy捗‘willbesignificant.ThatiS゛anyD,as
definedbyformula(E.I)・whichisequa:Lto01°greatei°thanD(Z°
譜=.207willbesignificantatthe05level(two-tailedtest)
TableE.1TableofB^(criticalvalueofD)intheKolmogorov-Smirnovone-sampletest.
Sample
size(N)
a.LevelofsignigicanceforD=max Fo(X)-SN(X)
0.20 0.ニL5 0.10 0.05 0.01
Over35 1.07
-\「N‾
1.14
一汗
1.22可N‾
1.36
ク‾1.63可『
-151-
![Page 155: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/155.jpg)
Appendix,F
TheAsymptoticDistributionoftheLikelihood
RatioTestCriterioninHultivariateAnalysis
ofVarianceforFour-FactorDesignwithRepeated
MeasurementsinWhichA11InteractionsAreDisregarded
Letus・considertheasymptot:icdistributionofthelikelihoodratio
testcriterionindicatedbyEq.(5.17).InEq.(5.12),arrange.x;りklni
(thenumberofwhichisn=abcde)inacertainorder,andassumea-th
vectortobeX£^(a=l,...,n)andχ'(pxn)゜(Xi',..・>・*■nz・Thenlet
F'(aXn)=[fiα],G'(bxn)=[g,μ],H'(cXn)=[hkα],andII'(dXn)
゜[ma],where.'尚`
fi(Z=
]
.,
0,
if-≫^tfI^ijklmlj=1,…,b;k=i,…,c;1=1,…,d;ni―1,…,e}
otherwise,
andgjαタhkα9n\\aaresimilarlydefined・
Meanwhi:Le,ifwedefinethatU(1xn)=(1,‥・,1),Bで(pxa)=((z11
‥・,αa'),B2'(pXb)゜(βら・・・,βb‘),Bろ'(pXc)=(r;,‥・.r')
andB4'(pXd)゜(δll9・・・,δd'),itispossib:letoalterEq.(5.:L2)
to
e(x)=lnμ十FB1+GB2十HB3十MBi,
=(lnFGml)(μ'Bi'Bo'Ba゛B4')'sZB
where6denotesI:expectedvalue";Z=(InFGHII)thedesignmatrixof
theexperiment;Bunknownparametermatrix.FromEq.(5.1A),the
densityfunctionofxis
.巴
(27T)2--
|Λ12exp{ 一手tΓΛ-1(X - ZB)1(X-ZB)},
wheretrindicatesthetraceofmatrix.Fromthisdensity,wewill
findBandK-themaximumlike:LihoodestimatesofB上andA,
-152-
![Page 156: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/156.jpg)
respectively・Bsatisfies.
z゛ZB=Z'X,
(SeeEq.(2.28).)
(F.I)
butsinceZ゛Zhasnotitsinverseir.atrix,weobtain%bycomparing
thebothsidesof(F.I)undertheconditionofEq.(5.13),thatis,
'n'Bi°0,‥・1・n'B,°O.
g=z‥‥.,沁'=(a11,…,もI)=(a・1…:-z….:....Xa...:-.χ・….:)
....B4(F●2)
Therefore,themaχimumlikelihoodestimateofAis
∧∧八∧∧A∧s岨=(X-ZB)゛(X-ZB)=(X-lnμ-FBi-GB2-HB,-l!B4)
べX-lj-F礼一・G礼一池3-M礼)
abcde°ぶ:
lj脊
I
Jj
11
召
ln
F
I(Xりklm~-'^i・...-X・j・・X.,J{..X,,・,.+5・X)
゜(゛ijklm-゛i・…-j;・j・・.-゛・・k・.-゛…l.4'3゛・・・・.)(F.3)
whichcoincideswiththeresidualR.itselfofEq.(5.15).
Subsequently,considerthehypothesisthatthereIsnoeffect
offactorA
HA
Letμ,
:(ZI°‥.゜αa°0,thatis,B,°0 (F.4)
B29B3,Bit,andAbethemax:imumlikelihoodestimatesof
μ9B2タB39BいandAunderthehypothesisl!A≫then.
;l°z・・…,n2'゜(z・1・..―X・・・.:,‥・,z・b.i:‾z・・・.:).¨・,n4・
nA=(X-lnβ一品2-H礼一MB4)'(χ-lnβ-GB2-HB3-MB4)=
(ぶり
dΣけ
。Σり1
bΣ・F
aΣJ
・・・・.)・(zijklm‾.x;・j・・.‾X
Σ
1m=1
・k..-AT...I.
-153-
![Page 157: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/157.jpg)
-i-2x)=bcde。Σ
' ド
abcde
1
(‘・V|・■・■a^a・■・゜) 'iXi...゜‾'X°)ナぶ
1j=1k=1
ぶ
lrT
y
l(£りk1“1
‾Zi・・‥‾Z・j・・.‾Z・・k‥‾X・・・|.+ろ.X;・‥‥)I(a;ijklm‾.x;i・・‥‾ぷ・j・..-X..k・・
一X..小斗ろぶ)゜Qi+R-
(F.5)
Thus,thelikelihoodratiocriterionforthehypothesisKais
w=-いに|旦1
い||Oi+r|.
(SeeEq.(2.37))
(F.6)
ThedistributionofZ∂underthehypothesis(F.4)isobtainedfron
Cochran'stheorem.、・
LetY=χ-ZB=χ-lnμ-FBi-GB,-KB3-MB4、andbreakdowny'Y
asfo1:Lows:
十Y'(
)Y十Y゛(FF曹
-bcde
in≪n-
)Y十Y゛(GG'In'n
ΞΥ'WoY+Y'WiY+Y'W2Y十YIW3Y+Y゛W4Y十■Y'W5Y
ΞVO+V1十V2十V3十V4十V5.
(ln:n-dlmensionalunitmatrix.)
WherenXnmatrices
forexample,as
x・712=(⊇?y
)Y十YI〔Ji;‾1戸〕Y
一響'÷ろ|五に|)Yabcen
WO゛゜..,V・areallidempotentmatrices.
-
'n'n°Wi
,n
Iχ
Theseareeasilyverifiedifwepayattentニiontofollowingrelations.
F'F=bcdela,≫nF=bcdela,Fla=ln,F'G=cdelalb・
-154-
![Page 158: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/158.jpg)
Therefore,takingthetraceofmatricesinordertoobtainthe
ranksofeachmatニrix,
tr(Wo)=1,tr(Wi)=jUヱj
l‰2‾そJji‾i♪a-1,tr(W2)=b一丿L,trCWa)=c-1,tr(W1,)=d-1,
trCWs)=n-a-b-c-d+3 (F.7)
areobtained.Andsincen=(1)+(a-1)十(b-1)十(c-1)十(d-1)+(n-a-b-c-d・
+3),Vo,・・・>V5canbeexpressedbyCochran'stheoremasfollows:
r.
Vi°Σ
に1"j"j' Uj(lxp)~N(O,A),(i=0,l,..5)
1・Jhere,T:istherankofWishowninEq.(F.T),and≪Jareassumedto
bedistributedindependentlyofeachother.(WhenT:>p,V:isdls-
tributedaccordingtoWishartdistributニionW(A,p,Ti).)
NOWタkeepingattentiont0・a'Bl°lb'B2°≫c'B3°ld'B4°07
substituteX-ZBforYinVi,andexpressViasthefunctionof
Zりklm°Then
V1=bcdeE(゛i‥‥-・x°…・-a.y(X.….-゛・….-CC,).1=1
Particularlywhen!lA:B1=Oistrue,Vi=Q1.
Ontheotherhand,
V5=
abcΣΣΣ
eΣ
dΣ
i=1i=1k=11=1m=1(xijklm‾゛i….‾゛・j・・.‾゛・・k‥‾゛…1.+3X..…)l
'(゛ijklm‾j:i….‾゛・j・・.‾゛・・k・.‾X…1.+3X…・・)=旦.
V5=RholdsnomatニterwhatBI~B4maybe,andR~W(A,p,て5)when__
T5=n-a-b-c-d+3≧p.
Therefore,themomentofw0fEq.(F.6)isaccuratelyobtained
sothatwecanlearnitsasymptoticdistributoinbyBoχ.(9)・
Namelv,リ=-{n-G-(p+らり)/2}1o肪iSdistributedasaChi-
squaredvariatewithpj!・1degreesof<^reerlomasthesamplesizentendsJ
toinfinity.VJhere£1=a-l,£2=b十c+d-2,n-a-b-c-d+3≧p.
-155-
![Page 159: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/159.jpg)
AppendixG
TheAsymptoticDistributionoftheLikelihoodRatioTestCriterioninIlultivariateAnalysisofVarianceforDivided-TypeTwo-・FactニorDesignwithRepeatedlニeasurements
Lとtusconsidertheasymptニoticdistributionofthelikelihood
ratiocriterionindicatedbyEa.(6.12),
'|■・・Arrange゛りk(thenumberofwhichisn=abr)ofE(!.(6.ね)inthe
●j・-lfollowingorder,inwhichletぷαbethea-thvector:
Z(Z°*ijk,
where a=(i-l)br十(j-l)r+k,
(1!i!a,1≦j≦b,1≦k≦r,anda=1,...,n)
Similarly,arranger4'inthefollowing:order:
rq°咄),(叫1副・瓦'7j゛)(a.゛1111a)・
whereq=.(i-l)b+j(q=!,..,ab).
LetG{(axn)=[g池],G2'(axn)=[g?11,1!・(bxn]=H;十H2;
Hi'(bxn)゜[h?ル],H,'(bxn)゜[h(ル)'L'゛(8b)q1)祀[ペル],and
L,'(abxn)=[11ユ.1,
where
巡=
瑞
1
0
1
0
if1≦i≦ai,andifj(1≦j≦b)andk(1≦k≦r)exist
suchthata=(i-l)br4-(j-l)r十kholds,
otherwise,
ifai+l<i<a,andifj(1≦j≦b)andk(1≦k≦r)exist
suchthata=(i-l)br十(J-l)r十kholds,
otherwise.
-156-
![Page 160: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/160.jpg)
hμ=
関戸h
(1)
qα
口がI
一一
一一
1
0
1
0
1
0
1
0
ifi(1≦i≦a-i)andk(1≦k≦r)existsuchthat
af=(i-l)br十(j-l)r十kholds,
otherwise,
ifi(ai+1≦i≦a)andk(1≦k≦r)existsuchthat
α=(i-l)br十(j-l)r十kholds
otherwise,
If1≦q≦a^b,andifk(1≦k≦r)・existssuchthat
a:=(q-l)r十kholds,
otherwise.
ifalb+11qiab,andifk(1≦k≦r)existssuch
thatcc=(q-l)r十kholds,
otherwise.
Meanwhile,wedefinethatX'(pxn)゜(zl゛,‥・9xnl)s
1:n(11)(11):n-d:imensiona:Lunitmatrix・ln'(1)(11)゜(1,..‘μ)パl[フ]z(1'"`)s
on(1xn)゜(09°..,0),Jali(1xn)゜(1いbc゛032bc)゛J327(1xn)
゜(Oalbc'la2bj),andTq,(axab)=[(Γal)iq],T(axab)゜[(Γ32)iq]'
Γbl(b)(3b)゜[(rbl)jq],r(bxab)゜[(Γb2)jq]^where
(Γal)iq°
1
0
if1≦i≦ai≫andifj(1≦j≦h)existssuch
thatq=(i-1)b+jholds,
otherwise,
-157-
≪..o
E
1
「IJ
O:0
![Page 161: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/161.jpg)
(Γa2)iq°
(Γbl
)jq --
(rb2)jq゛
1
0
1
0
1
0
ifal+1≦i<a,ar!difj(l≦j≦b)exists
sucht!latq=(i-l)b十jholds,
otherwise,
if1≦q≦“lb3andif
that・q=イi-1)b+j
otherwise.
i(1≦i≦a)existssuch
holds,
ifallbf1≦q≦ab,andifi(1≦i≦a)exists
suchthatq=(i¬1)b六十jholds,
otherwise.
Besides,definethatAi'(pxa)゜(α(1)'...・(Zjl)I・l(i112)・A2'(pxa)゜
(肱1タ屯伊1(2ぺ'りαj2)')9T'(pxb)=(β1い‥,βb'),Ti'(pxb)゜(il(1)1
‥'・lb(1)')5Tz'Cpxb)゜(11(2)≒゛゜'・ib(2)')9Ci'(pxab)゜(rl(1)≒"゛9ral(ぴ・
略12b),C2i(p)(8b)゜(皿11b,(9)'・‥゜・rj2)I)・(ai+a2=a).
ThenitispossibletoalterEq.(6.U)to
e(x)゜lnμ十Jalμ(1)十Ja2μ(2)十GiAi十G2A2+HT十HiTi十H2T2
十LiCi十L2C2ヽ
゜(inJaiJa2GiGjHH1H2L1L2)(μ`μ(1)゛μ(2)'Ai'A2'T'Ti'T2゛Cl'C2')'
ΞZB,(G.1)
where6denotes"expectedvalue":Z°('nJa-|Ja2GiG2HHiH2LiL2)
thedesignmatrixoftheexperiment;B°(μ゛μ(1)゛μ(2)'Ai'A2'T'Ti'T2'
Ci'Cz')'unknownparametermatrix.Eq・(6.6)correspondstoEq.(G.2):
alμ(1)十a2μ(2)゜0,
(la;・Oa2)AI°o・(Oal>lap)A2°o'
-158-
![Page 162: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/162.jpg)
Andalso
Ib
Γal
T=0
C1=
】b'Ti°091b'T2°0,31T1+32T2=0,
Γ゛12C2°ΓblCI°ΓbpCa°0
(la^b'9Oa2b)Ci°(031b9132bl)C2°0
holds.
FromEq.(6.3),thedensityfunctionofxis
(2TT)‾苧|川弓exp{一子tΓΛ'■"(X-ZB)'(X-ZB)},
(G.2)
(G.3)
wheretrindicatesthetraceofmatriχ.Fromthisdensity,wewill
findBand/?
respectively・
Bsatisfies
In'n
'n'H
Λμ
=n
一一
themaximumlikelihoodestimatesofBandA,
茫と=z‥jt)-;x;・・・9
-159-
(G.4)
(G.5)
Z'ZB=z'x,
(SeeEq.(2.28).)
butsincez'zhasnotitsinversematrix,weobtainBbycomparing
thebothsideof(G.4)undertheconditionofEq.(G.2)andEq.(G.3).
゜.χr・.・5
&i(1)゜:zi‥‾.x;‥.(tへ
ら(t‰(z・.(t)_゛…(1))-(゛・j.-゛‥.)・
βj=・・レーJC・・・・
'it"^lj・-Vi..X.j.十'x°‥'(t≒
(t=l,2).
Theseareverifiedifwepayattentiontofollowingrelations:
'lnIJal゛fllbr'lnIGI゛b「(1311'032)
arlu'・11.:Hi°¥Ib".In'H2°32「Ib''ln'Li°r(lalbl'Oa2b)
![Page 163: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/163.jpg)
1n'L2ニr(0°ibllI)'Ja;J31'゛a-ibr)J311J32ニO'J゛111Qiニb「(1311'032)゛
Ja;G2ニ0,J'H°air1b'゛Jal'HIニalrlbl'J31'H2ニO'J311L1ニ「(laib''0a2b)タ
J311L2ニO'Ja^^ao°aobc.Ja2'Gi゛O゛J32'G2°b「「031」32')'
Ja21Hニa2「lb'゛Ja2IHFO'J321H2ニ゜2「V.Ja2'Li°o'
J321L2ニ「(Oaib'1a2b')
G1゛G1=br10
L
Gi゛Hi=Gi'H,
G2゛G2=br
00
L
G11
100 G11G21=O・,Gi'H=r[
;lj
1b''
H2°09Gi'Li°rr11,タGjLo'°09
I
?
2
]
s
Ga'H=r[;jj]
1b'゛G21H1°O゛
Go'Rz°Gz'H,G2'Li°O゛G2'L2゛ΓΓ32゛
H'H°ar工b>H'Hi°alrlbタH'Hz°a2rlbタHi゛H1°a.rib
Hi'H2=0,H'L,=HiLi=rrK,H2'Li=0,Hz'Hz=a2rlb,
i七2°H2'L2°lΓΓb2'WL2ぷ
'Li'Li°lr[乱lbO]
・,--………『0
Li'La°0゛L2゛L2゛r[;1
2b
]
Therefore,themaximumlikelihoodestimateofAis
(G.6)
J°(X-ZB)
'(X-ZB)゜(X-1nA-Jil
l
β(1)
‾J・12
μ(2)
-GiAi-G2A2・-HT-Hifi-H2T2-LiCi-LoC2)"
(X-Inβ‾Jal
β(1)‾Ja2β
(2)
‾G1
λ
l‾G2
λ
2‾
�
-HiTi
-H212-LiCi―L2C2)
一一
bΣJbΣJ
・j.J
r●ぶ
1(ぶりk‾りj゛)'(ぶりk‾りj°)
「ぶ
1(‘x'ijk‾りr)'(ぶりk‾りj゛)
abr
゛ぶlj
lミF
I
JF
I(zりk‾りj.)'(Xりk‾りj°) (G.7)
whichcoincideswiththeresidualRitselfofEq.(6.10)・
Subsequently,considerthehypothesisthatthereisnoeffect
-160-
![Page 164: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/164.jpg)
(G.10)
1
offactorA(1)一A,(2)
Ko:μ(1)=μ(2)゜0 (G.8)
Letμ,Ai,A2,T,Ti,T2,Ci,C2andAbethemaximumlikelihood
estimatesofμ9Ai,A29T,Ti,T2,Ci,C2,andΛunderthehypothesis
Ho,then'
ぶ=
ll(叱
≪t≫a7
x・‥j?)゜zi‥_jp(t)
(゛゛j°(t)‾゛‥'(t))‾(゛・j・~゛...)
゜X.j.X...,Tjj-Xjj.―X-^.,-X..(t)十V,(t)
(t=l,2). (G.9)
nΛ=(X-ZB)゛(X-ZB)=(X-lnμ-GiAi-G2A2-HT-HiT,
-
H2T2-LiCi-L2C2)゛・(X-lnμ-GiAi-G2A2-HT-HiTi
,-H2T2 -LiCi-L2C2)
゜χ
j
y
l
ぶ{(範jl)‾゛...)+(り‾゛り゜)白(゛'¨(1)‾゛゜¨)
I
バ勺k-*ii°)}几乱もユ{(゛…(2)_゛…)゛(i印う町川'
りリ‥.(2)‾孔・.)十(ぷりk‾xり.)}
=br'lat(゛・・.(t)-゛・‥)'(゛‥.(t)-z...)
1=1
十Aχk
シりjk‾乱ゴ(xりk-*ii.)
-
-
ql十R
-161-
![Page 165: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/165.jpg)
Thus,thelikelihoodratiocriterionforthehypothesicHois
|。則|引w=一一こー-'-
ln別|(h+則
(G.ll)
。(SeeEq.(2.37))
●●●瀞●'●Thedistributionofz・・underthehypothesis(G.8)isobtainedfrom
Cochran'stheorem.
Let Y°X-ZB°X‾lnμ‾Jal μ(1)-J
a2μ(2)‾GIAI‾G2A2
- HT-HiTi-H2T2-LiCi-L2C2,
andbreakdownY'Yasfollows:,
Y'Y°Y゛(二万is-)Y十Y゛(ln一三ぺ這)Y
=Y'(-il≒い一一)クy十YI(j計J!L十
十Y゛(
十YI(
+
air
GiGi'-
br
ヰ -
l
匯)Y
-づ柚タ!')Y゛い(jVンj‾べjぶ;)↓
Inl;1一一
Ja2Ja21可こ
)Y十YI(
-162-
HiHi'+
H2H2'HH゛
a2「
LiLi'GiG'i
GoGj.-br-
a「
H2H2'-
a^r
)Y
HH゛
-a「
_JajJai-
‘a-ibr
一一HiHi'十j≒jlユー)Y十Y'(≒-り2'-
WherenXnmatrices-w..w,X'>,---,We
ces,forexample,as
a2「
ΞΥ'WoY十Y'WiY十YIW2(14十Y゛xjjY十Y゛F3Y
十Y'WuY十Y'WJI)Y十YI匈2)Y十Y'W6Y
≡vo十VI十V2(1)十V2(2)十V3十V4十V5(1)十V5哨十V6.(G.n)
areallidempotentmatri-
![Page 166: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/166.jpg)
●
-々斗ぎ!ユヅダ=?1_GI/-:^i_£3i_'=waibrbraibr2
Theseareverifiedifweattentiontothe!relationsofEq.(G.6)and
■・・・.・'・■,■り¶WII¶が倆,・■¶-・・4r一一j■--・●'・・
followingrelatニions.
LI
「乙al
.
E」
゜Li
i
引10
{L
IG
100
000yo
じ[卜~L
LG
I
a2
゜G1,
{――}
0
「I-~。」
T『
゜L2,
l
G2・i
I'1ぺ゜G
ム
L√几)yHI'LI
Q]
゜Jai
GI
CL」
゜・Jai゛Hilb°Ja!゛
L匹;=GいLがいり,L
ぐづ
=≒
G2
[7:1]
゜J32'H2'b=Ja2゛
HIb=In,HH'=(Hi十H2)(HI十H,)'=HiHi'十H2H2',etc.(G.12)
Therefore,takingthetraceofmこtricesinordertoobtainthe
ranksofeachmatriχ.
tr(Wo)=1,tr(Wi)=1,゛(匈1))゜1n1.(1)>2器
=ぺU牛一1=ai-1,tr(Wj2))=a2-1,tr(W3)=b-1,
tr(W4)=b-].,tir(WJ1))=alb-ai-b十1,tr(W5(2))゜32b'a2
- b+1 tr(U6)=n-ab (G.13)
areobtained.Andsincen・(1)十(1)十(ai-l)十(a2-1)十(b-1)十(b-1)十(aib-ai
-b+1)十(a2b-a2-b+1)十(n-ab),
`J。,Vいvr)9¨¨,V5canbeexpressedbyCochiran'stheoremasfollows:
163
![Page 167: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/167.jpg)
岬=Iい≪j'"j Uj(ixp)二~N(0・A)・
雪
(
じ・・・,6J(G.14)
where,7r7istherankofVI■shovヽ7ninEq.(G.13),and≪jareassumedto
bedistributedindependentlyofeachother.‘(WhenぐをP,V?is
distributedaccordinetoUishartdistributionU(Λ,p,T?)).)
Now,keepingattentiontoEq.(G.2),substitutex-ZBfor
YinVi,andexpressViasthefunctionofzりk Then
Visbr
jal(.x;!?.-x...-μ(t))I(x!?.-X...-μ(t))
1二1'
ParticularlywhenB.:μ(1)ごμ(2)゜Oistrue・9Vi°Qi
Ontheotherhand.
V6=
bΣ
aΣ E(Xijk-X・'.)I(zijk‾'x;'¨)゜RμIJニ1k=1
(G.15)
(G.16)
V6=RholdsnomatterwhatBmaybe.andR~U(Λ・PjTe)whenT6°n-ab
≧P●ノー,・
Therefore,themomentofwofEq・(G.ll)isaccuratelyobtained
sothatwecanlearnitsasymptoticdistributionbyEox.(9)
Namely,V=-{n一見2-(p十見1十:L)/2}10glむ・isdistributedas-
Chi-squared・variatewithP見1degreesoffreedomasthesamplesizen
tendstoinfinity,where£i°1,聡十見2°ab,n-ab≧p・
-164-
4
/
●'
![Page 168: Title Multivariate Statistical Analysis of Speech ….../rL /1 CV CVVV vcv cvcv \(t) Speechsound ・ ・一@一 ・一 @一 Matrix A(pxq) A(pxq+1) A゛ A‾1 国 tr(A) rank(A)](https://reader034.vdocuments.site/reader034/viewer/2022050508/5f991a127730897b676af647/html5/thumbnails/168.jpg)
1 1 1
1
♂ ・
AppendixH
MatrixofSumsofSquaresandCrossProductsinSubspace.
LetB(pxq)betheorthogonalprojectionmatrixofp-dimensional
spaceonitsq-dimenslnalsubspace.
X(1Xp)→X(1Xq)=xB (H.I)
And,letamatrixQbesuchamatrlχofsumsofsquaresandcross
productニSasshowninEq.(2.15).
Q(pxp)=2(xi-z.)'(xにX.),X.=!11Xi
i=lni=1
ThenQistransformedbytheaboveprojectionasfollows.
where
6(9・・9)=j(ii-i・.)・(Xi-X.)
1=1
=.ZB'(Xi-x.)'(Xi-z.)Bt=1
=B゛n-z.)'(xi-x.))Bt=1
°B'QB,
WZ
1
-n
n
Σ
i=1
W*i
(H.2)
(H.3)
NowletusconsidertheEq.(7.13).
SinceRisamatrixofsumsofsquaresandcrossproducts,and
B=A゛(AAI)‾1fromEq.(7.11),thentheresidualRinthesubspace1S
equalto
R=(A゛(AAI)‾1)゛R(A'(AA°)‾1)
-165-
(H.4)