Speech Processing in Mobile Environments

129 Pages·2014·4.726 MB·English

by  K. Sreenivasa Rao, Anil Kumar Vuppala (auth.)

SPRINGER BRIEFS IN ELECTRICAL AND
COMPUTER ENGINEERING  SPEECH TECHNOLOGY
K. Sreenivasa Rao
Anil Kumar Vuppala
Speech
Processing
in Mobile
Environments
123
SpringerBriefs in Electrical and Computer
Engineering
SpringerBriefs in Speech Technology
SeriesEditor
AmyNeustein
Forfurthervolumes:
http://www.springer.com/series/10043
Editor’sNote
The authors of this series have been hand selected. They comprise some of the
most outstanding scientists—drawn from academia and private industry—whose
researchismarkedbyitsnovelty,applicability,andpracticalityinprovidingbroad-
basedspeechsolutions.TheSpringerBriefsinSpeechTechnologyseriesprovides
the latest findings in speech technology gleaned from comprehensive literature
reviewsandempiricalinvestigationsthatareperformedinbothlaboratoryandreal
life settings. Some of the topics covered in this series include the presentation of
reallifecommercialdeploymentofspokendialogsystems,contemporarymethods
of speech parameterization, developments in information security for automated
speech, forensic speaker recognition, use of sophisticated speech analytics in call
centers,andanexplorationofnewmethodsofsoftcomputingforimprovinghuman-
computerinteraction.Thoseinacademia,theprivatesector,theselfserviceindustry,
lawenforcement,andgovernmentintelligenceareamongtheprincipalaudiencefor
thisseries,whichisdesignedtoserveasanimportantandessentialreferenceguide
for speech developers, system designers, speech engineers, linguists, and others.
In particular, a major audience of readers will consist of researchers and technical
experts in the automated call center industry where speech processing is a key
componenttothefunctioningofcustomercarecontactcenters.
Amy Neustein, Ph.D., serves as editor in chief of the International Journal of
Speech Technology (Springer). She edited the recently published book Advances
in Speech Recognition: Mobile Environments, Call Centers and Clinics (Springer
2010), and serves as quest columnist on speech processing for Womensenews.
Dr. Neustein is the founder and CEO of Linguistic Technology Systems, a NJ-
basedthinktankforintelligentdesignofadvancednaturallanguage-basedemotion
detectionsoftwaretoimprovehumanresponseinmonitoringrecordedconversations
ofterrorsuspectsandhelplinecalls.
Dr. Neustein’s work appears in the peer review literature and in industry and
mass media publications. Her academic books, which cover a range of political,
social, and legal topics, have been cited in the Chronicles of Higher Education
and have won her a pro Humanitate Literary Award. She serves on the visiting
facultyoftheNationalJudicialCollegeandasaplenaryspeakeratconferencesin
artificial intelligence and computing. Dr. Neustein is a member of MIR (machine
intelligence research) Labs, which does advanced work in computer technology
to assist underdeveloped countries in improving their ability to cope with famine,
disease/illness, and political and social affliction. She is a founding member of
the New York City Speech Processing Consortium, a newly formed group of NY-
basedcompanies,publishinghouses,andresearchersdedicatedtoadvancingspeech
technologyresearchanddevelopment.
K. Sreenivasa Rao (cid:129) Anil Kumar Vuppala
Speech Processing in Mobile
Environments
123
K.SreenivasaRao AnilKumarVuppala
IndianInstituteofTechnology InternationalInstituteofInformation
Kharagpur,WestBengal,India Technology
Hyderabad,Gachibowli,India
ISSN2191-8112 ISSN2191-8120(electronic)
ISBN978-3-319-03115-6 ISBN978-3-319-03116-3(eBook)
DOI10.1007/978-3-319-03116-3
SpringerChamHeidelbergNewYorkDordrechtLondon
LibraryofCongressControlNumber:2013956607
©SpringerInternationalPublishingSwitzerland2014
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,
broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology
nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection
with reviews or scholarly analysis or material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work. Duplication of
this publication or parts thereof is permitted only under the provisions of the Copyright Law of the
Publisher’slocation,initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.
PermissionsforusemaybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violations
areliabletoprosecutionundertherespectiveCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
While the advice and information in this book are believed to be true and accurate at the date of
publication,neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityfor
anyerrorsoromissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,with
respecttothematerialcontainedherein.
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Preface
Robust speech systems in mobile environment have gained a special interest in
recent years in order to enable access to remote voice-activated services. In this
context,threemajorchallengesthatneedtobeconsideredare:varyingbackground
conditions,speechcoding,andtransmissionchannelerrors.Inthisbook,wefocus
on improving the recognition performance of speech systems in the presence of
speechcodingandbackgroundnoiseconditionsbyusingvowelonsetpoints(VOPs)
asanchorpoints.VOPisanimportanteventinspeechproduction,anditisdefined
astheinstantatwhichtheonsetofvoweltakesplace.Speechcodersconsideredin
this work are GSM full rate (ETSI 06.10), GSM enhanced full rate (ETSI 06.60),
CELP(FS-1016),andMELP(TI2.4kbps).
Themajorworkspresentedinthisbookare:
(cid:129) MethodsareproposedforthedetectionofVOPsinthepresenceofspeechcoding
andbackgroundnoiseconditions.
(cid:129) A two-stage hybrid approach based on hidden Markov models (HMMs) and
supportvectormachines(SVMs)isproposedforimprovingtheperformanceof
consonant-vowel(CV)recognitionsystem.
(cid:129) Two-stage VOP detection method is proposed for spotting CV units from
continuousspeech.
(cid:129) Combinedtemporalandspectralpreprocessingmethodsareexploredtoimprove
theperformanceofCVrecognitionsystemunderbackgroundnoise.
(cid:129) A method based on VOPs is proposed to improve the performance of speaker
identification(SI)systeminthepresenceofcoding.
(cid:129) A method is proposed for nonuniform time scale modification using VOPs and
instantsofsignificantexcitation.
Someimportantconclusionsdrawnoutofthisworkare:(i)Performanceofthe
proposed VOP detection method based on spectral energy in the glottal closure
region is found to be better compared to existing methods under clean, coded and
noisyconditions.(ii)Performanceoftheproposedtwo-stagehybridCVrecognition
approachhasshownsignificantimprovementcomparedtootherapproaches,under
clean, coded, and noisy conditions. (iii) Performance of CV recognition system
v
vi Preface
under background noise is improved by using combined temporal and spectral
processing-based preprocessing method. (iv) Proposed two-stage VOP detection
method used for spotting CV segments from continuous speech has found to be
efficient in minimizing the missing and spurious VOPs. (v) In the presence of
coding, performance of SI system is improved by using features extracted from
steady vowel speech segments. Improvement in SI system performance is mainly
due to the presence of crucial speaker-specific information in the steady vowel
segments of speech, even after coding. (vi) Performance of the proposed time
scalemodificationmethodissuperiorcomparedtoexistingmethods.Thesuperior
performance of the proposed method is due to the nonuniform modification of
different speech segments and accurate detection of various speech segments with
thehelpofinstantsofsignificantexcitationandVOPs.
Thisbookismainlyintendedforresearchersworkingonbuildingrobustspeech
systemsinmobileenvironment.Thisbookisalsousefulfortheyoungresearchers,
who want to pursue the research in speech processing. Hence, this may be
recommended as the text or reference book for the postgraduate level advanced
speechprocessingcourse.
Many people have helped us during the course of preparation of this book. We
would especially like to thank all professors of G.S. Sanyal School of Telecom-
munication and School of Information Technology, IIT Kharagpur for their moral
encouragement and technical discussions during the course of editing and orga-
nization of the book. Special thanks to our colleagues at IIT Kharagpur for their
cooperationandcoordinationtocarryoutthework.Finally,wethankallourfriends
andwell-wishers.
WestBengal,India K.SreenivasaRao
Hyderabad,India AnilKumarVuppala
Contents
1 Introduction .................................................................. 1
1.1 Introduction............................................................. 1
1.2 ObjectiveoftheBook .................................................. 2
1.3 OrganizationoftheBook .............................................. 2
2 BackgroundandLiteratureReview ....................................... 5
2.1 ApproachesforDetectionofVowelOnsetPoints..................... 5
2.1.1 VOPDetectionUsingExcitationSource,Spectral
Peaks,ModulationSpectrum,andTheirCombination....... 6
2.2 SpeechProcessinginMobileEnvironment ........................... 9
2.2.1 SpeechandSpeakerRecognitionUnderCoding ............. 10
2.2.2 SpeechRecognitionUnderBackgroundNoise............... 11
2.3 RecognitionofCVUnitsofSpeechinIndianLanguages............ 12
2.4 TimeScaleModification ............................................... 13
2.5 Summary................................................................ 14
3 VowelOnsetPointDetectionfromCodedandNoisySpeech ........... 17
3.1 SpeechDatabasesforVOPDetection ................................. 17
3.1.1 TIMITDatabase................................................ 18
3.1.2 BroadcastNewsDatabase...................................... 18
3.2 VOPDetectionMethodforCodedSpeech............................ 18
3.2.1 ExtractionofGlottalClosureInstantsUsing
ZFFMethod .................................................... 20
3.2.2 SequenceofStepsintheProposedVOP
DetectionMethod .............................................. 21
3.2.3 ChoiceofFrameSize .......................................... 24
3.2.4 ChoiceofFrequencyBand..................................... 27
3.3 Performance of the VOP Detection Method
inthePresenceofSpeechCoding ..................................... 30
3.3.1 VOPDetectionfromContinuousSpeechUnderCoding..... 30
3.3.2 VOPDetectionfromCVUnitsUnderCoding................ 32
vii
viii Contents
3.4 VOPDetectionMethodforNoisySpeech............................. 35
3.4.1 FormantExtractionUsingGroupDelayFunction............ 35
3.4.2 Sequence of Steps in the Proposed VOP
DetectionMethodforNoisySpeech .......................... 37
3.5 Performance of the VOP Detection Method
inthePresenceofBackgroundNoise.................................. 37
3.5.1 VOPDetectionfromContinuousSpeechUnderNoise ...... 38
3.5.2 VOPDetectionfromCVUnitsUnderNoise ................. 39
3.6 Summary................................................................ 41
4 Consonant–VowelRecognitioninthePresenceofCoding
andBackgroundNoise ...................................................... 43
4.1 Consonant–VowelUnitDatabases..................................... 44
4.2 Two-StageCVRecognitionSystem................................... 45
4.2.1 MotivationsfortheProposedCVRecognitionApproach.... 45
4.2.2 ProposedCVRecognitionApproach.......................... 46
4.2.3 Framework...................................................... 47
4.2.4 PerformanceoftheCVRecognitionSystem.................. 47
4.3 ImpactofAccuracyinVOPDetectiononCVRecognition.......... 50
4.4 PerformanceofCVRecognitionSystemUnderCoding ............. 51
4.4.1 IsolatedCVUnitsRecognitionUnderCoding................ 51
4.4.2 CVUnitsRecognitionfromContinuousSpeech
inthePresenceofCoding...................................... 53
4.5 PerformanceofCVRecognitionSysteminthePresence
ofBackgroundNoise................................................... 54
4.6 Application of Combined Temporal and Spectral
ProcessingMethodsforCVUnitsRecognitionUnder
BackgroundNoise...................................................... 56
4.6.1 CombinedTSPMethodforEnhancement
ofNoisySpeech................................................ 56
4.6.2 CV Units Recognition Under Different
Background Noise Cases Using Temporal
andSpectralPreprocessingTechniques....................... 62
4.7 Summary................................................................ 63
5 Spotting and Recognition of Consonant–Vowel Units
fromContinuousSpeech.................................................... 65
5.1 Two-StageApproachforDetectionofVowelOnsetPoints .......... 66
5.1.1 SequenceofStepsintheProposedVOP
DetectionMethod .............................................. 66
5.1.2 ChoiceofDeviationThresholdforDetermining
theUniformityintheEpochIntervals......................... 67
5.1.3 Performance of the Proposed Two-Stage
VOPDetectionMethod........................................ 69
5.2 PerformanceofSpottingandRecognitionofCVUnits
inContinuousSpeech .................................................. 70
Contents ix
5.3 SpottingandRecognitionofCVUnitsfromCodedSpeech ......... 73
5.4 SpottingandRecognitionofCVUnitsfromNoisySpeech.......... 74
5.5 Summary................................................................ 75
6 SpeakerIdentificationandTimeScaleModificationUsing
VOPs .......................................................................... 77
6.1 SpeakerIdentificationinthePresenceofCodingUsing
VowelOnsetPoints..................................................... 78
6.1.1 SpeechDatabases .............................................. 78
6.1.2 SISystemUsingAANNModels.............................. 78
6.1.3 EffectofSpeechCodingonSpeakerIdentification........... 80
6.1.4 ProposedSpeakerIdentificationMethod...................... 82
6.1.5 PerformanceoftheSpeakerIdentificationSystem
UsingFeaturesExtractedfromSteadyVowelRegions....... 84
6.2 NonuniformTimeScaleModificationUsingInstants
ofSignificantExcitationandVowelOnsetPoints .................... 84
6.2.1 DurationAnalysisofVowelsinFastandSlowSpeech....... 85
6.2.2 DeterminationofDifferentSpeechSegments ................ 87
6.2.3 ProposedMethodforTimeScaleModification............... 88
6.2.4 EvaluationoftheProposedNonuniformTime
ScaleModificationMethod.................................... 92
6.3 Summary................................................................ 93
7 SummaryandConclusions ................................................. 97
7.1 SummaryofthePresentWork ......................................... 97
7.2 ContributionsofthePresentWork..................................... 100
7.3 DirectionsforFutureWork ............................................ 100
AppendixA MFCCFeatures.................................................. 103
AppendixB SpeechCoders ................................................... 107
B.1 GlobalSystemforMobileFullRateCoder(ETSIGSM06.10) ..... 107
B.2 GSMEnhancedFullRateCoder(ETSIGSM06.60)................. 107
B.3 CodebookExcitedLinearPrediction(CELPFS-1016)............... 107
B.4 MixedExcitedLinearPrediction(MELPTI2.4kbps) ............... 108
B.5 DegradationMeasures.................................................. 109
AppendixC PatternRecognitionModels.................................... 111
C.1 HiddenMarkovModels................................................ 111
C.2 SupportVectorMachines............................................... 112
C.3 Auto-AssociativeNeuralNetworkModels............................ 113
References......................................................................... 115