Nonlinear Analyses and Algorithms for Speech Processing

225 Pages·2016·2.95 MB·English

by  Drugman T.

Thomas Drugman
Thierry Dutoit (Eds.)
1
Advances in Nonlinear
1
9
7

I
A Speech Processing
N
L
6th International Conference, NOLISP 2013
Mons, Belgium, June 2013
Proceedings
123
Lecture Notes in Artificial Intelligence 7911
Subseries of Lecture Notes in Computer Science
LNAISeriesEditors
RandyGoebel
UniversityofAlberta,Edmonton,Canada
YuzuruTanaka
HokkaidoUniversity,Sapporo,Japan
WolfgangWahlster
DFKIandSaarlandUniversity,Saarbrücken,Germany
LNAIFoundingSeriesEditor
JoergSiekmann
DFKIandSaarlandUniversity,Saarbrücken,Germany
Thomas Drugman Thierry Dutoit (Eds.)
Advances in Nonlinear
Speech Processing
6th International Conference, NOLISP 2013
Mons, Belgium, June 19-21, 2013
Proceedings
1 3
VolumeEditors
ThomasDrugman
ThierryDutoit
UniversityofMons,TCTSLab
31,BouldevardDolez,7000Mons,Belgium
E-mail:{thomas.drugman,thierry.dutoit}@umons.ac.be
ISSN0302-9743 e-ISSN1611-3349
ISBN978-3-642-38846-0 e-ISBN978-3-642-38847-7
DOI10.1007/978-3-642-38847-7
SpringerHeidelbergDordrechtLondonNewYork
LibraryofCongressControlNumber:2013939661
CRSubjectClassification(1998):I.2.7,I.5,H.5,I.6,G.1
LNCSSublibrary:SL7–ArtificialIntelligence
©Springer-VerlagBerlinHeidelberg2013
Thisworkissubjecttocopyright.AllrightsarereservedbythePublisher,whetherthewholeorpartof
thematerialisconcerned,specificallytherightsoftranslation,reprinting,reuseofillustrations,recitation,
broadcasting,reproductiononmicrofilmsorinanyotherphysicalway,andtransmissionorinformation
storageandretrieval,electronicadaptation,computersoftware,orbysimilarordissimilarmethodology
nowknownorhereafterdeveloped.Exemptedfromthislegalreservationarebriefexcerptsinconnection
withreviewsorscholarlyanalysisormaterialsuppliedspecificallyforthepurposeofbeingenteredand
executedonacomputersystem,forexclusiveusebythepurchaserofthework.Duplicationofthispublication
orpartsthereofispermittedonlyundertheprovisionsoftheCopyrightLawofthePublisher’slocation,
initscurrentversion,andpermissionforusemustalwaysbeobtainedfromSpringer.Permissionsforuse
maybeobtainedthroughRightsLinkattheCopyrightClearanceCenter.Violationsareliabletoprosecution
undertherespectiveCopyrightLaw.
Theuseofgeneraldescriptivenames,registerednames,trademarks,servicemarks,etc.inthispublication
doesnotimply,evenintheabsenceofaspecificstatement,thatsuchnamesareexemptfromtherelevant
protectivelawsandregulationsandthereforefreeforgeneraluse.
Whiletheadviceandinformationinthisbookarebelievedtobetrueandaccurateatthedateofpublication,
neithertheauthorsnortheeditorsnorthepublishercanacceptanylegalresponsibilityforanyerrorsor
omissionsthatmaybemade.Thepublishermakesnowarranty,expressorimplied,withrespecttothe
materialcontainedherein.
Typesetting:Camera-readybyauthor,dataconversionbyScientificPublishingServices,Chennai,India
Printedonacid-freepaper
SpringerispartofSpringerScience+BusinessMedia(www.springer.com)
Preface
NOLISP, an ISCA tutorial and workshop on non-linear speech processing, is a
biannual event whose aim is to present and discuss new ideas, techniques, and
results related to alternative approaches in speech processing that may depart
from the mainstream. In order to work at the front-end of the subject area, the
following domains of interest have been defined for the NOLISP workshops:
1. Non-linear approximation and estimation
2. Non-linear oscillators and predictors
3. Higher-order statistics
4. Independent component analysis
5. Nearest neighbors
6. Neural networks
7. Decision trees
8. Non-parametric models
9. Dynamics for non-linear systems
10. Fractal methods
11. Chaos modeling
12. Non-linear differential equations
The initiative of organization of NOLISP 2013 at the University of Mons
(UMONS) came from the speech processing research group at TCTS Lab. The
factthatthiswasthe sixtheditionofNOLISPgivesevidencethatthe workshop
has already become an established international event.
TheOrganizingCommitteewouldliketosincerelythankoursponsorsAcapela
Group and Nuance, as well as FNRS, University of Mons, and ISCA for their
financial support.
April 2013 Thomas Drugman
Thierry Dutoit
Sponsors
Table of Contents
Speech and Audio Analysis
Evaluation of Automatic Glottal Source Analysis .................... 1
John Kane and Christer Gobl
NMF-Based Spectral Analysis for Acoustic Event Classification
Tasks........................................................... 9
Jimmy Luden˜a-Choez and Ascensi´on Gallardo-Antol´ın
Efficient GCI Detection for Efficient Sparse Linear Prediction.......... 17
Vahid Khanagha and Khalid Daoudi
Gender Detection in Running Speech from Glottal and Vocal Tract
Correlates....................................................... 25
Cristina Mun˜oz-Mulas, Rafael Mart´ınez-Olalla, Pedro G´omez-Vilda,
Agust´ın A´lvarez-Marquina, and Luis Miguel Mazaira-Ferna´ndez
An Efficient Method for Fundamental Frequency Determination of
Noisy Speech .................................................... 33
Mohamed Anouar Ben Messaoud, A¨ıcha Bouzid, and
Noureddine Ellouze
Glottal Source Model Selection for Stationary Singing-Voice by
Low-Band Envelope Matching ..................................... 42
Fernando Villavicencio
Contribution to the Multipitch Estimation by Multi-scale Product
Analysis ........................................................ 50
Jihen Zeremdini, Mohamed Anouar Ben Messaoud,
A¨ıcha Bouzid, and Noureddine Ellouze
Speech Signals ParameterizationBased on Auditory Filter Modeling.... 60
Youssef Zouhir and Ka¨ıs Ouni
Towards a Better Representation of the Envelope Modulation
of Aspiration Noise............................................... 67
Jo˜ao P. Cabral and Julie Carson-Berndsen
Speech Synthesis
Towards Physically Interpretable Parametric Voice Conversion
Functions ....................................................... 75
Daniel Erro, Agust´ın Alonso, Luis Serrano, Eva Navas, and
Inma Herna´ez
VIII Table of Contents
Reduced Search Space Frame Alignment Based on Kullback-Leibler
Divergence for Voice Conversion ................................... 83
Abdoreza Sabzi Shahrebabaki, Jamal Amini, Hamid Sheikhzadeh,
Mostafa Ghorbandoost, and Neda Faraji
Average Voice Modeling Based on Unbiased Decision Trees............ 89
Fahimeh Bahmaninezhad, Soheil Khorram, and Hossein Sameti
Non-linear Pitch Modification in Voice Conversion Using Artificial
Neural Networks................................................. 97
Bajibabu Bollepalli, Jonas Beskow, and Joakim Gustafson
Speech-Based Biomedical Applications
AnalysisandQuantificationofAcoustic Artefacts in Tracheoesophageal
Speech ......................................................... 104
Thomas Drugman, Myriam Rijckaert, George Lawson, and
Marc Remacle
Analysis of Speech from People with Parkinson’s Disease through
Nonlinear Dynamics.............................................. 112
Juan Rafael Orozco-Arroyave, Julia´n David Arias-London˜o,
Jesu´s Francisco Vargas-Bonilla, and Elmar N¨oth
Synthesis by Rule of Disordered Voices ............................. 120
Jean Schoentgen and Jorge C. Lucero
Towards a Low-Complex Breathing Monitoring System Based on
Acoustic Signals ................................................. 128
Pere Mart´ı-Puig, Jordi Sol´e-Casals, Gerard Masferrer, and
Esteve Gallego-Jutgl`a
Automatic Detection of Laryngeal Pathologies in Running Speech
Based on the HMM Transformation of the Nonlinear Dynamics ........ 136
Carlos M. Travieso, Jesu´s B. Alonso, Juan Rafael Orozco-Arroyave,
Jordi Sol´e-Casals, and Esteve Gallego-Jutgl`a
Feature Extraction Approach Based on Fractal Dimension for
Spontaneous Speech Modelling Oriented to Alzheimer Disease
Diagnosis ....................................................... 144
Karmele Lo´pez-de-Ipin˜a, Harkaitz Egiraun,
Jordi Sole-Casals, Miriam Ecay, Aitzol Ezeiza, Nora Barroso,
Pablo Martinez-Lage, and Unai Martinez-de-Lizardui
Table of Contents IX
Automatic Speech Recognition
Robust Hierarchical and Sparse Representation of Natural Sounds in
High-Dimensional Space .......................................... 152
Simon Brodeur and Jean Rouat
Onthe ImportanceofPre-emphasisandWindowShape inPhase-Based
Speech Recognition .............................................. 160
Erfan Loweimi, Seyed Mohammad Ahadi, Thomas Drugman, and
Samira Loveymi
Smoothed Nonlinear Energy Operator-Based Amplitude Modulation
Features for Robust Speech Recognition ............................ 168
Md. Jahangir Alam, Patrick Kenny, and Douglas O’Shaughnessy
Fuzzy Phonetic Decoding Method in a Phoneme Recognition
Problem ........................................................ 176
Lyudmila V. Savchenko and Andrey V. Savchenko
Improved EMD Usable Speech Detection for Co-channel Speaker
Identification .................................................... 184
Wajdi Ghezaiel, Amel Ben Slimane, and Ezzedine Ben Braiek
Speech Enhancement
Speech Enhancement: A Multivariate Empirical Mode Decomposition
Approach ....................................................... 192
Jordi Sol´e-Casals, Esteve Gallego-Jutgla`, Pere Mart´ı-Puig,
Carlos M. Travieso, and Jesu´s B. Alonso
Speech Denoising Based on Empirical Mode Decomposition and
Improved Thresholding ........................................... 200
Issaoui Hadhami and A¨ıcha Bouzid
A Fast Semi-blind Reverberation Time Estimation Using Non-linear
Least Squares Method ............................................ 208
Neda Faraji, Seyed Mohammad Ahadi, and Hamid Sheikhzadeh
Author Index.................................................. 217
Evaluation of Automatic Glottal Source Analysis
John Kane and Christer Gobl
Phonetics and Speech Laboratory,
School of Linguistic, Speech and Communication Sciences,
Trinity College Dublin,Ireland
Abstract. This paper documents a comprehensive evaluation carried
out on automatic glottal inverse filtering and glottal source parameteri-
sation methods.Theexperimentsconsist ofanalysis ofawide varietyof
synthetic vowels and assessment of the ability of derived parameters to
differentiate breathy to tense voice. One striking finding is that glottal
model-based parameters compared favourably to parameters measured
directly from theglottal sourcesignal, in termsof separation ofbreathy
totensevoice.Also,certaincombinationsofinversefilteringandparam-
eterisation methods were more robust than others.
1 Introduction
The productionof voicedspeech canbe consideredas:the sound source created
bythevibrationofthevocalfolds(glottalsource)inputtedthroughtheresonance
structure of the vocal tract and radiated at the lips. Most acoustic descriptions
typically used in speech processing involve characterisation of mainly the vocal
tractcontributiontothespeechsignal.However,thereisincreasingevidencethat
developmentofindependent feature sets for boththe vocaltractandthe glottal
source components can yield a more comprehensive description of the speech
signal. Recent developments in speech synthesis [1], voice quality modification
[2],voicepathologydetection[3]andanalysisofemotioninspeech[4]haveserved
to highlight the potential of features related to the glottal source.
However, approaches for analysing the estimated glottal source are at times
believedtolackrobustnessincertaincases.Forinstance,higherpitchvoicesare
knowntobeproblematicforinversefiltering[5]andparticularlywhencombined
with a low first formant frequency. There can be strong source-filter interaction
effects [6] which seriously affect the linear model of speech exploited in inverse
filtering. Furthermore, precise glottal source analysis is often said to require
the use of high-quality equipment to record in anechoic or studio settings [5].
Despite these claims, some studies have found that glottal source parameters
derived from speech recorded in less than ideal recording conditions contribute
positively to certain analyses [7].
It follows that the purpose of this paper is to investigate the performance of
both inversefiltering andparameterisationsteps typically usedin glottalsource
analysis.The evaluationofglottalsourceanalysismethods is knownto be prob-
lematic as it is not possible to obtain ‘true’ reference values. To deal with this,
T.DrugmanandT.Dutoit(Eds.):NOLISP2013,LNAI7911,pp.1–8,2013.
(cid:2)c Springer-VerlagBerlinHeidelberg2013