14-15 Janvier 2015 - lcpts - Université des Sciences et de la

Transcription

Université des Sciences et de la Technologie Houari Boumediene
2èmes Journées du Laboratoire de Communication Parlée et de
Traitement de Signal
JLCPTS 2015
14-15 Janvier 2015
Message du Directeur du LCPTS
Le laboratoire de communication parlée et de traitement des signaux (LCPTS) a été crée en 2000. Il est né de la fusion de deux
groupes: le groupe communication parlée appartenant à la post-graduation dirigée alors par le Professeur Abdelhamid ADANE
et le groupe traitement du signal, faisant partie de la post-graduation dirigée par le Professeur Boualem Sansal. Le nouveau
laboratoire était alors formé de quatre équipes: reconnaissance, codage et production de la parole ainsi qu'une équipe de
traitement du signal. LCPTS fut dirigé pendant une dizaine d'années par le Professeur Amar Djerardi. Il fut ensuite dirigé par le
Professeur Debyeche Mohamed, durant un mandat. Durant ces quatorze années d'existence, le LCPTS a formé plusieurs
docteurs d'état (actuellement professeurs), plusieurs docteurs d'université et récemment des docteurs troisième cycle (LMD). En
2014, l'effectif du laboratoire était composé de 10 professeurs, 07 maîtres de conférences classe A, 02 maîtres de conférences
classe B, 09 maîtres assistants classe A et de 06 maîtres assistants classe B. Plus de 60 doctorants sont encadrés par les 17
professeurs et MCA du LCPTS dans diverses thématiques touchant à la communication parlée et au traitement du signal. Ce
nombre important de chercheurs a conduit à une restructuration du LCPTS et la création d'un nouveau laboratoire dérivant de
l'équipe traitement du signal. Ces chercheurs ont tenu à marquer encore leur présence et à participer à ces 2èmes journées
JLCPTS-2015.
Ces 2èmes journées du laboratoire de communication parlée et de traitement du signal (JLPTS-2015) se déroulent au
cyberespace de l'USTHB les 14 et 15 janvier 2015 et vont donc être un espace de rencontre, d'expression et de débats
scientifiques de plus de 90 chercheurs du LCPTS associés aussi à des chercheurs d'autres laboratoires. Les thèmes concernés par
ces deux journées sont divers et variés et relèvent à la fois d'aspects fondamentaux et également de recherche appliquée. Ils
couvrent des domaines variés en communication parlée et en traitement des signaux, tels que: les télécommunications, le codage
et la sécurité, la synthèse et le dialogue homme-machine, la reconnaissance de la parole et du locuteur, l'identification, la
classification, la pathologie de la voix, le traitement des documents, l'analyse des signaux de capteurs, les communications
numériques sans-fil et Radio-Intelligents...
Les travaux de ces journées vont s'effectuer à travers des communications orales et des posters des chercheurs des différentes
équipes. Des conférences plénières sont également prévues et seront présentées par des spécialistes en communication parlée et
en traitement du signal.
Ces journées vont aussi être relevées par la présence de personnalités ayant des connaissances et un intérêt avéré pour nos
travaux de recherche.
Je voudrais exprimer ici mes vifs remerciements aux membres du comité de programme pour avoir examiné minutieusement les
papiers qui leur ont été attribués. Je remercie aussi tous les membres du comité d'organisation pour le temps qu'ils ont consacré
à la préparation de ces journées. Leur seul souci était la réussite des JLCPTS-2015.
Enfin, mon souhait est que ces journées se pérennisent et deviennent annuels.
Prof. Bachir BOUDRAA
Directeur du LCPTS
Prof. Bachir BOUDRAA
Comité d’organisation
Pr Boudraa Bachir
Dr Mekaoui Slimane
Dr Chelali Fatma Zohra
Dr Falek Leila
Mr Halilali Abderrezak
Dr Hamza Abdelkrim
Dr Merazi-Meksen Thouraya
Dr Nemmour Hassiba
Dr Djeradi Rachida
Mr. Krobba Ahmed
Comité de programme
Pr Amrouche Abderrahmane
Pr Boudraa Bachir
Pr Boudraa Malika
Dr Bouzid Merouane
Dr Chelali Fatma Zohra
Pr Chibani Youcef
Pr Chitroub Salim
Pr Djeradi Amar
Dr Djeradi Rachida
Pr Debyeche Mohamed
Dr Falek Leila
Dr Fergani Belkacem
Dr Fergani Lamia
Dr Hamza Abdelkrim
Pr Houacine Amrane
Dr Mekaoui Slimane
Dr Merazi Meksen Thouraya
Dr Metref Adel
Dr Nemmour Hassiba
Dr Ouamour-Sayoud Sihem
Pr Sayoud Halim
Pr Selouani Sid Ahmed
Pr Teffahi Hocine
Mercredi 14 Janvier 2015
Horaire
Salle 1
08h00-09h00
Accueil des participants
Salle 2
Cérémonie d’ouverture
09h00-10h00
Pr. B. Benzaghou : Allocution du Recteur de l’université
Pr. Z. Ali Mazighi : Allocution de la Doyenne de la Faculté
10h00 – 11h00
Session plénière 1 : Big Data Analysis
Pr. M. Oussalah
Président: Pr. M. Debyeche
11h00 – 11h30
11h30 – 12h50
Co-Président: Dr. B. Fergani
Session Poster 1 : ID6, 7,8, 9, 10, 12, 14, 18, 19, 21,26
Session orale 1 : Télécom, codage et sécurité 1
ID21, ID32, ID39, ID43
Session orale 2 : Classification et reconnaissance de formes 1
ID1, ID2, ID36, ID55
Président: Pr S.Mekaoui
Président: Pr. H. Teffahi
Co-Président: Dr. M. Bouzid
Co-Président: Dr. R. Djeradi
12h50 – 14h00
14h00 – 15h00
Session orale 3: Identification et reconnaissance de la parole 1
ID4, ID11, ID13
Session Orale 4 : Télécom, codage et sécurité 2
ID15, ID46, ID20
Président : Pr. M. Guerti
Président: Pr. A. Adane
Co-Président : Dr. L. Fergani
Co-Président: Dr. H. Nemmour
15h00 – 15h30
15h30 – 16h30
Session orale 5 : Identification et reconnaissance de la parole 2
ID29, ID30, ID40
Session orale 6 : Indexation et traitement des signaux
ID23, ID34, ID44
Président: Pr. B. Boudraa
Président: Pr. A. Houacine
Co-Président: Dr. G. Droua
Co-Président: Dr. S. Ouamour
Jeudi 15 Janvier 2015
Horaire
Salle 1
Salle 2
Session plénière 2 : Multi-modal Biometric Authentications : Concept Issues and
Applications Strategies
09h00 – 10h00
Pr. M. Soltane
Président : Pr. Y. Chibani
10h00 – 11h00
Session plénière 3
Pr . V. Aubergé
Président : Pr. A. Djeradi
11h00 – 11h30
Co-Président : Dr. S. Mekaoui
Co-Président : Dr. L. Falek
Session poster 2 : ID27, 31, 33, 35, 37, 41, 42, 45,52 ,57
Session orale 7 : Synthèse et Dialogue Homme-machine
ID3, ID5, ID17, ID25
11h30 – 12h50
Président : Pr. M. Boudraa
Co-Président: Dr. A. Metref
ID16, ID22, ID24, ID28
Président: Pr. M. Attari
Co-Président: Dr. A. Hamza
12h50 – 14h00
14h00-15H30
15h30-16h30
Session orale 9: Identification et reconnaissance de la parole 3
ID47, ID50, ID54
Session orale 10: Télécom, codage et sécurité 3
ID48, ID49, ID51, ID56
Président : Pr. A. Amrouche
Président: Pr. H. Sayoud
Cérémonie de clôture
Co-Président: Dr. T. Merazi
Co-Président: Dr. F.Z. Chelali
Session plénière 1 : Big Data Analysis
Mercredi 14 Janvier 2015: 09h30 à 11h
Abstract
With the advances in information retrieval and data storage technology together with enforcing regulations, the emergence of
big data databases of order of several terabytes from sensory information and online servers, including intelligent cities
equipped with thousands of internal/external sensors, traffic control systems, e-health monitoring, banking systems, has
raised several challenges to information processing community. Especially, from data mining perspective, the huge amount of
data collected from various sources of information created new forms of uncertainty that may prevent identification of
relevant cues or proper exploitation of the data by local authority or decision-makers. Besides, the huge amount of
information, without appropriate pre-processing, renders any real-time analysis rather a difficult or compromising task. This
highlights the importance of pre-processing and architectural design tasks for any analysis of big data. Especially, the
following research questions should be addressed:
i. How to concisely identify the purpose of the big data analysis?
This can be identifying possible anomaly in the overall system, or classifying the input (s) into some predefined category, or
quantify some instrumental variable of the system, etc. In each case a special methodology will be required.
ii. How to filter the raw data in order to eliminate any noisy information?
This includes how to distinguish the various type of noise (inherent, systematic, random, etc.) where at each case a special
approach is required to tackle it.
iii. How to choose a feature space that is relevant for the purpose of the analysis?
Feature space is a way to map high dimension data into lower dimension. The type of feature is much context dependent and
the expected outcome of the study (PCA, MFFC, energy levels, eigenface, binary instrumental variables...).
iv. How to model the context of interest in the feature space?
E.g., through a vector representation, matrix, symbolic or binary variable.
v. How to design an architecture for analysing the context interest?
This involves distinguishing redundant information from complementary information while taking into account of
environmental constraints and expected outcomes, e.g., distinguish between classification, association, clustering, diagnostic
problems.
vi. How to choose an uncertainty framework (deterministic, stochastic, rule-based, soft computing)?
vii. How to quantify the performance of the data analysis system?
This requires defining the ground truth if any or a reference model. Performances metrics can include accuracy, recall, false
positive, false negative, etc.
The talk will highlight the above methodology of big data analysis using two special case studies. The first one consists in the
use of smartphone data (accelerometer and gyroscope data) in order to classify user’s movements. The second case study
consists makes use of a set of array of visual sensors for object tracking and diagnosis. In both cases, several alternative
reasoning-based approach will be contrasted and compared.
Prof. M. Oussalah
Biography
Dr. Mourad Oussalah est Professeur
associé à l’université de Birmingham qu’il a
joint depuis 2003. Après son doctorat de
l’université de Paris XII en 1998, il a
travaillé à l’université catholique de
Louvain 1998-2000, puis il a rejoint Cite
Université de Londres (2000-2003) ou il
a travaillé dans les problèmes de fiabilité
des systemes home-machines. Mourad
Oussalah a publié plus de 200
publications internationales. Il est Fellow
de Royal Statistical Society et de IET
Society et Senior member de IEEE. Ses
activités de recherches se concentrent
principalement dans les thématiques de
fusion de données, de l'intelligence
artificielle, Il raisonnement à base de
logique floue et possibiliste, classification,
vision, robotique, raisonnement à base de
logique floue et possibiliste, classification,
vision, robotique.
Session
plénière
2 :09h
Multi-modal
Biometric Authentications : Concept Issues and Applications Strategies
Jeudi
15 Janvier
2015:
à 10h
ABSTRACT
As the information age matures, biometric identification technology will be at the heart of computer interaction with
humans and the biosphere in which they reside. an application dependent decision. These biometric systems for personal
Automated biometric systems for human identification measure a “signature” of the human body, compare the resulting
characteristic to a database, and render authentication and identification are based upon physiological or behavioral features
which are typically distinctive, although time varying, such as fingerprints, hand geometry, face, voice, lip movement, gait,
and iris patterns. Multi-biometric systems, which consolidate information from multiple biometric sources, are gaining
popularity because they are able to overcome limitations such as non-universality, noisy sensor data, large intra-user
variations and susceptibility to spoof attacks that are commonly encountered in uni-biometric systems.
The main challenging objectives of thisResearch Presentation are as follows: i) Present a Design of a working multimodal
biometric verification and identification system, using speech and signature as verification parameters, and applying the
proposed techniques to common scenarios and databases available for the biometrics research community; ii) Reviewing
and studying the problem of biometric system, the score normalization and score fusion of a multimodal biometrics
authentication system, considering the user specific information at a hand as well as the quality of the input biometric
signals; iii) Justification of the fusion performance: from a reference to the paradigm that in classification tasks, the use of
multiple observations and their judicious fusion at the data, hence the decision fusions at different levels improve the
correct decision performance; iv) The system performance, upon presentation of a biometric sample, a system should grant
access (if the person is a client/user) or reject the request (if the person is an impostor). In general terms, this decision is
made by comparing the system output with an operating threshold. In this process, two types of error can be committed:
falsely rejecting a genuine user or falsely accepting an impostor. The error rates are respectively called False Rejection Rate
(FRR) and False Acceptance Rate (FAR). These two errors are important measures to assess the system performance which
is visualized using a Detection Error Trade-off (DET) curve. A special point called Equal Error Rate (EER), where
FAR=FRR; v) the consequence, for novel strategies that have to be explored: the motivation and the perspective of
combining cryptography with biometrics; vi) the challenging applications strategies of multimodal biometrics systems from
the Philosophy That: In order to deploy biometrics effectively, it will need to understand the dynamics of the specific
applications. The End of the Presentation Focus on Molecular Biology Computational Design Research, Quantum DNA
mathematical modeling tools and high-performance computing that dramatically lowers the cost and shortens the DNA
sequencing and indexing times,….
Prof. M. Soltane
Biograhy
Mohamed SOLTANE (Assoc. Prof. Dr.)
received the M.Eng. degree in Electronics from
Badji Mokhtar University of Annaba-Algeria,
in 1995 and the M.Sc. degree in Electrical and
Electronics Engineering from UKM Malaysia in
2005, and the Ph.D. degrees in Electronics from
Badji Mokhtar University of Annaba-Algeria,
in 2010. He is currently an Associate Professor
at Electrical Engineering & Computing
Department, Faculty of Sciences & Technology,
Doctor yahia fares University OF MEDEA,
ALGERIA. His research interests include
statistical pattern recognition, biometric
authentication, cryptography and quantum
computing, computer vision and machine learning
and microcomputer based system design.
Session plénière 3 : La glu socio-affective au centre de l’Interaction Humain-Robot: vers une réalité co-construite
écologiquement.
Jeudi 15 Janvier 2015: 10h à 11h
Résumé
L’hypothèse forte sous-tendant ces travaux est quel que soit le rôle social, obligatoire, inventé ou
emprunté par le robot entrant dans la sphère sociale de l’humain, les compétences « professionnelles »
ou « compagnon » du robot ne seront intégrées par l’humain que si un lien relationnel pertinent
permet le véhicule dialogique, et que ce lien repose sur la construction d’une « glu socio-affective »
co-construite par l’humain relativement au rôle qu’il attribue au robot.
Le dialogue est considéré ici comme un système dynamique co-construit sur le tissu social qui ne peut
être décrit que dans la signature évolutive de la relation intrinsèquement inscrite dans le temps sociocognitif.
Une hypothèse forte est avancée sur le boostrap primitif du dialogue : la prosodie « pure » (micromots non linguistiques) serait la base de tout processus interactionnel.
Pour le démontrer, nous décrirons en particulier une expérience écologique menée dans le Living Lab
Domus du LIG : un scénario est dédié à capturer comment, pour un rôle de majordome domotique,
le robot Emox volontairement non humanoïde et non animaloïde (société Awabot) co-crée un
attachement relatif à son statut avec des personnes âgées, accompagnées de professionnels de
l’assistance, en se dotant d’expressions strictement para-linguistiques (bruits de bouche, fillers,
onomatopées, interjections, imitation prosodiées en glu positive) supposées croissantes dans leur
compétence à construire la glu socio-affective. Les formes et la nature des comportements
dialogiques des interactions humaines recueillies (corpus EEE Elderly Emoz Expressions) montrent
que si le rôle de majordome est accepté utile, voire nécessaire, par ce public fragilisé, la fonction jugée
fondamentale pour le robot est celui de la construction du lien (comme le confirment également les
enquêtes également menées). Le robot pourrait ainsi jouer en réalité le rôle d’ « entraîneur » à
l’interaction socio-affective et faciliter en conséquence les relations sociales de personnes en
isolement socio-affectif. La méthodologie expérimentale mise en place permet de construire, autour
des hypothèses qui sous-tendent la nature et les dimensions des primitives vocales de la « glu socioaffective », un système dialogique minimal construit autour de la personne et son comportement, et
potentiellement évolutif en apprentissage machine actif sur la paradigme de la télé-assistance.
Prof. V. Aubergé
Biographie
Véronique Aubergé est chercheur CNRS en Sciences Humaines et
Sociales au Laboratoire d’Informatique de Grenoble où elle est animateur
du Living Lab Domus du LIG, et co-porteur du défi du LIG sur
l’interaction et la robotique sociale. Elle est responsable du comité
d’étique SFTAG sur l’interaction et la robotique sociale. V Aubergé a
obtenu en 1991 une thèse Cifre d’Informatique et de Linguistique à
Grenoble (après un DEA en Sciences du Langage et en parallèle un
DEA en Informatique). Elle a appartenu 5 ans au Centre de Recherche
en Informatique des Sciences Sociales, puis pendant 4 ans dans la société
OROS, elle a été responsable d’un projet européen en synthèse vocale ; elle
a enseigné 2 ans à l’ENSIMAG. En parallèle, elle a été membre
partiel ou plein depuis 1983 de l’Institut de Phonétique de Grenoble
devenu ICP (où elle a été responsable de l’équipe Structure du Code) puis
devenu GIPSA Lab, qu’elle a quitté en 2012 pour le LIG. Elle est coporteur d’un master trans-disciplinaire Industries de la Langue. Ses
travaux ont porté initialement sur la modélisation de la transcription
orthographique-phonétique du français (dans les systèmes de synthèse
vocale du CNET) et sur la conception de systèmes de synthèse de la
parole guidés par la prosodie naturelle (modèle de Gestalts). Elle a
proposé une théorie cognitive de la communication socio-affective qui
inscrit le dialogue dans une dynamique évolutive. S’appuyant sur une
méthodologie expérimentale écologique de co-construction, ses travaux
portent sur la compréhension de la nature de la glu socio-affective
construite par les outils communicatifs de l’interaction dynamique face à
face, en particulier dans les lieux endommagés de la communication
Sommaire
Session orale 1 : Télécom, codage et sécurité 1
ID32: Implementation of Steganographic Method Based on IPv4 Identification Field
Hamza KHEDDAR, Merouane BOUZID
ID39: Impact of insertion based recovery techniques on echo canceller performance
Abdelhamid ARCHI, Amrane HOUACINE
ID43: A PCA based algorithm for detecting volume traffic anomalies in IP networks
Choukri BENAHMED, Slimane MEKAOUI
1
2
3
ID01: Writer’s gender classification using HOG and LBP features
Nesrine BOUADJENEK, Hassiba NEMMOUR, Youcef CHIBANI
ID02: Off-line handwritten signature verification using variants of local binary patterns
Yasmine SERDOUK, Hassiba NEMMOUR, Youcef CHIBANI
ID36: An efficient classifier using Smote-CSVM to imbalanced human activity datasets
M’Hamed Bilal ABIDINE, Belkacem FERGANI
ID55 : Implémentation d’un système de reconnaissance automatique de la parole visuelle par les SVM
Nadia BAKIR, Mohamed DEBYECHE, Abderrahmane AMROUCHE
4
5
6
7
Session orale 3: Reconnaissance et identification de la parole 1
ID04: Effect of the front-end processing on speaker verification using GMM-UBM and SVM
Nassim ASBAI, Abderrahmane AMROUCHE, Youcef AKLOUF
ID11: Caractérisation du systèmevocalique berbère
Fazia KARAOUI, Amar DJERADI
ID13: Detection of questions in berber language using prosodic features
Ramzi HALIMOUCHE, Hocine TEFFAHI, Leila FALEK
8
9
10
Session orale 4: Télécom, codage et sécurité 2
ID15: Frame loss recovery method for CELP-Codec by using RLS filter
11
Nadir BENAMIROUCHE, Bachir BOUDRAA
ID46:Implementing PLCtechniques with G729 coded to improve speech quality for VoIP transmission12
Adil BAKRI, Abderrahmane AMROUCHE
ID20: FPGA implementation of modular exponentiation for RSA cryptosystem
13
Mohamed ISSAD, Bachir BOUDRAA, Mohamed ANANE
ID29 : La prosodie dans la détection de l’état émotionnel
Mohamed Réda MOHAREM, Hocine TEFFAHI
ID30: Spectral cues based acoustic distinctive features for robust distributed speech recognition
Djamel ADDOU, Bachir BOUDRAA
ID40: A noise-robust front-end for multi-stream HMM based speech recognition systems
Mohamed Rafik DAALACHE, Djamel ADDOU, Malika BOUDRAA
14
15
16
Session orale 6: Indexation et traitement des signaux
ID23: Automatic speaker localization using two-ways sound source detection
Salah KHENNOUF, Halim SAYOUD, Siham OUAMOUR
ID34: Distance estimation versus fingerprinting methods for indoor localization
Zineb BELHADI, Lamia FEGRANI
ID44: Linear array sensors for speech localization and denoising database
Hamid MERAOUBI, Bachir BOUDRAA
17
18
19
Session orale 7: Synthèse et Dialogue Homme-machine
ID03: Interface for dynamic modification of the transformation parameters of the PSOLA algorithm 20
Lyes DEMRI, Leila FALEK, Hocine TEFFAHI
ID05: Une approche statistic-linguistique pour l’esxtraction de concepts sémantiques: Une première
étape vers un système générique de dialogue Homme-Machine
21
Mohamed LICHOURI, Amar DJERADI, Rachida DJERADI
ID17: Automatic classification of human body postures based on truncated SVD
22
Nabil ZERROUKI, Amrane HOUACINE
ID25: Speech enhancement using Kalman filter
23
Asma BOUCHAIR, Abderrahmane AMROUCHE
Session orale 8: Classification et reconnaissance de formes 2
ID16: Co-occurrence matrix for ancient degraded document image binarization
Abdenour SEHAD, Youcef CHIBANI, Mohamed CHERIET
ID22: Combination of off-line and On-line signature verification systems based on SVM and DST
Nassim ABBAS, Youcef CHIBANI
ID24: A feature generation method based on the contourlet transform and normalized energies for
line handwritten signature verification
Assia HAMADENE, Youcef CHIBANI
ID28: Multiple One-class classifer combination for multi-class classification
Bilal HADJADJI, Youcef CHIBANI
24
25
off26
27
ID47: SVM-based voice activity detection for distributed speech recognition system
Azzedine TOUAZI, Mohamed DEBYECHE
ID50: Combining auxiliary and cepstral features for speech recognition
AnissaImen AMROUS, Mohamed DEBYECHE
ID54: Parallel GMM sub-bandspeaker identification over AMR-NB codec
Hadjer SABOUNE, Mohamed DEBYECHE
28
29
30
Session orale 10 : Télécom, codage et indexation 3
ID48: Quantificateur vectoriel à divisions commutées appliqué au codage des paramètres ISF de la
parole en large bande
31
Salah-Eddine CHERAITIA, Merouane BOUZID
ID49: Automatic assessment of the dysarthria severity levels
32
KamilLahcene KADI, Sid Ahmed SELOUANI, Bachir BOUDRAA, Malika BOUDRAA
ID51: Request efficient channel estimation method for MIMO RFID systems
33
Fouad KHELLADI, Adel METREF, Belkacem FERGANI
ID56: Tatouage d’images par insertion aveugle et informée et détection par corrélation linéaire
34
Abdelhami SAIGHI, Salim CHITROUB
Session Poster 1 :
ID06: Facial expression identification system with Euclidean distance of facial adges
Sonia KHERCHAOUI, Amrane HOUACINE
ID07: Tracking characteristic points on a talking face
Noureddine CHERABIT, Amar DJERADI
ID08: American sign language recognition using probabilistic neural network
Khadidja SADEDDINE, F.Zohra CHELALI, Rachida DJERADI
ID09: Performance comparison of ALOHA and tree based anti-collision algorithms in RFID system
Mohamed Aïssa KALACHE, Lamia FERGANI
ID10: Arabic keyword spotting in ancient document images using curvelet descriptors
Youcef BRIK, Youcef CHIBANI
ID12: Language identification based on high frequency approaches
Kheireddine ABAINIA, Siham OUAMOUR, Halim SAYOUD
ID14: Writer-independent handwritten signature verification based on one-class SVM classifier
Yasmine GUERBAI, Youcef CHIBANI, Bilal HADJADJI
ID18: Historical document image binarization using contourlet transform
Et-Tahir ZEMOURI, Youcef CHIBANI, Youcef BRIK
ID19: L’acquisition compressée dans les systèmes mobiles
Houria HANECHE, Bachir BOUDRAA, Abdeldjalil OUAHABI
ID21: Digital watermarking in audio for copyright protection
Mustapha HEMIS, Bachir BOUDRAA
ID26: Utilisation des systèmes 3D pour la détection et quantification des mouvements au cours
examens scintigraphiques
Sofiane TCHOKETCH KEBIR, Slimane MEKAOUI
35
36
37
38
39
40
41
42
43
44
des
45
Session Poster 2:
ID27 : Contribution à la réalisation d’un système TTS pour la langue Arabe
46
Aïssa AMROUCHE, Leila FALEK, Hocine TEFFAHI
ID31: Improved packet loss recovery using multiple description coding in MELP coder for voice over IP
Mohammed SAIDI, Leila FALEK, Bachir BOUDRAA
47
ID33: Transmission sécurisée par le système PGP en vue de protéger les courriers électroniques
48
Yacine BOUFLIH, Salim CHITROUB
ID35: Automatic speech recognition in GSM network using the Bit-stream and auxiliary parameters 49
Lallouani BOUCHAKOUR, Mohamed DEBYECHE
ID37: Etude comparative de méthodes de correction de pertes de paquets en transmission de voix sur IP
Hamza KHEDDAR, Bachir BOUDRAA
50
ID41: A novel voice activity detection approach for automatic speaker verification
51
Meriem FEDILA, Abderrahmane AMROUCHE, Messaoud BENGHERABI
ID42: Enhancement of A5/1 stream cipher by overcoming its weaknesses
52
Mahdi MADANI, Salim CHITROUB
ID45: Contribution à l’implémentation d’algorithmes de traitement de la parole sur circuit FPGA
53
Abderrezak HALILALI, Imen TABATOUCHE, Amrane HOUACINE
ID52: SVM modeling for handling imbalanced human activity datasets
54
Nawel YALA, Belkacem FERGANI
ID57: Authentification vocale d’un locuteur arabophone en vue de la criminalistique
55
Ouassila KENAI, Mhania GUERTI
JLCPTS2015 14-15 Janvier 2015
Session orale 1
Télécommunications, Codage et Sécurité 1
ID32
1
Implementation of Steganographic Method
based in IPv4 Identification Field
Hamza Kheddar and Merouane Bouzid
Laboratoire de Communication Parlée et de Traitement des
Signaux (LCPTS)
Faculté d’électronique et d’informatique
Université des sciences et de Technologie Houari Boumediene
USTHB, Alger, ALGÉRIE.
Abstract: In this paper we present first a study of covert
channels (steganography) that may be applied for each TCP/IP
layer in VoIP application. Then, we present a steganographic
method which hide secret data in IP protocol header fields,
particularly the identification field. The IP protocol covert
channel implementation was carried out in NS-3 (Network
Simulator 3).
Keywords: Data hiding, network steganography, Secure VoIP,
TCP/IP protocols, Covert Channel.
ID39
2
Impact of Insertion Based Recovery
Techniques on Echo Canceller Performance
Abdelhamid Archi and A. Houacine
Signaux (LCPTS)
USTHB
Alger, ALGÉRIE.
Abstract: This paper presents a comparative study of the
impact of packet loss on the performance of an echo canceller.
More specifically it investigates the impact of insertion based
recovery techniques on the performance of a VoIP echo
canceller. The work is conducted using the well known NLMS
adaptive algorithm as well as its different improvements
PNLMS and IPNLMS. The simulation is carried out using a
sparse impulse response and an audio source signal taken from
ITU-G168 recommendation.
The measure of the echo canceller performances studied is
based on the ERLE.
ID43
3
A PCA Based Algorithm for Detecting
Volume Traffic Anomalies in IP Networks
Benhamed Choukri and Mekaoui Slimane
Signaux (LCPTS)
Abstract: In this paper we deal with the principal component
analysis technique that has been already applied to a large
variety of data. In this work we apply it to an IP traffic data for
detecting eventual anomalies in this data. The algorithm that we
have developed partitions data into two subspaces namely
estimated and residuals. The computing of statistical
parameters has provided us with tools that allowed us to
perform good results. We have studied the case of 24 traffic
matrix issued from a well known network, the Abilene
network; it appears that our algorithm yields good results of
anomalies detection.
Keywords- Principal Com ponent Analysis; Anomaly
Detection,
Session orale 2
Classification et Reconnaissance de Formes 1
ID01
4
HOG and LBP features for writer’s Gender
Classification
Nersrine Bouadjnek, Hassiba Nemmour and Youcef
Chibani
Signaux (LCPTS)
USTHB, Alger, ALGÉRIE
Abstract: Gender prediction based on the hanwritten text
becomes to earn a considerable importance for the document
analysis community. It provides information about anonymous
documents for which, we need to know if they were written by
a Man or a Woman. In this work, we propose a system for
writer’s gender classification that is based on local textural and
gradient features. Specifically, we employ Histogram of
Oriented Gradients (HOG) and Local Binary Patterns (LBP),
which are successful in various pattern recognition
applications.
The prediction task is achieved by SVM classifier. The results
obtained on samples extracted from IAM dataset show that the
proposed features provide quite promising results.
Keywords: Gender classification; HOG; LBP; SVM.
ID02
5
Orthogonal Combination and Rotation
Invariant ofLocal Binary Patterns for Offline Handwritten SignatureVerification
Yasmine Serdouk, Hassiba Nemmour and Youcef
Chibani
Signaux (LCPTS)
Abstract: This paper proposes a novel descriptor called,
Orthogonal Combination of Local Binary Pattern (OC-LBP)
used to solve off-line handwritten signature verification. It is
developed to reduce the size of LBP histogram while keeping
the same characterization power. The verification task is carried
out using Support Vector Machines (SVM) classifiers
whereOC-LBP is combined with the rotation invariant uniform
LBP to allow both, size reduction and accuracy improvement
over the classical LBP features. Experiments are conducted on
GPDS300 and CEDAR datasets. The results in terms of
average error rate show significant improvement over the state
of the art methods.
Keywords: Signature verification; Orthogonal combination of
local binary patterns; Rotation invariant uniform LBP; SVM.
ID36
6
An Efficient Classifier using Smote-CSVM
to Imbalanced Human Activity Datasets
M’Hamed Bilal Abidine and Belkacem Fergani
Signaux (LCPTS)
Abstract: Activity recognition datasets are generally
imbalanced, meaning certain activities occur more frequently
than others. Not incorporating this class imbalance results in
an evaluation that may lead to disastrous consequences for
elderly person. In this work, we evaluat e an efficient classifier
named SMO TE-CSVM combining the Synthetic Minority
Over-sampling Technique (SM OTE) with the discriminative
method Soft-Support Vector
Machines (CSVM ) in order to handle imbalanced data
problem. Our experiments carried out on different real world
activity recognition datasets, demonstrate that the SMO TECSVM is able to surpass the standard CSVM, CS-SVM and
CRF for classifying the minority activities.
Index Terms: Activity recognition, SVM, machine learning,
imbalanced data, sensor network.
ID55
7
Implémentation d’un Système de
Reconnaissance Automatique de la parole
Visuelle par les SVMs
Nadia Bakir, Med Debyeche and A. Amrouche
Signaux (LCPTS)
Résumé : Dans cet article, nous avons élaboré un système de
Reconnaissance Automatique de la Parole Visuelle (RAPV) par
les Supports à Vecteurs Machines (SVM) comme méthodes de
reconnaissance. Nous avons utilisé plusieurs méthodes de
paramétrisation du signal parole visuel, pour trouver la
meilleure méthode qui s’adapte mieux à la base de données
utilisée ainsi que le système de reconnaissance. La première
méthode est l’approche par estimation du gradient qui nécessite
plusieurs prétraitements, la deuxième méthode est basée sur les
contours actifs par approche des ensembles de niveaux, elle
permet une estimation moyenne des contours extérieurs des
lèvres, mes ne permet pas d’avoir une bonne détection sur tout
la base de données, car ces paramètres change d’une image à
une autre.
Nous avons opté pour la DCT comme solution à nos
problèmes, elle a permet d’avoir un très bon taux de
reconnaissance.
Mots clés : RAP visuelle, SVM, contours actifs, DCT.
Session orale 3
Reconnaissance et Identification de la Parole 1
ID04
8
Effect of the Front-End processing on
Speaker verification using GMM-UBM and
SVM
Nassim Asbai, Abderrahmane Amrouche and Youcef
Aklouf
Signaux (LCPTS)
Abstract: This paper presents a comparative study between
the statistical method based on Gaussian Mixture Model
(GMM) and Support Vector Machines (SVM) for textindependent speaker verification. The effect of the front-end
processing was investigated. Then, various features vectors are
extracted from TIMIT corpus : i) the Mel Frequency Cepstral
Coefficients
(MFCC) and their first and second derivatives plus energy
parameter, ii) the Linear Predictive Coefficients (LPC) and iii)
the Line Spectral Frequency (LSF). The results that achieved
show that, the use of MFCC parameters improves the
performance of speaker verification especially with the SVM
recognizer.
Keywords: Speaker verification; GMM; UBM; Expectation
minimization; SVM; MFCC; energy; LPC; LSF;TIMIT.
ID11
9
Caractérisation du Système Vocalique
Berbère (Kabyle)
Fazia Karaoui and Amar Djeradi
Signaux (LCPTS)
Résumé : Dans cette étude, en vue de caractériser les
voyelles du berbère, on a procédé à la mesure directe de la
fonction de transfert du conduit vocal par la méthode
d’excitation extérieur du conduit vocal par une séquence
binaire pseudo aléatoire
(Pseudo Random Eexcitation )[Djéradi et al 1991]. Les
voyelles [a, i, u] ont été prononcées par deux locuteurs ;
féminin et masculin de langue maternelle Kabyle. Les
enregistrements ont été effectués dans un environnement
ambiant au niveau du laboratoire de la communication parlée.
Le signal est numérisé à la fréquence d’échantillonnage de 10
kHz, les valeurs des trois premiers formants ont été mesurées.
Mots clé : fonction de transfert; conduit vocal; système
vocalique de la langue berbère.
ID13
10
Detecting Sentences Types in Berber Language
Ramzi Halimouche, Hocine Teffahi and Leila Falek
Laboratoire de Communication Parlée et de Traitement des Signaux
(LCPTS)
Abstract: The standard Arabic language, like many other languages,
contains a prosodic feature hidden in the speech signal. The studies in
this language are still preliminary, which immobilizes the realization
of communication tools. The prosodic study allows for each people to
have all tools to communicate in their native language. This paper
focuses on the acquisition of the tonal and prosodic structure of
affirmative and question sentences in Berber language. The study on
the prosodic differences between these two types of sentences in
Berber language, the detection and classification of sentence type is
the main subject of this paper. We’ve realized a system for
segmentation and automatic detection of sentence type based on both
prosodic, in Berber language, a language where all studies until now
are still preliminary. To this end, we developed a corpus made of 720
utterances that were extracted from 6 Berber spoken lectures. Prosodic
features are, then, extracted from each sentence. These features are
used as input to two different classifiers to classify each sentence into
either a question or affirmative sentence. We classified questions with
an accuracy of 93%. A feature-specific analysis further reveals that
energy and fundamental frequency (F0) features are mainly
responsible for discriminating between question and affirmative
sentences.
Keywords: question detection, prosodic analysis, Berber
language, classification.
Session orale 4
Télécommunications, Codage et Sécurité 2
ID15
11
Frame Loss Recovery Method for CELP-Codec
by Using Adaptive Filter
Nadir Benamirouche and Bachir Boudraa
Signaux (LCPTS)
Abstract: In this paper, we present a new frame loss recovery method
for Algebraic Code Excited Linear Prediction (ACELP) codecs in case of
speech transmission over erasure channels. Linear Prediction coding is
based on a strong interframe dependency which renders CELP codecs
very sensitive to packets loss. When the frames are properly received, the
speech signal is efficiently synthesized thanks to the good selection of
both: adaptive and fixed excitations. Indeed, it is otherwise; in case of
loss, a bad estimate of excitation introduces degradation on synthesized
speech signal and causes an error on the next correctly received frame,
causing error propagation. The proposed method tries to remedy this
problem and avoid the bad estimation of the introduced excitation by
packets loss concealment algorithm embedded in decoder. Therefore, the
goal of the proposed method is to appeal to adaptive filtering by using
RLS filtering to learn the changes between two successive excitations
and to be a halfway between the past and current excitation.
Subsequently, once the frame is lost, its excitation will be estimated by
filtering the past excitation by the previously adjusted RLS filter. Thus,
our method offers a real time frame loss recovery for speech transmission
over lossy packets channels.
Keywords: Speech coding, frame loss recovery, excitation
reconstruction, RLS filter, Lossy network transmissions
ID46
12
Implementing the PLC Techniques with G
729 Coded to Improving the Speech Quality
for VoIP Transmission
Adil Bakri and Abderrahmane Amrouche
Signaux (LCPTS)
Abstract: Speech compression technology is widely used in
digital communication systems such as wireless systems and
VoIP. The compression technique described in the ITU-T
G.729
Recommendation is commonly employed in speech
transmission systems because of the reconstructed speech
signal in reciever. To increase the quality of speech on the
Voice Over Internet Protocol (VoIP), we propose in this paper
the technique of Packet Loss Concealment (PLC) and analyzes
the performance results of ITUT G.729 codec with this
technique. The PLC algorithms used ranged from simply
inserting silence and repeated for the missing audio, to the use
of the G.711 Appendix I algorithm that does a good job of
generating a synthetic speech signal to cover missing data in a
received bit stream, to ensure a smooth transition between the
real signal and the synthetic signal.
Keywords: VoIP; OLA; PLC; G729; ITU-I G711 Appendix I;
PESQ
ID20
13
FPGA Implementation of Modular Exponentiation
for RSA Cryptosystem
Mohamed Issad*, Bachir Boudraa** and Mohamed
Anane***
*Centre de Développement des Technologies Avancées, Baba
Hassen, Alger, ALGÉRIE
**Laboratoire de Communication Parlée et de Traitement des
Signaux (LCPTS)
*** Ecole Supérieur d’informatique, Alger, Algérie
Abstract: In this paper, the implementation of RSA (Rivest,
Shamir and Adleman) cryptosystem using Hardware/Software
(HW/SW) co-design approach is presented. The main function
of RSA is the Modular Exponentiation (ME) which is
accomplished by repeated Modular Multiplications (MM’s) on
large integers. In this work, we propose the implementation of
the ME as a Programmable System on Chip (PSoC), using the
soft processor core MicroBlaze of Xilinx. In the proposed
embedded system, the control of the ME algorithm is executed
by the embedded processor, while the MM’s are computed
within specific hardware. Because, the MM is usually
considered as timeconsuming, the Montgomery modular
multiplication algorithm is employed to realize our accelerator
customized IP (Intellectual Property). The results show that the
application to RSA 1024bits, the execution time of the ME is
about 15.14 ms.
Keywords: RSA, Montgomery Modular Multiplication, FPGA,
PSoC
Session orale 5
ID29
14
Détection de l’état émotionnel grâce aux
paramètres prosodiques
Moharem Mohamed réda and Hocine Teffahi
Signaux (LCPTS)
Résumé: Depuis quelques années, les études sur la parole
émotionnelle vont au delà d’une analyse des manifestations
vocales des différents états émotionnels et commencent à
s’intéresser au développement de systèmes de classification
automatique des émotions.
Un grand nombre d’application de la reconnaissance des
émotions dans la voix concerne le domaine de l’interaction
homme machine,les systèmes de dialogue en sont un premier
exemple, la détermination de l’état émotionnel de l’utilisateur
permet d’adapter la stratégie dialogique.
Déterminer uels descripteurs sont essentiels et quels sont ceux
à éviter est un défi complexe puisqu’il nécessite de travailler
sur un grand nombre de variabilités et donc d’avoir à sa
disposition des corpus riches et variés. Les principaux résultats
portent à la fois sur la collecte et l’annotation de corpus
émotionnels réalistes avec des locuteurs variés (enfants,
adultes, personnes âgées).
Mots clés : émotion, prosodie, parole.
ID30
15
Spectral Cues based Acoustic Distinctive
features for Robust Distributed Speech
Recognition
Djamel Addou and Bachir Boudraa
Signaux (LCPTS)
Abstract: In this paper, a new framework integrating a noise
robust Front end in distributed speech recognition (DSR) is
presented. Using the Aurora 2 speech database, we evaluate the
impact of the proposed multidimensional acoustical features on
the performance of the Mel-frequency based ETSI advanced
front end. From the denoised acoustic frame using the wiener
filter, we optimize the stream weights of multi-stream HMMs
by deploying a discriminative approach based in LikelihoodRatio Maximization (LRM). Finally, these features are
adequately transformed and reduced in a multi-stream scheme
using Karhunen-Loeve Transform (KLT). The proposed front
end is shown to exhibit a relative error rate reduction and
provides comparable recognition performance compared with
the current DSR-FE (Distributed Speech RecognitionFrontend) available in GSM mobile communications.
Keywords: Distributed speech recognition, noise robustness,
front end processing, wiener filter, LRM, KLT
ID40
16
A Noise-Robust Front-end for Multi-stream
HMM based Speech Recognition Systems
Mohamed Rafik Daalache, Djamel Addou and Malika
Boudraa
Signaux (LCPTS)
Abstract: This paper present new robust front-end processing
whose main aim is the improvement of Distributed Speech
Recognition (DSR) systems performance when perturbed by
severe environmental noise. The improvement is performed by
using a combination of post proceeded Mel-cepstral
coefficients (MFCCs) and line spectral frequencies features
(LSFs) projected in linear discriminate analysis (LDA) space.
The experiments carried out on the Aurora 2.0 database show
that, even with fewer parameters, the proposed front-end
provides comparable recognition results to the standard ETSI
WI008 advanced frontend, nowadays used as codec in 3rd
generation mobile communication systems, while achieving
higher accuracy when the signal-to-noise ratio (SNR) is very
low.
Keywords: distributed speech recognition; linear discriminate
analysis; multi-stream hidden Markov models; Mel-cepstral
coefficients; line spectral frequencies.
Session orale 6
Indexation et Traitement des Signaux
ID23
17
Automatic Speaker Localization using twoways Sound Source Detection
Salah Khenouf, Halim sayoud and Siham Ouamour
Signaux (LCPTS)
Abstract: The problem treated in the present paper is the
audio speaker localization, which consists in locating the
position of an active speaker in a meeting-room. We mean by
an
active speaker, a person who is speaking. This phase represents
the first step of speaker tracking, which is the main purpose of
our research work. Our approach, consisting in a two-channelbased (stereo) estimation of the active speaker location, is
achieved by comparing the two signals received by two
cardioids microphones placed one against the other and spaced
by a fixed distance. This localization technique is inspired from
the human ears, which act as two different sound observation
points, enabling humans to estimate the location of the active
speaker with a very good precision. The off-line experiments of
speaker tracking have been done in a small-meeting room
without echo cancelation. Results show the good performances
of the proposed localization methods.
ID34
18
Distance Estimation versus Fingerprinting
Methods for Indoor Localization
Zineb Belhadi and Lamia Fergani
Signaux (LCPTS)
Abstract: Different approaches based on various wireless
technologies have been proposed so far for indoor localization.
Radio frequency Identification (RFID) indoor localization
seems to be a promising way of research. The identification
capability of this technology combined to localization methods
improves the results obtained by other wireless technologies
such as Wifi, GPS,Zigbee... This paper details some
localization techniques used for RFID Tags localization in
Indoor environment. In particular, Fingerprinting methods are
compared to Distance estimation methods. We will show
through several simulation experiments, using NS2 and Matlab
software, that fingerprinting techniques outperform Distance
estimation techniques for localization and tracking tasks.
Keywords: RFID technology; Localization; Indoor vironment;
Fingerprinting.
ID44
19
Data Base for Speech Localisation and
Denoising using Microphone Array
Hamid Meraouibi and Bachir Boudraa
Signaux (LCPTS)
Abstract: The MUSLOD (Multi-microphone Source LOcalisation
Database) data base presented in this document is the result of work
carried out in the center of development of advanced technologies in
collaboration with the laboratory of «communication parlée» at USTHB.
The database thus obtained contains 63,360 seconds (18 hours). Inspired
from existing databases of specialized community, we propose a platform
for sound acquisition using four microphones in order to develop a
database called MUSLOD. It will be used for localization of sound
sources in reverberant environments. We present the experimental device
used for the development of basic MUSLOD data and place of
registration and the terms of implementation. It consists of actual data
desktop environment from a network of linear microphones. For this, we
use different values of sampling and different inter-microphone distances
frequencies, location data are presented for different emission angles of
the sound source. The database is divided into two broad categories
registration: The first category contains records acquired but denoised by
a method of spectral subtraction. The second category contains rough
recordings. The sampling frequencies used are 8, 16, 44.1, and 64 KHz.
Keywords: source; localization; database; microphone array; denoising
Session orale 7
Synthèse et Dialogue Homme-Machine
ID03
20
Interface for Dynamic Modification of the
Transformation Parameters of the PSOLA Algorithm
Lyes Demri, Leila Falek and Hocine Teffahi
Signaux (LCPTS)
Abstract: The prosody of a speech signal is related to many
factors: the social and geographical origin of the speaker, his or
her emotional state, his physiological state (weariness, sickness,
…) and the type of the sentence (interrogative, affirmative,
etc.).
A good synthesis or speech transformation system must
account for all of these factors in order to produce a speech that
sounds natural. In this paper, we propose a graphical interface
for the modification of the prosodic features of the speech
signal (the melodic curve - fundamental frequency and
temporal organization of the syllables - and the formantic
trajectories) using the PSOLA algorithm. The interface allows
the user to manually introduce the desired trajectories of the
transformation parameters of the PSOLA algorithm in order to
produce a transformed signal which has the desired prosody.
The results are acceptable, especially for the modification of
the fundamental frequency and of the temporal organization of
the source signal.
Keywords: speech synthesis, PSOLA algorithm, PSOLA
parameters, prosodic parameters,
ID05
21
Une approche Statistico-Linguistique pour l’extraction
de concepts sémantiques: Une première étape vers un
système générique de dialogue Homme-Machine
Mohamed Lichouri, Amar Djeradi and Rachida Djeradi
Signaux (LCPTS)
Résumé : Dans le présent papier, nous proposons la mise en
oeuvre d’un système de Compréhension Automatique de
l’énoncé en Communication Homme-Machine. L’architecture
que nous adoptons est basée sur une approche StatisticoLinguistique qui suppose que la compréhension d’un énoncé
n’est autre qu’une traduction du langage naturel vers le langage
conceptuel. Pour cela, nous présentons une nouvelle méthode
d’extraction de concepts sémantiques basée sur la technique de
recherche documentaire LSA. La méthode proposée a été
validée sur une plate-forme offrant des renseignements de
gestion de la scolarité universitaire (interrogation d’une base de
données) en considérant une entrée textuelle. Cette méthode a
achevée un taux de reconnaissance (extraction de concept) de
l’ordre de 51.85%, ainsi un taux de compréhension de l’ordre
de 73.33%. En plus de ça, afin de concevoir un modèle
générique de compréhension automatique de l’énoncé, nous
avons considéré un autre domaine d’application qui est
l’Extraction d’Information.
Keywords: Compréhension, énoncé, Communication, HommeMachine, Statistico-Linguistique, Conceptuelle.
ID17
22
Automatic Classification of Human Body
Postures Based on the Truncated SVD
Nabil Zerrouki and Amrane Houacine
Signaux (LCPTS)
Abstract: In this experimental study, we propose to use
Singular Value Decomposition (SVD) coefficients as features
to automatically classify human body postures. The
classification process uses images extracted from a fixed
camera video. A background subtraction technique is applied
for human body segmentation. A truncated SVD is performed
by selecting significant magnitude coefficients. And the heightwidth ratio of the human body is also included in the set of
features. The classification is then performed using an Artificial
Neural Network (ANN). Four body postures are considered in
our experiments, namely: standing, bending, sitting, and lying.
For evaluation, different state of the art statistical measures
have been considered. All of these evaluation measures
demonstrate that the proposed approach provides a significant
recognition rate compared to well-known classification
methods.
Keywords: Human body postures, classification, SVD
coefficients, neural network.
ID25
23
Speech Enhancement using Kalman Filter
Asma Bouchair and Abderrahmane Amrouche
Signaux (LCPTS)
Abstract: In this paper, we restrict in the use of the Kalman
filter to enhance speech signal degraded by noisy environment,
the complexity of the enhancement task resides in the nature of
the speech signal and the type of noise. In the past three
decades, interest has focused on developing methods more
efficient and rigorous in eliminating noise while preserving the
quality and intelligibility of the speech signal. For that we have
described two types of Kalman filter to enhance the speech
signal: The Kalman filter without noise modeling and the
Kalman filter with noise modeling.
Keywords: speech enhancement; noisy environment; Kalman
filter.
Session orale 8
Classification et Reconnaissance de Formes 2
ID16
24
Co-occurrence Matrix for Ancient Degraded
Document Image Binarization
A.Sehad, Y.Chibani, Mohamed Cheriet and .Yaddaden
Signaux (LCPTS)
Abstract: In this paper, we present a promising method for
binarization of historical and degraded document images, based
on co-occurrence matrix. The proposed method is computed by
using a threshold based on co-occurrence matrix descriptor.
The proposed method is tested objectively, using DIBCO
dataset degraded documents and compared against the classical
methods. The results are satisfactory and promising.
Keywords: binarization; degraded document; texture;
threshold.
ID22
25
Combination of Off-Line and On-line Signature
Verification Systems Based on SVM and DST
Nassim Abbas and Youcef Chibani
Signaux (LCPTS)
Abstract: The objective of this work is to present a signature
verification system based on combination of off-line and online systems
for managing conflict provided by the Support
Vector Machine (SVM) classifiers. This system is basically divided into
three parts: i) off-line verification stage, ii) on-line verification stage
and iii) combination module using
Dempster-Shafer theory (DST). The proposed framework allows
combining the normalized SVM outputs and uses an estimation
technique based on the dissonant model of Appriou to compute the
belief assignments. Combination is performed using Dempster-Shafer
(DS) rule followed by the likelihood ratio based decision making.
Experiments are conducted on the well known NISDCC signature
collection using false rejection and false acceptance criteria. The
obtained results show that the proposed combination framework using
DST yields the best verification accuracy compared to the sum rule
even when individual off-line and on-line classifications provide
conflicting results.
Keywords: Off-line signature verification; On-line signature
verification; Support Vector Machines; Dempster-Shafer
theory; belief assignments; conflict.
ID24
26
A Feature Generation Method Based on the
Contourlet Transform and Normalized
Energies for Offline Handwritten Signature
Verification
Assia Hamadene and Youcef Chibani
Signaux (LCPTS)
Abstract: We propose in this paper a new feature generation
method based on the contourlet transform (CT) for offline
handwritten signature verification (HSV). The CT allows
capturing segments of the handwritten signature contours on
each direction. In order to generate a reduced size of feature
vector, we use normalized energies computed on each
direction.
Experiments are conducted on the well known CEDAR dataset
using the support vector machines (SVM) classifier. The
obtained results show that the proposed approach is more
effective compared to the state of the art.
Keywords: contourlet transform; offline handwritten sifnature
verification;support vector machines;normalyzed enengy.
ID28
27
Dynamic One-Class Classifier Combination
for Multi-Class Classification
Bilal Hadjadji, Youcef Chibani and yasmine Guerbai
Signaux (LCPTS)
Abstract: The One-Class Classifier (OCC) has been widely used for
solving the one-class and multi-class classification problems. Its
main advantage for multi-class is offering an open system and
therefore allows easily extending new classes without retraining
OCCs. However, extending the OCC to the multi-class classification
achieves less accuracy comparatively to other multiclass classifiers.
Hence, in order to improve the accuracy and keep the offered
advantage we propose in this paper a Multiple Classifier System
(MCS), which is composed of different types of OCC. Usually, the
combination is performed using fixed or trained rules. Generally, the
static weighted average is considered as straightforward combination
rule. In this paper we propose a dynamic weighted average rule that
calculates the appropriate weights for each test sample. Experimental
results conducted on several real-world datasets proves the effective
use of the proposed multiple classifier system where the dynamic
weighted average rule achieves the best results for most datasets
versus the mean, max, product and the static weighted average rules.
Keywords: one class classifier, multiple classifier system,
multi-class classification, dynamic weighted average rule.
Session orale 9
ID47
28
SVM- Based Voice Activity Detection for
Distributed Speech Recognition System
Azzedine Touazi and Mohamed Debyeche
Signaux (LCPTS)
Abstract: Voice Activity Detection (VAD) algorithms based
on machine learning techniques have shown competitive results
in the area of automatic speech recognition. This paper
describes a new approach of VAD based on Support Vector
Machines (SVM) for Distributed Speech Recognition (DSR)
system. In the proposed scheme, the speech and the non-speech
frames are detected from the compressed Mel Frequency
Cepstral Coefficients (MFCCs), at the back-end (e.g. server)
side, with the aim of improving the VAD performance and
reducing the compression bit-rate from the front-end side (e.g.
client). By using the trained SVM with polynomial kernel, the
SVM-based VAD show an encouraging detection results. The
classification task conducted from the Aurora-2 speech
database, using different noise conditions, illustrates
comparable VAD performance, with respect to ETSI Advanced
Front-End (ETSI-AFE) VAD algorithm.
Keywords: voice activity detection; support vector machines;
DSR system; mel frequency cepstral coefficients.
ID50
29
Combining Auxiliary and Cepstral Features
for Speech Recognition
Amrous Anissa Imen and Debyeche Mohamed
Signaux (LCPTS)
Abstract: This paper presents our work about the integration
of auxiliary features in Hidden Markov Model (HMM) based
automatic speech recognition systems. Our experiments were
done using the HTK Toolkit and two data bases, the
ARADIGIT corpus which is an isolated data base of Arabic
digits and the TIMIT corpus which is a continuous speech data
base.
The obtained results show that: while the integration of the
auxiliary features with the standard parameters by SI (Separate
Integration) strategy leads to small improvements in the two
test environments (clean and noisy), their integration by DI
(Direct Integration) strategy leads to significant improvements
of the recognition system performance in noisy environment.
Keywords: ASR system, HMM, MFCC, auxiliary features.
ID54
30
Parallel GMM Sub-bandspeaker
Identification over AMR-NB Codec
Saboune Hadjer and Debyeche Mohamed
Signaux (LCPTS)
Abstract: The main goal of this work is to improve speaker
recognition performance in mobile Communications. For this
purpose, we investigate an exploratory paradigm named Multiband speaker recognition. The general principle of this concept
is to split the whole frequency domain into several sub-bands
on which statistical sub-recognizers based on Gaussian mixture
models (GMM) are independently applied and then recombined
to yield a global score and a global recognition decision. The
choice of the sub-band architecture and the recombination
strategy are discussed.
This technique had been shown to be robust for speech
recognition when narrow band noise degradation occurs. We
verify this robustness for the speaker identification task in the
case of AMR-NB transcoded speech.
Keywords: Speaker Identification, Sub-band, AMR-NB,
GMM, fuzzy integral
Session orale 10
Télécommunications Codage et Indexation 3
ID48
31
Quantificateur Vectoriel à Divisions
Commutées Appliqué au Codage des
Paramètres ISF de la Parole en Large Bande
Cheraitia Salah Eddine and Bouzid Merouane
Signaux (LCPTS)
Abstract : Ce document présente une technique du codage
des paramètres Immittance de Fréquence (ISF) d’un codeur de
parole en large bande, il s’agit de l’Adaptive Multi Rate Wide
Band AMR-WB. Cette technique exploite la stabilité de la
direction du commutateur du célèbre codeur des paramètres
ISF, c’est le quantificateur vectoriel à divisions commutées
(SSVQ).
Le codeur proposé présente les mêmes performances du codeur
SSVQ avec un nombre moins de bits et un débit par trame
variable.
Mots clé : Codage efficace, Codeur de la parole, Prédiction
Linéaire, paramètres ISF.
ID49
32
Automatic Assessment of the Dysarthria
Severity Levels
Kamil Lahcene Kadi, Sid Ahmed Selouani, Bachir
Boudraa and Malika Boudraa
Signaux (LCPTS)
Abstract: In this paper, Linear Discrimination Analysis (LDA)
is combined with two approaches of automatic classification,
Gaussian Mixture Model (GMM) and Support
Vector Machine (SVM) to perform an automatic assessment of
dysarthric speech. The front-end processing uses a set of
prosodic features selected with LDA on the basis of their
discriminative ability. The Nemours database of American
dysarthric speakers is used throughout experiments. Results
show a best classification rate with LDA/SVM system of 93%
that was achieved over four severity levels of dysarthria: no
dysarthric L0, mild L1, severe L2 and severe L3. This tool can
help clinicians to assess dysarthria, can be used in remote
diagnosis and may reduce some of the costs associated with
subjective tests.
Keywords : Dysarthria, gmm, lda, nemours-database, prosodicfeatures, severity-level-assessment, SVM
ID51
33
Channel Estimation Using Superimposed
Sequences for MIMO RFID Systems
Khelladi Fouad, Metref Adel and Fergani Belkacem
Signaux (LCPTS)
Abstract: In this paper, we propose a new channel estimation
method for multiple-input multiple-output (MIMO) passive
radio frequency identification (RFID) systems. The basic idea
consists in superimposing pilot symbols on the tag information
data.Moreover,atthereaderside,aregularizedleastsquaresalgorith
m is applied on the preamble of the tag response to estimate the
channel matrix coefficients. We derive a solution for the
regularized optimization problem and the closed-form
expression of the mean square error (MSE) for the proposed
method.
Our approach reduces the total number of requests needed for
the identification of tags in the reading area and provides
robustness against pilot contamination effect. The performance
of the proposed approach is compared with the conventional
pilot estimation approach for different numbers of tags and
receiving antennas. Numerical results show acceptable
enhancements in terms of the mean square error (MSE) and the
total number of requests compared to the conventional pilot
method.
ID56
34
Tatouage d’Images par Insertion Aveugle et
Informée et Détection par Corrélation
Linéaire
Abdelhamid Saighi and Salim Chitroub
Signaux (LCPTS)
Résumé : L’article présente deux systèmes de tatouage
d’images. Le premier système ne fait pas référence à l’image
originale dans le processus d’insertion de la marque. La
détection se fait par corrélation linéaire.
Le deuxième système utilise, par contre, l’image originale dans
la phase d’insertion de la marque. La détection se fait aussi par
corrélation linéaire. Les deux systèmes ont été testés face à
différentes attaques et leurs performances discutées et
comparées.
Session Poster 1
ID06
35
Euclidean Distance of Facial Edges for
Facial Expression Identification System
Kherchaoui Sonia* and Amrane Houacine**
Signaux (LCPTS)
Abstract: In this paper we present a facial expression
recognition system (FER). This system does the identification
and classification of the seven basic expressions: happy,
surprise, fear, disgust, sad, anger and a neutral state. The FER
system consists of three main parts. The first part is the
detection of the face and facial features to extract the face
centered region. Next step consists of a normalization of this
interest region and edge extraction. At this step we have a face
edge image that we use to calculate the Euclidean distance of
all pixels that constitute edges that make of our feature vector.
The third step is the classification of different emotional state
by the SVM method.
Keywords: facial expression identification; edge detection;
euclidian distance; support vector machine (SVM).
ID07
36
Tracking Characteristic Points on a
Talking Face
Noureddine Cherabit and Amar Djeradi
Signaux (LCPTS)
Abstract: Our paper presents a method of tracking
characteristic points on a speaking face and the reconstruction
of a face model from speaking descriptive vectors followed.
After capturing a video of a face speaking using a CCD camera,
facial expressions and the position of each characteristic point
on the face changes over time. The essential of this work is to
track this points using the LucasKanade method. Using
delaunay triangulation and descriptive vectors we reconstruct a
speaking face model.
Key words: talking head, tracking points, coding of visual
speech, Delaunay triangulation, feature point.
ID08
37
American Sign Language Recognition using
Probabilistic Neural Network
Khadidja Sadeddine , Rachida Djeradi and Fatma
zohra CHELALI
Signaux (LCPTS)
Abstract: The recognition of the signs that use the deaf
people to communicate is the most difficult than other kind of
gestures recognition. In this paper, we propose a probabilistic
neural network (PNN) for American Sign Language (ASL)
recognition for static signs realized with one hand. For that, we
use the Jochen Triesch dataset that contains ten static signs of
ASL with simple background. The PNN network trains sign
images in gray scale, their Discrete Cosine Transform (DCT)
features and transformed into binary. The PNN achieves a
recognition rate of 96%for both images in gray scale and DCT
features with an error MSE of 0.8%, and 90% for binary
images with MSE of 2% .
Keywords: Hand Gestures Recognition; Probabilistic Neural
Network; Sign Language Recognition.
ID09
38
Performances Comparison of ALOHA and
Tree Based Anti-Collision Algorithms in
RFID System
Mohamed Aissa Kalache and lamia Fergani
Signaux (LCPTS)
Abstract: In RFID systems, to resolve the problem of collision
between tags they are two main approaches of anti-collision
algorithm: ALOHA and Tree based algorithm. In this paper we
review several existing anti-collision algorithms and we
evaluate and compare their performances in terms of the total
interrogation cycles and the time required for identifying all the
Tags. It states that K-TAS shows the better performances in
this context.
Keywords: RFID; Tag; anti-collision; ALOHA; Tree.
ID10
39
Arabic Keyword Spotting in Ancient
Document Images Using Curvelet
Descriptors
Youcef Brik, Youcef Chibani, Bilal Hadjadji and EtTahir Zemouri
Signaux (LCPTS)
Abstract: This paper deals with the contribution of Curvelet
transform to generate more accurate word image descriptors for
Arabic keyword spotting in ancient documents. Due to its
properties, Curvelet can tolerate more scale distortions and
more directional features in images. The process of Curvelet
descriptor generation is applied to each word image in the
dataset. Therefore, dynamic time warping algorithm is
employed to match corresponding coefficients from Curvelet
descriptor matrices. Experimental results on ancient Arabic
document demonstrate that the characterization of the word
image from the Curvelet descriptors offers a significant
improvement in the word spotting system.
Keywords: feature generation, word spotting, ancient ocument,
curvelet transform, dynamic time warping.
ID12
40
Language Identification Based on High
Frequency Approaches
Kheireddine Abainia, Siham Ouamour and Halim Sayoud
Signaux (LCPTS)
Abstract: This paper deals with the problem of automatic
language identification of noisy texts, which represents an
important task in natural language processing. Actually, there exist
several works in this field, which are based on statistical and
machine learning approaches for different categories of
texts.Unfortunately, most of the proposed methods work fine on
clean texts or long texts, but often present a failure when the text is
corrupted or too short. In this research work, we use a typical
dataset consisting of short texts collected from several discussion
forums containing several types of noises. Our dataset contains 32
different languages; where we notice that some languages are quite
different while some others are too closed. In this investigation, we
propose two types of methods to identify the text language: termbased method and character-based method. Moreover, we propose
two hybrid methods to enhance the performances of those
techniques. Experiments show that the proposed hybrid methods
are quite interesting and present good language identification
performances in noisy texts.
Keywords: Natural Language Processing; Text categorization;
Automatic Language Identification; Noisy Text; Hybrid Approach.
ID14
41
The Effective Use of Writer-Independent
Handwritten Signature Verification Based
on One-Class SVM Classifier
Yasmine Guerbai, Youcef Chibani and Bilal Hadjadji
Signaux (LCPTS)
Abstract: The limited number of writers and the lack of
forgeries as counterexample to construct the systems is the
main difficulty task for designing a robust off-line Handwritten
Signature Verification System (HSVS). In this paper, we
propose to study the influence of writer’s number using
conjointly the curvelet transform and the One-Class Support
Vector Machine (OC-SVM), which takes in consideration only
genuine signatures. The design of the HSVS is based on the
writer-independent approach. Experimental results conducted
on the standard CEDAR and GPDS datasets demonstrate that
the proposed method allows achieving the lowest Average
Error Rate with a limited number of writers.
ID18
42
Historical Document Image Binarization
using Contourlet Transform
Et-Tahir Zemouri, Youcef Chibani and Youcef Brik
Signaux (LCPTS)
Abstract: in this paper we propose the use of the Contourlet
Transform for evaluating the quality of the degraded historical
document. We first improve the quality of the document image
using the Contourlet Transform. After reconstruction, we use
the local thresholding method for extracting the foreground
text. The proposed method is evaluated on the benchmarking
dataset used in the international document image binarization
contest (DIBCO
2009/2011 and H-DIBCO) according to the type of
degradation. Promising results are obtained comparatively to
the standard methods.
Keywords: historical document; binarization; enhancement;
contourlet transform; thresholding.
ID19
43
L’Acquisition Compressée dans les Systèmes
Mobiles
Houria Haneche*, Bachir Boudraa* and Abdeldjalil
Ouahabi**
*Laboratoire de Communication Parlée et de Traitement des
Signaux (LCPTS)
**Polytech de l’Université de TOURS, France
Résumé : L’acquisition compressée (Compressed Sensing ou
Compressive Sampling en anglais) est un nouveau paradigme
qui permet d’acquérir des signaux à un taux plus élevé.
Beaucoup de recherches étudient les applications possibles
dans les télécommunications.
Ce travail présente nos travaux préliminaires dans les systèmes
de communications mobiles, grâce aux algorithmes de
reconstruction et la parcimonie du signal la reconstruction est
possible et l’erreur de reconstruction est maintenue faible.
Mots clefs : Compressed Sensing, Parcimonie, Signal épars,
Systèmes mobiles, Minimisation l1, Optimisation convexe.
ID21
44
Digital Watermarking in Audio for
Copyright Protection
Mustapha Hemis and Bachir Boudraa
Signaux (LCPTS)
Abstract: Digital watermarking is a technique that consists in
hiding binary information within a signal in an imperceptibly
way. This technique plays an important role for copyright
protection of multimedia data. In this paper an efficient robust
audio watermarking algorithm based on double transforms is
introduced. In a first step, the original signal is decomposed by
discrete wavelet transform (DWT). Then the prominent
approximation coefficients are segmented in non-overlapping
2D blocks. Singular value decomposition (SVD) is applied on
each one. The watermark is embedded in the singular values
(SVs) for each block. Watermark extraction is non-blind and
it’s done by performing inverse operation of embedding
process. Experimental results show that this scheme provides
high robustness against common signal processing attacks. In
addition, this method uses a double key for insertion and
extraction, making it suitable for secure application such as
copyright protection.
Index Terms: Multimedia security, audio watermarking,
discrete wavelet transform, singular value decomposition.
ID26
45
Utilisation des Systèmes 3D pour la
Détection et Quantification des Mouvements
au Cours des Examens Scintigraphiques
Sofiane Tchoketch Kebir and Slimane Mekaoui
Signaux (LCPTS)
Résumé : La Scintigraphie est l’une des techniques de l’imagerie
médicale qui consiste à produire des images médicales par émission de
photons produit par des radio éléments tel que l’iode 131 ou le
Technétium et captés par des caméras pour caractériser certaines
pathologies cancéreuses. Les images fournies par ces caméras doivent
se rapprocher le plus possible de la réalité. Plusieurs facteurs peuvent
interférer sur les images acquises, notamment les mouvements
volontaires ou involontaires des patients pendant l’examen
Scintigraphique. Dans ce papier nous proposons une approche de
détection et de suivi du mouvement des patients durant l’acquisition des
images Scintigraphiques, en exploitant des caméras RGB-D.
Cette approche, nous permet d’opérer après avoir déterminé les
caractéristiques globales du mouvement, des corrections sur les images
acquises. Ces corrections sont alors applicables aussi bien au cas du
dédoublement des contours qu’à celui des aspects flous.
Mots clés : Imagerie médicale, traitement d’image, quantification de
mouvement, modélisation et suivi de mouvement, cameras RGB-D,
Session Poster 2
ID27
46
Contribution à la Réalisation d’un Système
TTS pour la Langue Arabe
Aissa Amrouche, leila Falek and Hocine Teffahi
Signaux (LCPTS)
Résumé : Par ses propriétés morphologiques et syntaxiques la
langue arabe est considérée comme une langue très difficile à
maîtriser dans le domaine du traitement automatique de la
parole et les systèmes de synthèse à partir du texte arabe sont
donc très peu nombreux. Le but de notre travail est une
réalisation d’un système de synthèse de la parole par
concaténation (sélection) dynamique d’unités dans un corpus
pour la langue arabe baptisé GArabic TTS sous
l’environnement Matlab. Le texte à introduire est un texte non
voyellé qui facilite l’utilisation du système, la sortie est
disponible uniquement pour une voix masculine. L’évaluation
de GArabic TTS est basée sur un test subjectif et objectif; en ce
qui concerne l’intelligibilité, aspects naturels (l’écoute) et la
qualité (PESQ). L’évaluation finale de la qualité globale du
système est jugée satisfaite.
ID31
47
Improved Packet Loss Recovery using
Multiple Description
Coding in MELP Coder for Voice over IP
Mohammed Saidi, Leila falek, Bachir Boudraa and
Hocine Teffahi
Signaux (LCPTS)
Abstract: In VoIP systems, CELP coders, such as G.729, are
commonly used to obtain a good speech quality in the absence
of packet losses. However, harmonic coders such as MELP
may be a good alternative for VoIP due to their higher
resilience to packet losses. In this work we have proposed a
VoIP application with two MELP coders supported by
packetization scheme based on Multiple Description Coding
(MDC). A packet will contain information two MELP coders
operating at 2.4 and 1.2Kbps respectively. The packetization is
achieved using 135 bits in 22.5 ms corresponding to a total rate
of 6 kbps. The results show that under typical VoIP operating
conditions, the method performs well and outperforms CELP
coders operating without MDC at 8 kbps.
Keywords: VoIP, MELP, MDC, packet loss
ID33
48
Transmission Sécurisée par le Système PGP
en vue de Protéger les Courriers
Electroniques
Yacine Bouflih and Salim Chitroub
Signaux (LCPTS)
Résumé : Dans ce travail, nous présentons un système de
transmission sécurisée en utilisant le système PGP «Pretty
Good Privacy» dans le but d’offrir un moyen de préserver la
confidentialité des messages du courrier électronique qui
circulent dans un canal de transmission donné. C’est un outil
qui utilise une technique de chiffrement hybride basée sur le
mode de chiffrement symétrique et le mode de chiffrement
asymétrique pour le cryptage et le décryptage des données.
Cette hybridation constitue un système de chiffrement très
solide et rapide du fait que les attaquants, ou la cryptanalyse,
trouvent des difficultés pour casser l’algorithme du cryptage.
Mots clés : Cryptographie; Chiffrement; Signature numérique;
Empreinte électronique ; PGP ; Temps de calcul ; Espace
mémoire.
ID35
49
Using the Prosodic Features and Formant for
Speech Recognition System over Mobile Network
Lallouani Bouchakour and Debyeche Mohamed
Signaux (LCPTS)
Abstract: This paper investigates the contribution of formants and
prosodic features like pitch and energy on automatic speech
recognition system performance in mobile networks especially the
GSMEFR (Global System for Mobile Enhanced Full Rate) codec. The
front-end of the speech recognition system combines feature extracted
by converting the quantized spectral information of speech coder,
prosodic information and formant frequencies. The quantized spectral
information is represented by the LPC (Linear Predictive Coding)
coefficients, the LSF (Line Spectral Frequencies) coefficients, the
approximation of the LSF's to the LPC Cepstral Coefficients
(LPCC’s) that are the Pseudo Cepstral Coefficients (PCC) and the
Pseudo-Cepstrum (PCEP) coefficients. The achieved speakerindependent speech recognition system is based on Continuous
Hidden Markov Model (CHMMs) classifier. The obtained results
show that the resulting multivariate feature vectors lead to a
significant improvement of the speech recognition system
performance in mobile environment, compared to speech coder bitstream system alone.
Keywords: ASR, GSMEFR, CHMM, ARADIGIT, bit-stream,
Formant, Pitch.
ID37
50
Etude Comparative de Méthodes de
Correction de Pertes de Paquets en
Transmission de la Voix sur IP (VoIP)
Hamza Kheddar and Bachir Boudraa
Signaux (LCPTS)
Résumé : Le travail consiste à mettre au point sous
MATLAB deux codeurs MELP fonctionnant respectivement à
1.2 kbit/s et à 2.4 kbit/s. Nous apportons ensuite des
améliorations à ces codeurs par l’implémentation de techniques
de dissimulation des trames perdues basées sur le récepteur.
Ces techniques consistent en l’entrelacement des trames
d’information. A cet effet, nous avons d’abord effectué une
étude comparative de plusieurs méthodes d’entrelacement. Par
la suite, nous avons étendu cette comparaison à une méthode
déjà mise au point au sein de notre laboratoire, intitulée codage
par descriptions multiples (MDC). Pour cela, nous avons utilisé
la technique d’évaluation normalisée par ITU-T et appelée
PESQ (Perceptual Evaluation of Speech Quality).
Mots clés : Entrelacement, Codage MELP, MDC, PLC, PES
ID41
51
A Novel Voice Activity Detection Approach
for Automatic Speaker Verification
Meriem Fedila, Abderrahmane Amrouche and
Messaoud Bengherbi
Signaux (LCPTS)
Abstract: The main goal of this paper is to propose a novel
voice activity detector (VAD) approach for Speaker
Verification applications which is both faster and robust to
noise environment. The idea is to use Support Vector Machine
(SVM) parameters estimated using speech and non speech of a
long utterance to compute a function decision of the proposed
VAD approach. Speaker Verification results on TIMIT
database show that the proposed VAD approach greatly
improves Verification speed in features extraction step. for
processing frames of female recording, a time of 0.0015 second
was achieved using SVM-based VAD, which is nearly six
times smaller than the recently proposed self adaptive VAD
method. Further simplicity which results by a speed-up process,
our VAD also possesses a satisfying degree of robustness
against noise.
Index Terms: Voice Activity Detection, Automatic Speaker
Verification, Real-time, Support Vector Machine.
ID42
52
A New Proposed A5/1 Stream Cipher to
Improve GSM Security
Mahdi Madani and Salim Chitroub
Signaux (LCPTS)
Abstract: In all cellular systems, the confidentiality of the
user’s calls is based on cryptographic algorithms. And each
system is encrypted by its specific algorithm. For example, the
Global System for Mobile communication use the algorithm
named A5/1, the Universal Mobile for Telecommunication
System use another algorithm named A5/3. In this paper we
present developments in the security architecture of the A5/1
algorithm used by GSM network. But, we begin with a brief
description of the A5/1 original architecture, then listing the
most attacks from which it has cracked and then showing the
weaknesses from which suffer.
Index Terms: GSM Networks; Stream Cipher;Cryptographic
Algorithm; Mobile Security;; Cryptanalysis; Non-linear
Boolean ,Functions; Linear Feedback Shift Register.
ID45
53
Contribution à l’Implantation
d’Algorithmes de Traitement de la Parole
sur Circuit FPGA
Halilali Abderrezak and Houacine Amrane
Signaux (LCPTS)
Résumé : Certaines applications de traitement du signal de
parole, telles que la transformation, la reconnaissance ou la
synthèse, peuvent nécessiter un traitement temps réel.
L’utilisation de circuits spécifiques peut être la solution à ce
genre de problèmes.
Le travail présenté propose une méthodologie basée sur
l’utilisation d’un circuit FPGA pour l’implantation d’une
application de transformation de la parole.
Mots clés : Parole, VHDL, FPGA, audio codec.
ID52
54
SVM Modeling for Handling Imbalanced
Human Activity Datasets
Yala Nawel and Fergani Belkacem
Signaux (LCPTS)
Abstract: Activity recognition datasets are generally
imbalanced, meaning certain activities occur more frequently
than others. Not incorporating this class imbalance results in an
evaluation that may lead to disastrous consequences for elderly
persons. In this work, we evaluate various types of resampling
methods: at algorithmic level using CS-SVM and at data level
using OS-CSVM combined with the discriminative classifier
named Soft-Margin Support Vector Machines (CSVM) in order
to handle imbalanced data problem. We conduct several
experiments using three real world activity recognition datasets
and show that the OS-CSVM are able to surpass CSVM and
CSSVM.
OS-CSVM is better than CS-CSVM for classifying the
activities using binary and ubiquitous sensors.
Keywords: Activity Recognition, SVM, Cost Sensitive
Learning, Machine Learning, Imbalanced Data.
ID57
55
Authentification Vocale d’un Locuteur
Arabophone en Vue de la Criminalistique
Ouassila Kenai and Mhania Guerti
Signaux (LCPTS)
Résumé : Dans cet article, nous nous intéressons à
l’Authentification Vocale des Locuteurs Arabophones en vue de la
Criminalistique ou Forensique (AVC) en mode indépendant du
texte. A partir des signaux vocaux de ces locuteurs des
informations relatives à leurs identités sont extraites par l’analyse
cepstrale MFCC (Mel Frequency Cepstral Coefficients) avec ces
derniers en estimant des modèles GMM (Gaussian Mixture
Models) de locuteurs robustes.Ainsi une trace vocale, peut être
analysée et par la suite comparée avec les GMM en appliquant
l’approche Bayésienne, Like-Lihood Ratio (LLR), afin de
permettre son authentification. Nos expériences réalisées sur le
système AVC montrent qu’un GMM composé de 32 Gaussiennes
est largement suffisant pour représenter la distribution des vecteurs
d’un seul locuteur (qui est le criminel) ainsi que le matériel
d’enregistrements qui donne de meilleures performances de ce
système élaboré.En effet, nous avons obtenu des résultats
satisfaisants. Ceux-ci peuvent aider la justice à prendre une
décision afin de résoudre des problèmes criminalistiques.
Mots clés : Authentification Vocale Criminalistique AVC,
MFCC, GMM et Approche LLR.

14-15 Janvier 2015 - lcpts - Université des Sciences et de la

Transcription

Documents pareils

voir le CV

curriculum vitae - Ordre des experts internationaux

Session Septembre 2016/2017

Traitement Automatique des Chèques Postaux Algériens

Houari-Ben-Slimane_vente-vehicule-occasion

GUEMRAS Nadjet

Participation à la foire Batimatec du 3 au 8 Mai 07 à Alger, Algérie

Le fonctionnement du GSM

Fiche de Participation

L`optimisation dans les réseaux de transport et de distribution