Dans cette thèse, nous avons proposé une approche analytique multi-échelle pour le word spotting dans les documents manuscrits. Le modèle proposé fonctionne selon deux niveaux différents. Un module de filtrage global permettant de définir plusieurs zones candidates de la requête dans le document testé. Ensuite, l’échelle de l’observation est modifiée à un niveau inférieur afin d’affiner les résultats et sélectionner uniquement ceux qui sont vraiment pertinents. Cette approche de word spotting est basée sur des familles généralisées de filtres de Haar qui s’adaptent à chaque requête pour procéder au processus de spotting et aussi sur un principe de vote qui permet de choisir l’emplacement spatial où les réponses générées par les filtres sont accumulées. Nous avons en plus proposé une autre approche pour l’extraction de texte du graphique dans les bandes dessinées. Cette approche se base essentiellement sur les caractéristiques pseudo-Haar qui sont générées par l’application des filtres généralisés de Haar sur l’image de bande dessinée. Cette approche est une approche analytique et ne nécessite aucun processus d’extraction ni des bulles ni d’autres composants. / The presented thesis follows two directions. The first one disposes a technique for text and graphic separation in comics. The second one points out a learning free segmentation free word spotting framework based on the query-by-string problem for manuscript documents. The two approaches are based on human perception characteristics. Indeed, they were inspired by several characteristics of human vision such as the Preattentive processing. These characteristics guide us to introduce two multi scale approaches for two different document analysis tasks which are text extraction from comics and word spotting in manuscript document. These two approaches are based on applying generalized Haar-like filters globally on each document image whatever its type. Describing and detailing the use of such features throughout this thesis, we offer the researches of document image analysis field a new line of research that has to be more explored in future. The two approaches are layout segmentation free and the generalized Haar-like filters are applied globally on the image. Moreover, no binarization step of the processed document is done in order to avoid losing data that may influence the accuracy of the two frameworks. Indeed, any learning step is performed. Thus, we avoid the process of extraction features a priori which will be performed automatically, taking into consideration the different characteristics of the documents.

In this dissertation, three research directions were explored to alleviate two major issues, i.e., the use of incorrect models and training/test condition mismatches, in the modeling frameworks of modern spoken keyword search (KWS) systems. Each of the three research directions, which include (i) data-efficient training processes, (ii) system optimization objectives, and (iii) data augmentation, utilizes different types and amounts of training resources in different ways to ameliorate the two issues of acoustic and language modeling in modern KWS systems. To be more specific, resource-dependent keyword modeling, keyword-boosted sMBR (state-level minimum Bayes risk) training, and multilingual acoustic modeling are proposed and investigated for acoustic modeling in this research. For language modeling, keyword-aware language modeling, discriminative keyword-aware language modeling, and web text augmented language modeling are presented and discussed. The dissertation provides a comprehensive collection of solutions and strategies to the acoustic and language modeling problems in KWS. It also offers insights into the realization of good-performance KWS systems. Experimental results show that the data-efficient training process and data augmentation are the two directions providing the most prominent performance improvement for KWS systems. While modifying system optimization objectives provides smaller yet consistent performance enhancement in KWS systems with different configurations. The effects of the proposed acoustic and language modeling approaches in the three directions are also shown to be additive and can be combined to further improve the overall KWS system performance.

We propose subword spotting, a generalization of word spotting where the search is for groups of characters within words. We present a method for performing subword spotting based on state-of-the-art word spotting techniques and evaluate its performance at three granularitires (unigrams, bigrams and trigrams) on two datasets. We demonstrate three applications of subword spotting, though others may exist. The first is assisting human transcribers identify unrecognized characters by locating them in other words. The second is searching for suffixes directly in word images (suffix spotting). And the third is computer assisted transcription (semi-automated transcription). We investigate several variations of computer assisted transcription using subword spotting, but none achieve transcription speeds above manual transcription. We investigate the causes.

We propose subword spotting, a generalization of word spotting where the search is for groups of characters within words. We present a method for performing subword spotting based on state-of-the-art word spotting techniques and evaluate its performance at three granularitires (unigrams, bigrams and trigrams) on two datasets.We demonstrate three applications of subword spotting, though others may exist. The first is assisting human transcribers identify unrecognized characters by locating them in other words. The second is searching for suffixes directly in word images (suffix spotting). And the third is computer assisted transcription (semi-automated transcription). We investigate several variations of computer assisted transcription using subword spotting, but none achieve transcription speeds above manual transcription. We investigate the causes.

This thesis addresses the problems of phonemic variability and confusability from the pronunciation modeling perspective for an automatic speech recognition (ASR) system. In particular, several research directions are investigated. First, automatic grapheme-to- phoneme (g2p) and phoneme-to-phoneme (p2p) converters are developed that generate alternative pronunciations for in-vocabulary as well as out-of-vocabulary (OOV) terms. Since the addition of alternative pronunciation may introduce homophones (or close homophones), there is an increase of the confusability of the system. A novel measure of this confusability is proposed to analyze it and study its relation with the ASR performance. This pronunciation confusability is higher if pronunciation probabilities are not provided and can potentially severely degrade the ASR performance. It should, thus, be taken into account during pronunciation generation. Discriminative training approaches are, then, investigated to train the weights of a phoneme confusion model that allows alternative ways of pronouncing a term counterbalancing the phonemic confusability problem. The objective function to optimize is chosen to correspond to the performance measure of the particular task. In this thesis, two tasks are investigated, the ASR task and the KeywordSpotting (KWS) task. For ASR, an objective that minimizes the phoneme error rate is adopted. For experiments conducted on KWS, the Figure of Merit (FOM), a KWS performance measure, is directly maximized.

A Mandarin keyword spotting system based on MFCC, discrete-time HMM and Viterbi algorithm with DTW is proposed in this thesis. Joining with a dialogue system, this keyword spotting platform is further refined to a prototype of natural speech patient registration system of Kaohsiung Veterans General Hospital. After the ID number is asked by the computer-dialogue attendant in the registration process, the user can finish all relevant works in one sentence. Functions of searching clinical doctors, making and canceling registration are all built in this system. In a laboratory environment, the correct rate of this speaker-independent patient registration system can reach 97% and all registration process can be completed within 75 seconds.

While significant progress has been made in automatic speech recognition (ASR) during the last few decades, recognizing and understanding unconstrained conversational speech remains a challenging problem. In this dissertation, five methods/systems are proposed towards a robust conversational speech recognition and understanding system. I. A non-uniform minimum classification error (MCE) approach is proposed which can achieve consistent and significant keyword spotting performance gains on both English and Mandarin large-scale spontaneous conversational speech tasks (Switchboard and HKUST Mandarin CTS). II. A hybrid recurrent DNN-HMM system is proposed for robust acoustic modeling and a new way of backpropagation through time (BPTT) is introduced. The proposed system achieves state-of-the-art performances on two benchmark datasets, the 2nd CHiME challenge (track 2) and Aurora-4, without front-end preprocessing, speaker adaptive training or multiple decoding passes. III. To study the specific case of conversational speech recognition in the presence of competing talkers, several multi-style training setups of DNNs are investigated and a joint decoder operating on multi-talker speech is introduced. The proposed combined system improves upon the previous state-of-the-art IBM superhuman system by 2.8% absolute on the 2006 speech separation challenge dataset. IV. Latent semantic rational kernels (LSRKs) are proposed for spotting the semantic notions on conversational speech. The proposed framework is generalized using tf-idf weighting, latent semantic analysis, WordNet, probabilistic topic models and neural network learned representations and is shown to achieve substantial topic spotting performance gains on two conversational speech tasks, Switchboard and AT&T HMIHY initial collection. V. Non-uniform sequential discriminative training (DT) of DNNs with LSRKs is proposed which directly links the information of the proposed LSRK framework to the objective function of the DT. The experimental results on the subset of Switchboard show the proposed method can lead the acoustic modeling to a more robust system with respect to the semantic decoder.

Thesis (MSc )--University of Stellenbosch, 2001. / ENGLISH ABSTRACT: Washing and cleansing in an aqueous washing solution is a complex process involving the interaction of numerous physical and chemical processes. Investigations indicate that various factors present during repeated laundering might have an adverse effect on textile fabrics. Little is known, however, about the effect of the different mechanical wash actions on the deterioration of textiles. The automatic washing machines commonly used in South African households are often broadly classified as top loaders and front loaders. On closer inspection, however, it becomes clear that marked differences exist between the horizontal drum machines (H), vertical drum agitator type (V(A)) and vertical drum impeller type machines (V(/)). The effect of laundering on the wear life of textile fabrics has been extensively investigated, but a serious need for research on the effect of the different mechanical washing actions of the different washing machines on textile fabrics exists on both national and international level. A number of factors have an effect on textile deterioration during washing, e.g. number of washes, temperature of wash liquid, detergent type as well as fabric finishes and water quality. Fabric properties like changes in tensile strength, print deterioration and fraying, as well as electron microscope photographs, were used as indicators of mechanical deterioration to textiles. The broad aim of this exploratory study was firstly, to compare the effect of repeated washing in different domestic automatic washing machines (commonly used in South Africa) on the possible mechanical damage to textile fabrics and, secondly, to compare the soil removal efficiency of the mechanical wash actions of the different machines. An experimental study was therefore decided on. The washing of the test fabrics, as well as the testing, was carried out under controlled conditions in a laboratory. Apart from the different mechanical wash actions of the individual machines, the effect of the variables wash temperature, level of detergent and number of washes, was also investigated. This was done by measuring tensile strength, print deterioration and fraying propensity on samples laundered repeatedly at 40°C or 60°C, with and without detergent in the washing liquid, and comparing the colour change measured on laboratory-soiled test fabrics after washing. Analysis of the mechanical wash actions of the seven washing machines compared in this study indicated conspicuous differences among machines. It can be concluded that the mechanical wash action in the category V(A) machines is the most severe, and causes the highest reduction in tensile strength, the greatest print deterioration and the highest degree of fraying. These machines did not, however, exhibit greater soil removal efficiency than the other two categories of washing machines. The machines from category H did not exhibit a significantly different soil removal efficiency than those from category V(A), but their mechanical wash action proved to be less severe. Category V(/) machines seem to have the most delicate wash action and will probably cause the slightest fabric deterioration over the long term, but unfortunately produces poor soil removal efficiency results. This study also confirmed that washing fabrics in water alone causes more deterioration of tensile strength in fabrics than washing with detergent in the wash solution. Due to the limited scope and exploratory nature of this research/study, certain aspects of automatic washing machine processes could not be tested empirically. Recommendations for future research and implications for consumers were formulated. / AFRIKAANSE OPSOMMING: Was en skoonmaak in 'n waterige oplossing is 'n komplekse proses wat 'n interaksie tussen verskeie fisiese en chemiese prosesse behels. Ondersoeke dui aan dat verskeie van die faktore teenwoordig tydens herhaalde wasprosesse 'n nadelige effek op tekstielstowwe mag he. Baie min is egter bekend oor die effek van die verskillende meganiese wasaksies op die degenerasie van tekstiele. Die outomatiese wasmasjiene wat algemeen in Suid-Afrikaanse huishoudings gebruik word, word dikwels breedweg geklassifiseer as bolaaiers en voorlaaiers. By nadere ondersoek word dit egter duidelik dat daar opvallende verskille bestaan tussen die horisontale drom masjiene (H), die vertikale drom roerder tipe ((V)A)) en die vertikale drom stuwer tipe (V(/)) masjiene. Die effek van was op die draleeftyd van tekstielstowwe is uitvoerig ondersoek, maar daar bestaan op beide nasionale en internasionale vlak In ernstige behoefte aan navorsing oor die effek van die verskillende meganiese wasaksies van die verskillende wasmasjiene op tekstielstowwe. Faktore 5005 die aantal wasse, temperatuur van die wasoplossing, detergent tipe, water kwaliteit en tekstielstoftipe en -afwerking(s) het In effek of tekstieldegenerasie gedurende was. Tekstielstofeienskappe 5005 veranderinge in die breeksterkte van die tekstielstof of die tekstielstofmassa per eenheidsarea, agteruitgang van bedrukking, rafeling, sowel as elektronmikroskoop foto's word as aanduiding van meganiese degenerasie t.o.v. tekstiele beskou. Die oorkoepelende doe I van hierdie eksploratiewe studie was eerstens, om die effek te vergelyk van herhaalde was in verskillende outomatiese wasmasjiene (algemeen in gebruik in Suid-Afrika) op die moontlike meganiese beskadiging aan tekstielstowwe, en tweedens, om die doeltreffendheid van vuilverwydering op die wasaksies van die verskillende masjiene te vergelyk. Daar is gevolglik op In eksperimentele studie besluit. Die was van die tekstielstowwe, sowel as die laboratoriumtoetse, is onder gekontroleerde atmosferiese toestande in In laboratorium uitgevoer. Afgesien van die verskillende meganiese wasaksies van die individuele masjiene, is die effek van die veranderlikes westemperetuur, detergentv!ak en aanta! wasse ook ondersoek. Dit is gedoen deur die bepaling van breeksterkte en die meting van die mate van agteruitgang van bedrukking en mate van geneigdheid tot rafeling van kledingstowwe na herhaalde was by 40°C en 60°C, met en sonder detergent in die wasvloeistof. Die kleurverandering na was is ook gemeet op laboratorium gevlekte monsters. Die analise van die meganiese wasaksie van die sewe wasmasjiene wat in hierdie studie vergelyk is, het opvallende verskille tussen masjiene uitgewys. Daar kan tot die gevolgtrekking gekom word dat die meganiese wasaksie in kategorie V(A) masjiene die ergste is, wat gevolglik die grootste verlaging in breeksterkte en die meeste afieruitgang van bedrukking en rafeling veroorsaak het. Hierdie masjiene het egter nie meer doeltreffende vuilverwydering tot gevolg gehad in vergelyking met die ander kategoriee van wasmasjiene nie. Die masjiene in kategorie H het nie 'n betekenisvolle verskil in doeltreffenheid van vuilverwydering van die van kategorie V(A) getoon nie, maar die meganiese wasaksie was minder straf. Kategorie V(/) masjiene het blykbaar die mees delikate wasaksie en sal moontlik die minste tekstielstof degenerasie oor die langtermyn toon, maar ongelukkig is die resultate t.o. v. die doeltreffendheid van vuilverwydering swak. Die studie bevestig dat die was van tekstielstowwe in water aileen 'n groter verlaging in breeksterkte van tekstielstowwe veroorsaak as wat die geval is wanneer met 'n detergent in die wasvloeistof gewas word. Weens die beperkte omvang en eksploratiewe aard van die studie kon sekere aspekte van die outomatiese wasprosesse nie empiries getoets word nie. Aanbevelings vir toekomstige navorsing en implikasies vir verbruikers is geformuleer.

L’occurrence en service commercial de fissures macroscopiques dans certains disques de frein TGV a pu être reliée au type de garniture utilisé. L’objectif de cette thèse est de comprendre cette relation, d’identifier les paramètres d’influence et de proposer des voies d’amélioration pour la conception de garnitures à risque de fissuration réduit. Le comportement thermique de quatre couples disque-garnitures est d’abord analysé par le biais d’une campagne expérimentale de freinage originale. Elle met en évidence différents types de localisations thermiques et permet d’identifier des signatures thermiques caractéristiques des garnitures. Le lien entre localisations thermiques et risques de fissuration est ensuite établi à l’aide d’une modélisation thermomécanique. Des indicateurs tenant compte des caractéristiques spatiales et temporelles des localisations thermiques ainsi que des niveaux de température atteints sont proposés. Ils permettent de classer les garnitures testées dans un graphe de « criticité ». Enfin, une étude d’influence des caractéristiques mécaniques et thermiques des garnitures sur les localisations engendrées permet de dégager des préconisations et des voies d’amélioration pour la conception de nouvelles garnitures. La caractérisation expérimentale du comportement de deux nouvelles garnitures, l’une s’approchant le plus des préconisations faites, l’autre s’en éloignant fortement, montre la pertinence de l’approche développée et la validité des préconisations / The occurrence of macroscopic cracks in some TGV brake discs in commercial service has been linked to the pad type used. The objective of this thesis is to understand this relationship, to identify sensitive parameters and to propose guidelines to the design of pads reducing the risk of cracking.The thermal behavior of four disc-pad couples is first analyzed through an original experimental campaign of braking. It highlights various types of thermal localisations and enables to identify thermal signatures characteristic of the various pads. The relationship between thermal localisation and risk of cracking is then determined using thermomechanical modeling. Some indicators taking into account spatial and temporal characteristics of thermal localisations and temperature levels achieved are proposed. They enable to classify the tested pads in a graph of "criticity". Finally, a study of the influence of mechanical and thermal properties of pads on the thermal localisations occurrence enables to propose guidelines and improving ways for the design of new pads. An experimental characterization of the behavior of two new pads, one very close to the given recommendations, the other strongly away from them, shows the relevance of the approach and the validity of the present recommendations

This thesis is aimed on design keyword detector. The work contains a description of the methods that are used for these purposes and design of algorithm for keyword detection. The proposed detector is based on the method of DTW (Dynamic Time Warping). Analysis of the problem was performed on the module programmed in ANSI C, which was created within the thesis. The results of the detector were evaluated using the metrics WER (word error rate) and AUC (area under curve).