proceedings - gbv · michael finke, alex waibel carnegie mellon univ., usa th3a.5 a prosody-only...

15
EUROPEAN SPEECH COMMUNICATION ASSOCIATION CESCA) / 5th EUROPEAN CONFERENCE ON SPEECH COMMUNICATION AND TECHNOLOGY RHODES - GREECE 22l25lSeptembeiiil9971 UNDER THE AUSPICES OF THE MINISTRY OF CULTURE THE MINISTRY OF THE AEGEAN THE GENERAL SECRETARIAT OF SCIENCE AND RESEARCH PROCEEDINGS VOLUME 5 ORGANIZER: UNIVERSITY OF PATRAS WIRE COMMUNICATIONS LABORATORY 261 10 Rion - Patras - Greece

Upload: others

Post on 12-Jun-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

EUROPEAN SPEECH COMMUNICATION ASSOCIATION CESCA)

/

5th EUROPEAN CONFERENCE ON SPEECH COMMUNICATION AND TECHNOLOGY

RHODES - GREECE

22l25lSeptembeiiil9971

UNDER THE AUSPICES OF

THE MINISTRY OF CULTURETHE MINISTRY OF THE AEGEAN

THE GENERAL SECRETARIAT OF SCIENCE AND RESEARCH

PROCEEDINGSVOLUME 5

ORGANIZER:

UNIVERSITY OF PATRASWIRE COMMUNICATIONS LABORATORY261 10 Rion - Patras - Greece

Page 2: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

SESSION: ThMDSpeaker Recognition and Language IdentificationChair: Douglas Reynolds, MiT, USA

ThMD.l Gaussian Mixture Models with Common PrincipalAxes and Their Application in Text-Independent SpeakerIdentification 2279Kuo-Hwei Yuo, Hsiao-Chuan WangNational Tsing Hua Univ., ROChina

ThMD.2 Speaker Models Designed from Complete DataSets: A New Approach to Text-Independent SpeakerVerification 2283Dominik R. Dersch, *Robin W. KingUniv. of Sydney, Australia*Univ. of South Australia, Australia

ThMD.3 A Double Gaussian Mixture Modeling ApproachTo Speaker Recognition 2287Vergin Rivarol, Douglas CShaughnessyINRS Telecommunications, Canada

ThMD.4 An Acoustic Subword Unit Approach to Non-Linguistic Speech Feature Identification 2291Mohamed Afify, Yifan Gong, Jean-Paul HatonCRIN-CNRS, France

ThMD.5 N-Best GMM's for Speaker Identification 2295Chakib Tadj, *Pierre Dumouchel, fYu FangEcole de Technologie Superieure, Canada*Centre de Recherche Informatique, Canadaflnstitut Universitaire de Technologie, Canada

ThMD.6 Model Dependent Spectral Representations forSpeaker Recognition 2299Guillaume Gravier, *Chafic Mokbel, Gerard CholletENST/SIG, France*CNET-DIH/RCP, France

ThMD.7 Equalizing Sub-Band Error Rates in SpeakerRecognition 2303Roland Auckenthaler, "John S. MasonTechnical Univ. Graz, Austria*Univ. of Wales Swansea, UK

ThMD.8 Automatic Gender Identification Under AdverseConditions 2307Stefan Slomka, Sridha SridharanQueensland Univ. of Technology, Australia

ThMD.9 Acoustic Features and Perceptive Processes in theIdentification of Familiar Voices 2311Yizhar Lavner, Isak Gath, Judith RosenhouseIsrael Institute of Technology, Israel

ThMD.10 On the Use Acoustic Segmentation in SpeakerIdentification 2315Leandro Rodriguez-Linares, Carmen Garcia-MateoUniv. ofVigo, Spain

ThMD.ll Speaker Recognition by Humans and MachinesHerman J.M. Steeneken, David A. Van LeeuwenTNO-HFRI, The Netherlands 2319

Page 3: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThMD.12 Foreign Speaker Accent Classification UsingPhoneme-Dependent Accent Discrimination Models andComparisons with Human Perception BenchmarksKarsten Kumpf, *Robin W. KingUniv. of Sydney, Australia*Univ. of South Australia, Australia 2323

ThMD.13 A Comparison of Human and Machine InSpeaker Recognition 2327Li Liu, Jialong He, Giinther PalmUniv. ofUlm, Germany

ThMD.14 Evaluation of Second Language Learners'Pronunciation Using Hidden Markov Models 2331Simo M.A. Goddijn, *Guus de KromForensic Science Laboratory, Rijswijk*Univ. of Utrecht, The Netherlands

ThMD.15 Delta Vector Taylor Series EnvironmentCompensation for Speaker Recognition 2335Brian Eberman, Pedro J. MorenoDigital Equipment Corp., USA

ThMD.16 Wavelet-Like Regression Features in theCepstral Domain for Speaker Recognition 2339Jonathan HumeUniv. of Wales Swansea, UK

ThMD.17 Minimum Classification Error Linear Regression(MCELR) for Speaker Adaptation Using I£MM with TrendFunctions 2343Rathinavelu ChengalvarayanBell Labs-Lucent Technologies, USA

ThMD.18 A Continuous HMM Text Independent SpeakerRecognition System Based on Vowel Spotting 2347Nikos Fakotakis, *Anastasios Tsopanoglou, Kallirroi GeorgilaUniv. ofPatras, GreeceKNOWLEDGE SA, Greece

ThMD.19 On the Independence of Digits in ConnectedDigit Strings 2351Johan W. Koolwaaij, Lou BovesNijmegen University, The Netherlands

ThMD.20 A New Procedure for Classifying Speakers inSpeaker Verification Systems 2355Johan W. Koolwaaij, Lou BovesNijmegen University, The Netherlands

ThMD.21 Sound Channel Video Indexing 2359Claude Montacie, Marie-Jose CaratyUniv. Pierre et Marie Curie - CNRS, France

ThMD.22 CDHMM Speaker Recognition By Means ofFrequency Filtering of Filter-Bank Energies 2363Javier Hernando, Climent NadeuUniversitdt Politecnica de Catalunya, Spain

Page 4: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

SESSION: Tf$3AStyle and Accent RecognitionCftair: Gerard Choitet, ENST/SIG, Switzerland

Th3A.l Using Accent-Specific Pronunciation Modelling forImproved Large Vocabulary Continuous SpeechRecognition 2367J. J. Humphries, P. C. WoodlandCambridge Univ., UK

Th3AJ2 Automatic Speech Recognition for ChildrenAlexandras Potamianos, Shrikanth Narayanan, Sungbok LeeAT&T Labs-Research, USA 2371

Th3A.3 Recognition of Non-Native Accents 2375Carlos Teixeira, Isabel Trancoso, Antonio SerralheiroINESC, Portugal

Th3A.4 Speaking Mode Dependent PronunciationModeling in Large Vocabulary Conversational SpeechRecognition 2379Michael Finke, Alex WaibelCarnegie Mellon Univ., USA

Th3A.5 A Prosody-Only Decision-Tree Model forDisfluency Detection 2383Elizabeth Shriberg, *Rebecca Bates, Andreas StolckeSRI International, USA*Boston Univ., USA

Th3A.6 A Novel Training Approach for Improving SpeechRecognition Under Adverse Stressful Conditions 2387Sahar E. Bou-Ghazale, John H.L HansenDuke Univ., USA

SESSION: Th3BPhoneticsClutir: Joaqttim, Ltistetri, Unw, of Barcelona, Spain

Th3B.l From Phone ^identification to Phone ClusteringUsing Mutual Information 2391Peter CBoyle, Ji Ming, Marie Owens, F.Jack SmithQueen's Univ. of Belfast, N. Ireland

Th3B.2 Phonetic Code Emergence in a Society of SpeechRobots: Explaining Vowel Systems and the MUAFPrinciple 2395Ahmed-Reda Berrah, Rafael LaboissiereInstitut de la Communication Parlee, France

Th3B.3 Effects of Voicing on /t,d7 Tongue/Palate Contact inEnglish and Norwegian 2399Inger Moen, Hanne Gram SimonsenUniv. of Oslo, Norway

Th3B.4 Fieldwork Techniques for Relating FormantFrequency, Amplitude and Bandwidth 2403Peter Ladefoged, *Gunnar FantUCLA, USA*KTH, Sweden

Th3B.5 Word Juncture Modelling Based on the TTMITDatabase 2407Xue Wang, Louis C.W. PolsUniv. of Amsterdam, The Netherlands

Page 5: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

Th3B.6 The Phonology and Phonetics of Second LanguageIntonation: The Case of "Japanese English" 2411Motoko UeyamaUCLA, USA

SESSION: T&3C (SPECIAL SESSION)Toward* Robust ASR for Car and TelephoneApplicationsChair: Jean-Claud Junqua, Panasonic TechnologiesInc., California, USA

Th3C.l Methods for Microphone Equalization in SpeechRecognition 2415L. Fissore, Giorgio Micca, C. VairCentra Studi e Laboratori Telecomunicazioni (CSELT), Italy

Th3C2 Room Acoustics and Reverberation: Impact onHands-Free Recognition 2419Satoshi Nakamura, Kiyohiro ShikanoNara Institute of Science and Technology, Japan

Th3C3 Echo and Noise Reduction for Hands-FreeTerminals -State of the Art- 2423Gerard Faucon, Regine Le Bouquin-JeannesUniv. de Rennes I, France

Th3C4 Robust Speech Recognition for Wireless Networksand Mobile Telephony 2427Reinhold Haeb-UmbachPhilips GmbH, Germany

Th3C.5 Robust ASR for the Cellular EnvironmentJay NaikNynex, USA(Not arrived in time to be included in the Proceedings)

Th3C6 Speech Recognition in the Car From Phone Dialingto Car Navigation 2431Dirk Van CompernolleLernout & Hauspie Speech Products NV, Belgium

: Th3DLanguage Specific SystemsChair: ChmielSvnn, CNET, hmnion, France

Th3D.l A Keyvowel Approach to the Synthesis of RegionalAccents of English 2435Briony Williams, Stephen IsardUniv. of Edinburgh, UK

Th3D.2 Experimental Implementation of Pitch-Synchronous Synthesis Methods for the ROMVOX Text-to-Speech System 2439Attila Ferencz, Radu Arsinte, *Istvan Nagy, Teodora Ratiu,Maria Ferencz, tGavril Toderean, fDiana Zaiu, Tunde-CsillaKovacs, Lujos SimonSoftware ITC SA, Romania*Music Academy Gh.Dima, RomaniatTechnical Univ. of Cluj-Napoca, Romania

Th3D3 The Bell Labs German Text-to-Speech System: AnOverview 2443Bernd Mobius, Richard Sproat,Jan P.H van Santen Joseph POliveBell Labs-Lucent Technologies, USA

Page 6: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

Th3D.4 The Generation of Regional Pronunciations ofEnglish for Speech Synthesis 2447Susan FittUniv. of Edinburgh, UK

Th3D.5 Bell Laboratories Russian Text-To-Speech SystemElena Pavlova, Yuri Pavlov, Richard Sproat, Chilin Shih, JanP.H. van SantenBell Labs-Lucent Technologies, USA 2451

Th3D.6 A Bilingual Text-To-Speech System in Spanish andCatalan 2455Antonio Bonafonte, Ignasi Esquerra, Albert Febrer, FrancescVallverduUniversita't Politecnica de Catalunya, Spain

SESSION: Th4APronunciation ModelsChair; Jean-Paul Haron, OUN/CNRS-lNRtA, France

Th4A.l Automatic Rule-based Generation of WordPronunciation Networks 2459Nick Cremelie, Jean-Pierre MartensUniv. of Gent, Belgium

Th4A.2 Creating User Defined New Vocabularies for VoiceDialing 2463Jose Maria Elvira, Juan Carlos Torrecilla, Javier CamineroTelefonica I+D, Spain

Th4A.3 Automatic Generation of Context-DependentPronunciations 2467Mosur Ravishankar, Maxine EskenaziCarnegie Mellon Univ., USA

Th4A.4 Automatic Generation of a PronunciationDictionary Based on a Pronunciation Network 2471Toshiaki Fukada, Yoshinori SagisakaATR TTL, Japan

Th4A.5 What is Wrong with the Lexicon-An Attempt toModel Pronunciations Probabilistically 2475Uwe Jost, Henrik Heine, Gunnar EvermannHamburg Univ., Germany

Th4A.6 Lexical Tuning Based on Triphone ConfidenceEstimation 2479Kevin L. Markey, *Wayne WardBerdy Medical Systems, USA*Carnegie Mellon Univ., USA

SESSION: Th4BAuditor}1 Modelling and PsychoacousticsChair: William Ainsworth Keele Univ,, UK

Th4B.l Improving of Amplitude Modulation Maps for FO-Dependent Segregation of Harmonic Sounds 2483Frederic Berthommier, *Georg MeyerICP, INPG. France*Univ. of Keele, UK

Th4B2. Psychophysical Evaluation of PSOLA: NaturalVersus Synthetic Speech 2487Reinier Kortekaas, Armin KohlrauschIPO, The Netherlands

Page 7: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

Th4B.3 Perception of Noised Words by Normal Childrenand Children with Speech and Language ImpairmentsValentina V. Lublinskaja, *Inna V. Koroleva, A.N. Kornev,Elena V. Iagounova 2491Pavlov Institute of Physiology, Russiainstitute of Ear, Throat, Nose and Speech Pathology, Russia

Th4B.4 Modeling the Perception of Simultaneous Semi-Vowels 2495Georg F. Meyer, William.A AinsworthKeele Univ., UK

Th4B.5 Properties of Auditory Model RepresentationsFernando Santos Perdigao, Luis V. SaUniversidade de Coimbra, Portugal 2499

Th4B.6 Impact of "Ascending Sequence" AI (AuditoryPrimary Cortex) Cells on Slop Consonant Perception"Marta Eduardo Sa, de Sa Luis VieiraUniversidade de Coimbra, Portugal 2503

SESSION: Th4CVoice Conversion and Data Driven FO-ModeisChair: Yashinori Sagisaka, ATR Interpret. Telecom.Res. Labs., Japan

Th4C.l Application-Dependent Prosodic Models for Text-To-Speech Synthesis and Automatic Design of LearningDatabase Corpus Using Genetic Algorithm 2507Olivier Boeffard.Emerard F.France Telecom-CNET, France

Th4C-2 Combinatorial Issues in Tcxt-To-Speech SynthesisJan P.H. van SantenBell Labs-Lucent Technologies, USA 2511

Th4C-3 Automatic Corpus-Based Training of Rules forProsodic Generation in Text-To-Speech 2515Eduardo Lopez-Gonzalo, Jose M. Rodriguez-Garcia, LuisHernandez-Gomez, Juan M. VillarETS/T-UPM, Spain

Th4C4 Hidden Markov Model Based Voice ConversionUsing Dynamic Characteristics of Speaker 2519Eun-Kyoung Kim, Sangho Lee, Yung-Hwan OhKA1ST, Korea

Th4C5 Speaker Interpolation in HMM-Based SpeechSynthesis System 2523Takayoshi Yoshimura, Takashi Masuko, Keiichi Tokuda,*Takao Kobayashi, Tadashi KilamuraNagoya Institute of Technology, Japan*Tokyo Institute of Technology, Japan

Th4C6 Designing a Speaker Adaptable Formant-BasedText-To-Speech System 2527Vassilios Darsinos, Dimitrios Galanis, George KokkinakisUniv. of Patras, Greece

Page 8: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

SESSION; Th4DVocal Traet AnalysisCluxir: Atdreas Paoloni, Fondayom UgoBordom, Italy

Th4D.l On Using Fractal Features of Speech Sounds inAutomatic Speech Recognition 2531Petros Maragos, *Alexandros PotamianosILSP & Georgia Tech, Greece & USA•AT&T Labs-Research, USA

Th4D.2 Dynamic Constraint Weighting in the Context ofArticulatory Parameter Estimation 2535Hywel B. Richards, John S. Bridle, Melvyn J. Hunt, *John S.MasonDragon Systems, UK*Univ. of Wales Swansea, UK

Th4DJ Estimation of Vocal Tract Front Cavity Resonancein Unvoiced Fricative Speech 2539Minkyu Lee, Donald G. ChildersUniv. of Florida, USA

Th4D.4 A Software Tool to Study Portuguese VowelsAntonio Teixeira, Francisco Vaz, *Jose Carlos PrincipeINESC, Portugal*Univ. of Florida, USA 2543

Th4D.5 Post-Synchronization Via Formant-to-AreaMapping of Asynchronously Recorded Speech Signals andArea Functions 2547Jean Schoentgen, Sorin CioceaUniv. Libre de Bruxelles, Belgium

Th4D.6 Geometrically and Acoustically OptimizedCodebook for Unique Mapping from Formants to Vocal-Tract Shape 2551Zhenli L. Yu, P.C. ChingThe Chinese Univ. of Hong Kong, Hong Kong

ShSSIOV: ThAANoise Mitigation, Spew-li KnliiincvmcnL IIChair: Hayva Irgniinarayarui. II'I M.WK.\S. India

ThAA.l Noisy Speech Enhancement by Fusion of Auditoryand Visual Information: A Study of VowelTransitions 2555Laurent Girin, Gang Feng, Jean-Luc SchwartzUniv. of Stendhal, France

ThAA.2 Spectral Subtraction Using a Non-CriticallyDecimated Discrete Wavelet Transform 2559Andreas Engelberg, Thomas GulzowUniv. of Kiel, Germany

ThAA3 Bayesian Affine Trasformation of HMMParameters for Instantaneous and Supervised Adaptationin Telephone Speech Recognition 2563Jen-Tzung Chien, Hsiao-Chuan Wang, *Chin-Hui LeeNational Tsing Hua Univ., ROChina*BellLabs, USA

ThAA.4 Integrated Bias Removal Techniques for RobustSpeech Recognition 2567Craig Lawrence, Mazin RahimUniv. of Maryland, USA

Page 9: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAA.5 Acoustic Front Ends for Speaker-IndependentDigit Recognition in Car Environments 2571Detlev Langmann, Alexander Fischer, Friedhelm Wuppermann,Reinhold Haeb-Umbach, Thomas EiselePhilips GmbH, Germany

ThAA.6 Signal Bias Removal Using the Multi-PathStochastic Equalization Technique 2575Lionel Delphin-Poulat, Chafic MokbelFrance Telecom, France

ThAA.7 Subband Echo Cancellation in Automatic SpeechDialog Systems 2579Andrej Miksic, Bogomir HorvatUniv. ofMaribor, Slovenia

ThAA.8 Speech Enhancencement Via Energy SeparationHeshamTolba, Douglas OShaughnessyUniv. du Quebec, Canada 2583

ThAA.9 A Method of Signal Extraction from Noisy SignalMasashi Unoki, Masato Akagi 2587Japan Advanced Institute of Science and Technology, Japan

ThAA.10 Multi-Channel Noise Reduction Using WaveletFilter Bunk 2591Jiri Sika, Vratislav DavidekCzech Technical Univ., Czech Republic

ThAA.ll Speech Signal Detection in Noisy EnvironmentUsing a Local Entropic Criterion 2595Imad Abdallah, Silvio Montresor, Marc BaudryLaboratoire d'Informatique de IVniv. du Maine, France

ThAA.12 A New Algorithm for Robust Speech Recognition:The Delta Vector Taylor Series ApproachPedro J. Moreno, Brian EbermanDigital Equipment Corp., USA 2599

ThAA.13 Robust Enhancement of Reverberant SpeechUsing Iterative Noise Removal 2603David Cole, Miles Moody, Sridha SridharanQueensland Univ. of Technology, Australia

ThAA.14 A Network Speech Echo Canceller with ComfortNoise 2607David J. Jones, Scott D. Watson, Kenneth G. Evans, BarryM.G. Cheetham, *Robert A. ReevesUniv. of Liverpool, UK*BT Laboratories, UK

ThAA.15 A Metric for Selecting Sub-Band Processing inAdaptive Speech Enhancement Systems 2611Amir Hussain, Douglas R. Campbell, Thomas J. MoirUniv. of Paisley, UK

ThAA.16 Estimation of LPC Cepstrum Vector of SpeechContaminated by Additive Noise and its Application toSpeech Enhancement 2615Hidefumi Kobatake, Hideta SuzukiTokyo Univ. of Agriculture & Technology, Japan

ThAA.17 Multi-Band and Adaptation Approaches toRobust Speech Recognition 2619Sangita Tibrewala, Hynek HermanskyOregon Graduate Institute of Science and Technology, USA

Page 10: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAA.18 Non-Quadratic Criterion Algorithms for SpeechEnhancement 2623Enrique Masgrau, Eduardo Lleida, Luis VicenteUniversidad de Zaragoza, Spain

SESSION: ThABFO and Duration Modelling, Spoken languageprocessingCluiir: Richard Schwartz, BBN Systems andTeclis, USA

ThAB.l Modeling Segmental Duration with MultivariateAdaptive Regression Splines 2627Marcel RiediETH Zentrum TIK, Switzerland

ThAB.2 High Quality Speech Synthesis for Phonetic SpeechSegmentation 2631Fabrice Malfrere, Thierry DutoitCircuits Theory and Signal Processing Lab, Belgium

ThAB 3 Factors Affecting Perceived Quality andIntelligibility in the CHATR Concatenative SpeechSynthesiser 2635Nick Campbell, Itoh Yoshiharu, Wen Ding, Norio HiguchiATR Interpreting Telecommunications Res. Labs., Japan

ThAB .4 Reduced Lexicon Trees for Decoding in a MMI-Connectionist/HMM Speech Recognition System 2639Christoph Neukirchen, Daniel Willett, Gerhard RigollGerhard-Mercator-Univ. Duisburg, Germany

ThAB.5 A Stochastic Model of Intonation for French Text-to-Speech Synthesis 2643Jean Veronis, Philippe Di Cristo, Fabienne Courtois, BenoitLagrueUniv. de Provence & CNRS, France

ThAB.6 Phonetic Rules for a Phonetic-to-Speech SystemAngelien A. Sanderman, *Rene Collier 2647KPN Research, The Netherlandsinstitute for Perception Research, The Netherlands

ThAB.7 Multi-Lingual Duration Modeling 2651Jan P.H van Santen, Chilin Shih, Bernd Mobius, EvelyneTzoukermann, Michael TanenblattBell Labs-Lucent Technologies, USA

ThAB.8 A Model of Segment (and Pause) DurationGeneration for Brazilian Portuguese Text-to-SpeechSynthesis 2655Plinio A. BarbosaState Univ. of Campinas, Brazil

ThAB.9 Parsing Strategy for Spoken Language Interfaceswith a Lexicalized Tree Grammar 2659Ariane Halber, David RousselThomson-CSF, France

ThAB.10 What's in a Word Graph - Evaluation andEnhancement of Word Lattices 2663Jan W. Amtrup, Henrik Heine, Uwe JostHamburg Univ., Germany

Page 11: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAB.ll Accelerated DP Based Search for StatisticalTranslation 2667Christoph Tillmann, Stefan Vogel, Hermann Ney, A. Zubiaga,H. SawafRWTH Aachen, Germany

ThAB.12 Use of Pitch Pattern Improvement in the CHATRSpeech Synthesis System 2671Ken Fujisawa, Toshio Hirai, Norio HiguchiATRITL, Japan

ThAB.13 Generating Segment Durations in a Text-to-Speech System: A Hybrid Rule-Based/Neural NetworkApproach 2675Gerald Corrigan, Noel Massey, Orhan KaraaliMotorola, USA

ThAB.14 On the Global FO Shape Model Using aTransition Network for Japanese Text-To-Speech SystemsYasushi Ishikawa, Takashi EbiharaMitsubishi Electric Corporation, Japan 2679

ThAB.15 An Alternative and Flexible Approach in RobustInformation Retrieval Systems 2683Jose Colas, Juan M. Montero, Javier Ferreiros, Jose M. PardoUniversidad Politecnica de Madrid, Spain

ThAB.16 A Probalistic Approach to Analogical SpeechTranslation 2687Keiko Horiguchi, Alexander FranzSony, Japan

ThAB.17 Dynamic Lexicon for a Very Large VocabularyVocal Dictation 2691Marie-Jose Caraty, Claude Montacie, Fabrice LefevreUniv. Pierre et Marie Curie - CNRS, France

SESSION: ThACLanguage ModellingOiair: Ronald Rosenfeld, Carnegie Mellon Univ., USA

ThAC.1 Construction of Language Models Using theMorphic Generator Grammatical Inference (MGGI)Methodology 2695Encama Segarra, Luis HurtadoUniversidad Politecnica de Valencia, Spain

ThAC.2 An Integrated Language Modeling with N-GramModel and WA Model for Speech Recognition 2699Shuwu Zhang, Taiyi HuangChinese Academy of Sciences, China

ThAC3 Statistical Analysis of Dialogue Structure 2703Ye-Yi Wang, Alex WaibelCarnegie Mellon Univ., USA

ThAC.4 Statistical Language Modeling Using the CMU-Cambridge Toolkit 2707Philip Clarkson, *Ronald RosenfeldCambridge Univ., UK*Camegie Mellon Univ., USA

ThAC.5 Text Normalization and Speech Recognition inFrench 2711Gilles Adda, Martine Adda-Decker, Jean-Luc Gauvain, LoriLamelLJMSI, France

Page 12: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAC.6 A Novel Tree-Based Clustering Algorithm forStatistical Language Modeling 2715Geraldine Damnati, Jacques SimoninFrance Telecom, France

ThAC.7 Variable-Length Language Modeling IntegratingGlobal Constraints 2719Shoichi Matsunaga, Shigeki SagayamaNTT, Japan

ThAC.8 An Hybrid Language Model for ContinuousDictation Prototype 2723Kamel Smaili, Imed Zitouni, Francois Charpillet, Jean-PaulHatonCRIN-CNRS & INRIA, Lorraine, France

ThAC.9 Dealing with Pronunciation Variants at theLanguage Model Level for the Continuous AutomaticSpeech Recognition of French 2727Laure Pousse, Guy PerennouIRIT-Equipe IHMPT, France

ThAC.10 Rational Interpolation of Maximum LikelihoodPredictors in Stochastic Language Modeling 2731Ernst Gunter Schukat-Talamazzini, *Florian Gallwitz, *StefanHarbeck, *Volker WarnkeUniv. of Jena, Germany*Univ. of Erlangen, Germany

ThAC.ll N-Gram Language Model Adaptation UsingSmall Corpus for Spoken Dialog Recognition 2735Akinori Ito, Hideyuki Saitoh, Masaharu Katoh, Masaki KohdaYamagata Univ., Japan

ThAC.12 Variable N-Gram Language Modeling andExtensions for Conversational Speech 2739Man-Hung Siu, *Mari OstendorfBBNInc, USA*Boston Univ., USAThAC.13 Fuzzy Class Rescoring: A Part-of-SpeechLanguage Model 2743Petra GeutnerUniv. of Karlsruhe, Germany

ThAC.14 Speech Understanding Based on IntegratingConcepts By Conceptual Dependency 2747Akito Nagai, Yasushi IshikawaMitsubishi Electric Corporation, Japan

ThAC.15 Dynamic Language Models for InteractiveSpeech Applications 2751Fabio Brugnara, Marcello FedericoIstitutoper la Ricerca Scientifica e Tecnologica (IRST), Italy

ThAC.16 Large-Scale Lexical Semantics for SpeechRecognition Support 2755George Demetriou, Eric Atwell, Clive SouterUniv. of Leeds, UK

ThAC.17 Integration of Grammar and Statistical LanguageConstraints for Partial Word-Sequence Recognition....2759Hajime Tsukada, Hirofumi Yamamoto, Yoshinori SagisakaATR Interpreting Telecommunications Res. Labs., Japan

Page 13: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAC.18 Using Intonation to Constrain Language Modelsin Speech Recognition 2763Paul Taylor, Simon King, Stephen bard, Helen Wright,Jaqueline KowtkoUniv. of Edinburgh, UK

ThAC.19 Incorporating POS Tagging into LanguageModeling 2767Peter A. Heeman, *James F. AllenFrance Telecom, France*Univ. of Rochester, USA

ThAC.20 Confidence Metrics Based on N-Gram LanguageModel Backoff Behaviors 2771Carl Uhrik, *Wayne WardBerdy Medical Systems, USA* Carnegie Mellon Univ., USA

ThAC.21 Structure and Performance of a DependencyLanguage Model ...2775Ciprian Chelba, *David Engle, Frederick Jelinek, fVictor M.Jimenez, Sanjeev Khudanpur, Iidia Mangu, nHarry Printz,**Eric Ristad, ttRonald Rosenfeld, ^Andreas Stolcke,BHDekai WuJohn Hopkins Univ., USA*Dept.of Defense Fort Meade,MD, USAfUniversitdt Politecnica de Valencia, SpainUIBM, USA**Princeton Univ., USAffCamegie Mellon Pittsburgh.PA, USAttSRlInternational, USAVOHong Kong Tech University, Hong Kong

ThAC.22 Modeling Linguistic Segment and TurnBoundaries for N-Best Rescoring of Spontaneous SpeechAndreas StolckeSRI International, USA 2779

ThAC.23 Hybrid Language Models: Is Simpler better?Peter E. Kenne, Mary OTCaneUniv. of Adelaide, Australia 2783

ThAC.24 Internal and External Tagsets in Part-of-SpeechTagging 2787Thorsten BrantsUniv. of the Saarland, Germany

SESSION: ThADAuditory Modelling and Psychoacoustics,NeuralNetworks for Speech Processing and RecognitionChair: Phil D. Green, Unm of Sheffield, UK

ThAD.l A Probabilistic Model of Double-VowelSegregation 2791Laurent Varin, Frederic BerthommierICP, INPG, France

ThAD.2 Stimulus Signal Estimation From Auditory-NeuralTransduction Inverse Processing 2795Houshang Habibzadeh Vaneghi, Shigeyoshi KitazawaShizuolca Univ., Japan

Page 14: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

*Institut Universitaire Professionnalise, France

ThAD.4 The Initial Time Span of Auditory Processing Usedfor Speaker Attribution of the Speech Signal 2803Valentina V. Lublinskaja, "Christian SappokPavlov Institute of Physiology, Russia*Ruhr Universitat, Germany

ThAD.5 Sparse Connection and Pruning in Large DynamicArtificial Neural Networks 2807Nikko StromKTH, Sweden

ThAD.6 A Modular Initialization Scheme for Better SpeechRecognition Performance Using Hybrid Systems ofMLPsVHMMs 2811Roxana Teodorescu.Dirk Van Compernolle, Ioannis DologlouK.U Leuven-ESAT, Belgium

ThAD.7 Lateralization for Auditory Perception of ForeignWords 2815Tatiana ChernigovskayaRussian Academy of Sciences, Russia

ThAD.8 The Structural Weighted Sets Method forContinuous Speech and Text Recognition 2819Yuri Kosarev, Pavel Jarov, Alexander OsipovRussian Academy of Sciences, Russia

ThAD.9 Lateral Inhibitory Networks for AuditoryProcessing 2823Christian J. Sumner, Duncan F. GilliesImperial College, UK

ThAD.lO Missing Fundamentals:A Problem of Auditory orMental Processing? 2827Henning ReetzUniv. ofKonstanz, Germany

Page 15: PROCEEDINGS - GBV · Michael Finke, Alex Waibel Carnegie Mellon Univ., USA Th3A.5 A Prosody-Only Decision-Tree Model for Disfluency Detection 2383 Elizabeth Shriberg, *Rebecca Bates,

ThAD.12 Empirical Comparison of Two MultilayerPerccptron-Based Keyword Speech RecognitionAlgorithms 2835Suhardi, 'Klaus FellbaumTechnical Univ. of Berlin, Germany*Brandenburg Technical Univ. ofCottbus, Germany

ThAD.13 Segment Boundary Estimation Using RecurrentNeural Networks 2839Toshiaki Fukada, 'Sophie Avelinc, Mike Schuster, YoshinoriSagisakaATR Interpreting Telecommunications Res. luibs.,*ENST, France

ThAD.14 Incorporation of IIMM Output Constraints inHybrid NN/IIMM Systems During Training 2843Mike SchusterATR LTL Japan

ThAD.15 Principles of the Hearing Periphery Fuctioning inNew Methods of Pitch Detection and SpeechEnhancement 2847Ludmila Babkina, 'Sergey Koval, Alexander MolchanovResearch Institute ofEar,Nose, Ihroat and Speech Disorders,Russia*Speech Technology Centre, Russia

ThAD.16 The Locus or the Syllable Effect: Prclexical orLexical? 2851Christine Meunier, *Alain Content, Uli H. Frauenfeldcr, fRuthKeamsUniv. of Geneva, Switzerland*Univ. Libre de Bruxelles, BelgiumtMedical Research Council, UK

ThAD.17 On Not Remembering Disflucncies 2855Ellen Gurman Bard, Robin J. LickleyUniv. of Edinburgh, UK

ThAD.18 Using an Auditory Model and LeakyAutocorrelators to Tune In to Speech 2859Tjeerd AndringaUniv. ofGroningen, The Netherlands