Dissertations / Theses on the topic 'And impressive and expressive speech'

Orientador: José Mario De Martino<br>Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Elétrica e de Computação<br>Made available in DSpace on 2018-08-26T21:43:57Z (GMT). No. of bitstreams: 1 Costa_PaulaDornhoferParo_D.pdf: 15894797 bytes, checksum: 194a20ae502dfc7198a008d576e23e4c (MD5) Previous issue date: 2015<br>Resumo: O desenvolvimento da tecnologia de animação facial busca atender uma demanda crescente por aplicações envolvendo assistentes, vendedores, tutores e apresentadores de notícias virtuais; personagens realistas de videogames, agentes sociais e ferramentas para experimentos científicos em psicologia e ciências comportamentais. Um aspecto relevante e desafiador no desenvolvimento de cabeças falantes, ou "talking heads", é a reprodução realista dos movimentos articulatórios da fala combinados aos elementos de comunicação não-verbal e de expressão de emoções. Este trabalho presenta uma metodologia de síntese de animação facial baseada em imagens, ou animação facial 2D, que permite a reprodução de uma ampla gama de estados emocionais de fala expressiva, além de suportar a modulação de movimentos da cabeça e o controle de elementos faciais tais como o piscar de olhos e o arqueamento de sobrancelhas. A síntese da animação utiliza uma base de imagens-protótipo que são processadas para obtenção dos quadros-chave da animação. Os pesos utilizados para a combinação das imagens-protótipo são derivados de um modelo estatístico de aparência e formas, construído a partir de um conjunto de imagens de treinamento extraídas de um corpus audiovisual de uma face real. A síntese das poses-chave é guiada pela transcrição fonética temporizada da fala a ser animada e pela informação do estado emocional almejado. As poses-chave representam visemas dependentes de contexto fonético que implicitamente modelam os efeitos da coarticulação na fala visual. A transição entre poses-chave adjacentes é realizada por um algoritmo de metamorfose não-linear entre imagens. As animações sintetizadas aplicando-se a metodologia proposta foram avaliadas por meio de avaliação perceptual de reconhecimento de emoções. Dentre as contribuições deste trabalho encontra-se a construção de uma base de dados de vídeo e captura de movimento para fala expressiva em português do Brasil<br>Abstract: The facial animation technology experiences an increasing demand for applications involving virtual assistants, sellers, tutors and newscasters; lifelike game characters, social agents, and tools for scientific experiments in psychology and behavioral sciences. A relevant and challenging aspect of the development of talking heads is the realistic reproduction of the speech articulatory movements combined with the elements of non-verbal communication and the expression of emotions. This work presents an image-based, or 2D, facial animation synthesis methodology that allows the reproduction of a wide range of expressive speech emotional states and also supports the modulation of head movements and the control of face elements, like the blinking of the eyes and the raising of the eyebrows. The synthesis of the animation uses a database of prototype images which are combined to produce animation keyframes. The weights used for combining the prototype images are derived from a statistical active appearance model (AAM), which is built from a set of sample images extracted from an audio-visual corpus of a real face. The generation of the animation keyframes is driven by the timed phonetic transcription of the speech to be animated and the desired emotional state. The keyposes consist of expressive context-dependent visemes that implicitly model the speech coarticulation effects. The transition between adjacent keyposes is performed through a non-linear image morphing algorithm. To evaluate the synthesized animations, a perceptual evaluation based on the recognition of emotions was performed. Among the contributions of the work is also the building of a database of expressive speech video and motion capture data for Brazilian Portuguese<br>Doutorado<br>Engenharia de Computação<br>Doutora em Engenharia Elétrica

Beginning at birth, a child's receptive and expressive language skills are developing in stages. Likewise, the child's socialization skills are progressing in stages. However, it does not seem that communication and socialization are developing independently of each other. Rather, it seems that their development is interrelated.Children learn to speak in a social context, and social situations are necessary for the development of a variety of language structure~ On the same note, in order for those language structures to develop normally, it is necessary for the child to participate in different social situations. Social interactionists have theorized for some time that human language develops out of the social-communicative functions that language serves in human relations. Vygotsky (1962) theorized that language development, social development, and cognitive development all overlap. He stated that a child's social means of thought is language and referred to this as "verbal thought." This verbal thought process serves a major social function. It is through this verbal thought process that children have the ability to be socialized by others and to socialize with others: If, in fact,Expressive language skills and socialization skills do develop together, it would then seem logical that the child who is late to begin talking would also experience initial deficits in the development of socialization Subsequently, it would seem that the late-talking child (L T) who has persistent deficits in language would, in turn, maintain chronic deficits in socialization. Results of a study which set out to investigate the differences between two and three-year old subjects with a history of LT and their normal language peers indicated that subjects with a history of LT are, in fact, at risk for persistent delays in both expressive language and socialization (Paul, Spangle Looney, and Dahm, 1991). The purpose of this study was to compare the language and socialization skills of a group of five-year olds with a history of LT to a group of normal subjects of the same age. If significant differences were found between the two groups in either area, the scores of the subjects with a history of LT at age two would be correlated with their scores at age five to investigate whether a significant relationship existed between their scores at both ages. It was hypothesized that the subjects with a history of LTwould be at risk for longterm delays in both language and socialization. More specifically, the group of subjects with a history of LT, as a whole, would show significant delays in the areas of expressive language and socialization as compared to the normal controls. It was further hypothesized that the subjects with a history of LTs' scores at the age of two would reliably predict their scores at five, given a significant deficit in either area. The Vineland Adaptive Behavior Scales VABS (Sparrow, Balla, & Cicchetti, 1984) was the test instrument used to gather the data at both age levels, five years and two years. Parents of 25 subjects with a history of L T and 25 normal subjects were interviewed by a trained graduate researcher on their child's communication, daily living and socialization skills using the VABS. Results of an ANOVA and Tukey multiple comparisons indicated that the subjects with a history of LT, as a whole, scored significantly lower than the normal subjects in the areas of expressive communication and socialization at age five. Since a proportion of the test items in the socialization domain of the VABS require the child to verbalize, an item analysis between the verbal and the nonverbal test items was performed to determine the influence of the verbal test items on the subjects with a history of LTs' socialization scores. Results of the item-analysis indicated that the subjects with a history of L T's poor performance on the socialization scale was due to their deficits in social skills not their deficits in expressive language. Lastly, a Pearson Product Moment Correlational Test was conducted to investigate the relationship between the subjects with a history of LTs' scores at age two on the communication and the socialization scales and their scores at age five on the same scales. Results indicated that the subjects with a history of LTs' scores on both the socialization scale and the communication scale at age two correlated significantly with their scores on the socialization scale at age five. Therefore, the subjects with a history of LTs' socialization and communication scores at age two are good predictors of their adaptive social skills at the age of five.

Researchers are seeking more information . on how and why language disorders tend to run in families, particularly siblings of language disordered children. The purpose of this study was to examine the prevalence of language and related disorders in the siblings of two groups of children: those with slow expressive language development (SELD) and those with a normal language history. This study sought to answer the following questions: 1) Is there a significant difference in prevalence of language problems in two groups of children: those with SELD and those with a normal language history?, and 2) Is there a greater probability of language problems in the siblings of children in the SELD group who have receptive/expressive language disorders when compared to those SELD children with pure expressive language deficits or to those with a normal language history? Subjects were 45 7-and 8-year old children participating in a longitudinal study at Portland State University. The children were divided into two groups, normal and SELD, based on test scores administered at intake to the original study. In order to look at the effect of a receptive component on heritability of language disorders, the SELD group was subgrouped into pure expressive language disorders and receptive/expressive language disorders based on tests administered at intake. A family history questionnaire was the method of data collection, asking parents to report on ten areas of language and related disorders in the siblings of subjects. Results of one-sided z-tests and a chi-square test were computed and consistently found a highly significant difference between groups, with families of SELD subjects more likely to report a history of language problems over the normal group. These results are consistent with previous research in showing the heritability of language disorders. Results may also indicate that a receptive language component is associated with heritability of specific language disorders among children.

The relationships between different types of adult-support inferential book reading strategies and young children’s language and literacy competence were examined in 18 studies that included 1134 study participants. van Kleeck’s (2006) descriptions of two levels of inferencing and different types of inferential strategies at each level were used to code and analyze the patterns of correlations between the book reading strategies and the child outcomes. Results showed that parents’ and teachers’ use of different types of inferencing strategies were related to variations in the child outcomes, and that the effects of inferencing were conditioned on the children’s ages. Implications for practice are described.

Predicting language outcomes in children who at age two are "late talkers" is a concern of Speech Language Pathologists. Currently, there is no conclusive data allowing specialists to predict which children will outgrow their delays and which children will not. The purpose of the present study is to analyze the effect of a receptive language delay on the outcome of the slow expressive language delayed child, and determine whether or not it is a viable predictor of poor outcomes. The subject information used in this project was compiled from the data collected and reported by Paul (1991) during the Portland Language Development Project (PLDP). Children in the PLDP first participated in the longitudinal study between the ages of twenty to thirtyfour months. They were categorized as being slow in expressive language development if they produced fewer that fifty intelligible words during this age range. They were then subgrouped into an expressive-receptive delayed group if they scored more than one standard deviation below the mean on the Reynell Developmental Language Scales. Of the twenty-five subjects with complete data over the five years of the study, nineteen were considered to be solely expressively delayed, while the remaining six were classified as having both an expressive and a receptive language delay. Lee's Developmental Sentence Scoring (DSS) (1974) was used to track the subject's expressive language abilities to the age of seven. DSS scores were analyzed yearly, using the Mann-Whitney nonparametric statistical test. This would determine whether the subjects considered to be both expressively and receptively delayed were exhibiting more difficulties in their expressive language abilities than those subjects with expressive delays alone. The results of the study indicated that significant differences did not exist between the two groups. Therefore, there was insufficient evidence to conclude that a receptive language delay at twenty to thirty-four months of age is a feasible predictor of lasting expressive language delays. This leads to the recommendation that additional research be conducted focusing on areas other than receptive language abilities as being predictors of poor expressive language outcomes.

This study was part of the Portland Language Development Project. The purpose was to establish reliability for the Goodenough Draw-A-Man Test. Additionally, nonlinguistic cognitive performance scores were correlated with soores from expressive and receptive language test soores. Finally, scores of overall cognitive function and of nonlinguistic cognitive function in children with normally developing language (NL) and with expressive language delay (ELD) were compared. The original group size was 60 children, 30 with ELD at the age of 20 months, and 30 who were a matched control group. These subjects were reevaluated during Kindergarten. The Draw-A-Man Test was administered to assess the subjects' nonlinguistic cognitive functioning. The McCarthy Scales of Children's Abilities CMCSA) was administered to assess the subjects' overall cognitive functioning. A free speech sample was analyzed using the Developmental Sentence Scoring (DSS) criteria to assess expressive language skills, and the Vineland Adaptive Behavior Scales

The contemporary research suggests that some children who present with early language delays as toddlers outgrow their delays while others continue to develop long-term language difficulties. Several studies over the years have focused on factors that might aid in predicting the outcome of late talkers. This current study emphasized exploring gender as a possible predictive factor. The purpose of this study was to determine if significant differences exist in the rate of growth in language skills, as indexed by scores on the Developmental Sentence Scoring (DSS) procedure (Lee, 1974) of boys versus girls who are late to start talking as toddlers. The research hypothesis was that boys who present as LT toddlers would score significantly higher than LT girls at each age level tested. The DSS is a norm-referenced instrument that assesses age-appropriate morphological development and syntax. The LT subjects used were part of the Portland Language Development Project, a longitudinal study. Spontaneous speech samples were collected, transcribed, and analyzed using the DSS procedure once each year from the time they were approximately 3 years of age, until the age of 7. Late talking children in this present study were grouped by gender. A Chi Square test was used to determine if the proportion of males scoring above the 10th percentile on the DSS was significantly different than the proportion of females scoring above the 10th percentile at each age. Results from this analysis indicated that at the age of 3 years, more boys than girls scored above the 10th percentile on the DSS. There were no significant differences found at the ages of 4, 5, 6, and 7. At-test was used to compare average DSS scores between the two genders for each year of the study. This test revealed a significant difference between the LT girls' and LT boys' scores at the age of 3 years. No significant differences were found for the subsequent years. However, difference between boys' and girls' scores at age 7 approached significance, with boys again scoring higher.

Research suggests that the linguistic environment of the expressive language delayed child is different from that of his peers. Does this difference actually exist and if so, what are its characteristics? The purpose of this study was to describe the linguistic characteristics of mothers' input to children with normal language acquisition and those of mothers of expressively delayed toddlers; and to identify any differences between these groups. Though considerable research exists in this area, few studies have dealt specifically with large groups of expressively delayed toddlers.

This study attempts to further our understanding of linguistic politeness by focusing on both a Western and a non-Western language. It is based on two sets of data (one spontaneous and the other elicited) and provides a comparative analysis of three expressive speech acts produced by native speakers of British English and Thai. At face value, compliments, apologies and thanks may seem to have little referential meaning, yet these speech acts can be crucially important in originating, maintaining or even terminating social relationships. The data reveal a tendency for the two groups of speakers to use the three politeness devices in a different manner, reflecting cross-cultural differences in social norms and value systems. This project follows earlier studies of similar nature, in particular those carried out on different varieties of English. The findings are interpreted within pragmatic and sociolinguistic theoretical frameworks, and are discussed in the following format: linguistic structures of the speech acts, their functions, the topics of compliments, apologies and thanks, interpersonal and contextual factors influencing the production of these expressives, and the responses given to them. The analysis has implications for language specialists and lay people alike, in that it brings together a number of important insights with regard to these speech features that may result in miscommunication if and when British and Thai speakers converse in intercultural situations.

Defining the genre 'elegy' in the Anglo-Saxon cultural horizon and at the same time delimiting the boundaries within which some Old English lyrical compositions might be grouped is a far from trivial operation. Indeed, contrary to the classical tradition where the elegiac meter – a dactylic hexameter verse followed by a dactylic pentameter- represents the distinctive element of the genre, the Anglo-Saxon poetry lacks a characterising meter for the elegies. Since the early nineteenth century scholars have been applying the word “elegy” to nine Old English poems found in the Exeter Book manuscript that have traditionally been treated as genre for their general sense of lament. The current study is part of this field of research and proposes a new attempt to delimit the elegy genre through a pragmatic analysis. In details, the “speech act theory" will be applied since it provides an interesting and productive new way of looking at the elegiac discourse. The analysis of speech acts will be conducted on two levels: at a micro-structural level, namely by focussing on individual illocutionary acts expressed in single sentences, and at a macro-structural level, by focussing on the so-called macro-speech acts, where it seems possible to recognise a global discourse illocutionary point together with a discourse illocutionary force. The identification of an expressive illocutionary force, in particular that of lamenting, at the macro level of speech will represent a possible demonstration of the elegiac nature of these compositions.

This investigation was conducted to determine whether the removal of hearing aids from these children for eighteen hours (+ 1/2 hour and including sleep time) would result in reduced speech intelligibility as perceived by a panel of listening judges who were unfamiliar with the speech of the deaf.

Durant segles, la interpretació oral de contes i històries ha sigut una tradició mundial lligada a l’entreteniment, la educació, i la perpetuació de la cultura. En les últimes dècades, alguns treballs s’han centrat en analitzar aquest estil de parla ric en matisos expressius caracteritzats per determinats patrons acústics. En relació a això, també hi ha hagut un interès creixent en desenvolupar aplicacions de contar contes, com ara les de contacontes interactius. Aquesta tesi està orientada a millorar aspectes claus d’aquest tipus d’aplicacions: millorar la naturalitat de la parla sintètica expressiva a partir d’analitzar la parla de contacontes en detall, a més a més de proporcionar un millor llenguatge no verbal a un avatar parlant mitjançant la sincronització de la parla i els gestos. Per aconseguir aquests objectius és necessari comprendre les característiques acústiques d’aquest estil de parla i la interacció de la parla i els gestos. Pel que fa a característiques acústiques de la parla de contacontes, la literatura relacionada ha treballat en termes de prosòdia, mentre que només ha estat suggerit que la qualitat de la veu pot jugar un paper important per modelar les subtileses d’aquest estil. En aquesta tesi, el paper tant de la prosòdia com de la qualitat de la veu en l’estil indirecte de la parla de contacontes en diferents idiomes és analitzat per identificar les principal categories expressives que la composen i els paràmetres acústics que les caracteritzen. Per fer-ho, es proposa una metodologia d’anotació per aquest estil de parla a nivell de oració basada en modes de discurs dels contes (mode narratiu, descriptiu, i diàleg), introduint a més sub-modes narratius. Considerant aquesta metodologia d’anotació, l’estil indirecte d’una història orientada a una audiència jove (cobrint versions en castellà, anglès, francès, i alemany) és analitzat en termes de prosòdia i qualitat de la veu mitjançant anàlisis estadístics i discriminants, després de classificar els àudios de les oracions de la història en les seves categories expressives. Els resultats confirmen l’existència de les categories de contes amb diferències expressives subtils en tots els idiomes més enllà dels estils personals dels narradors. En aquest sentit, es presenten evidències que suggereixen que les categories expressives dels contes es transmeten amb matisos expressius més subtils que en les emocions bàsiques, després de comparar els resultats obtinguts amb aquells de parla emocional. Els anàlisis també mostren que la prosòdia i la qualitat de la veu contribueixen pràcticament de la mateixa manera a l’hora de discriminar entre les categories expressives dels contes, les quals son expressades amb patrons acústics similars en tots els idiomes analitzats. Cal destacar també la gran relació observada en la selecció de categoria per cada oració que han fet servir els diferents narradors encara quan, que sapiguem, no se’ls hi va donar cap indicació. Per poder traslladar totes aquestes categories a un sistema de text a parla basat en corpus, caldria enregistrar un corpus per cada categoria. No obstant, crear diferents corpus ad-hoc esdevé un tasca molt laboriosa. En la tesi, s’introdueix una alternativa basada en una metodologia d’anàlisi orientada a síntesi dissenyada per derivar models de regles des de un petit però representatiu conjunt d’oracions, que poden poder ser utilitzats per generar parla amb estil de contacontes a partir de parla neutra. Els experiments sobre suspens creixent com a prova de concepte mostren la viabilitat de la proposta en termes de naturalitat i similitud respecte un narrador de contes real. Finalment, pel que fa a interacció entre parla i gestos, es realitza un anàlisi de sincronia i èmfasi orientat a controlar un avatar de contacontes en 3D. Al tal efecte, es defineixen indicadors de força tant per els gestos com per la parla. Després de validar-los amb tests perceptius, una regla d’intensitat s’obté de la seva correlació. A més a més, una regla de sincronia es deriva per determinar correspondències temporals entre els gestos i la parla. Aquests anàlisis s’han dut a terme sobre interpretacions neutres i agressives per part d’un actor per cobrir un gran rang de nivells d’èmfasi, com a primer pas per avaluar la integració d’un avatar parlant després del sistema de text a parla.<br>Durante siglos, la interpretación oral de cuentos e historias ha sido una tradición mundial ligada al entretenimiento, la educación, y la perpetuación de la cultura. En las últimas décadas, algunos trabajos se han centrado en analizar este estilo de habla rico en matices expresivos caracterizados por determinados patrones acústicos. En relación a esto, también ha habido un interés creciente en desarrollar aplicaciones de contar cuentos, como las de cuentacuentos interactivos. Esta tesis está orientada a mejorar aspectos claves de este tipo de aplicaciones: mejorar la naturalidad del habla sintética expresiva a partir de analizar el habla de cuentacuentos en detalle, además de proporcionar un mejor lenguaje no verbal a un avatar parlante mediante la sincronización del habla y los gestos. Para conseguir estos objetivos es necesario comprender las características acústicas de este estilo de habla y la interacción del habla y los gestos. En cuanto a características acústicas del habla de narradores de cuentos, la literatura relacionada ha trabajado en términos de prosodia, mientras que sólo ha sido sugerido que la calidad de la voz puede jugar un papel importante para modelar las sutilezas de este estilo. En esta tesis, el papel tanto de la prosodia como de la calidad de la voz en el estilo indirecto del habla de cuentacuentos en diferentes idiomas es analizado para identificar las principales categorías expresivas que componen este estilo de habla y los parámetros acústicos que las caracterizan. Para ello, se propone una metodología de anotación a nivel de oración basada en modos de discurso de los cuentos (modo narrativo, descriptivo, y diálogo), introduciendo además sub-modos narrativos. Considerando esta metodología de anotación, el estilo indirecto de una historia orientada a una audiencia joven (cubriendo versiones en castellano, inglés, francés, y alemán) es analizado en términos de prosodia y calidad de la voz mediante análisis estadísticos y discriminantes, después de clasificar los audios de las oraciones de la historia en sus categorías expresivas. Los resultados confirman la existencia de las categorías de cuentos con diferencias expresivas sutiles en todos los idiomas más allá de los estilos personales de los narradores. En este sentido, se presentan evidencias que sugieren que las categorías expresivas de los cuentos se transmiten con matices expresivos más sutiles que en las emociones básicas, tras comparar los resultados obtenidos con aquellos de habla emocional. Los análisis también muestran que la prosodia y la calidad de la voz contribuyen prácticamente de la misma manera a la hora de discriminar entre las categorías expresivas de los cuentos, las cuales son expresadas con patrones acústicos similares en todos los idiomas analizados. Cabe destacar también la gran relación observada en la selección de categoría para cada oración que han utilizado los diferentes narradores aun cuando, que sepamos, no se les dio ninguna indicación. Para poder trasladar todas estas categorías a un sistema de texto a habla basado en corpus, habría que grabar un corpus para cada categoría. Sin embargo, crear diferentes corpus ad-hoc es una tarea muy laboriosa. En la tesis, se introduce una alternativa basada en una metodología de análisis orientada a síntesis diseñada para derivar modelos de reglas desde un pequeño pero representativo conjunto de oraciones, que pueden ser utilizados para generar habla de cuentacuentos a partir de neutra. Los experimentos sobre suspense creciente como prueba de concepto muestran la viabilidad de la propuesta en términos de naturalidad y similitud respecto a un narrador de cuentos real. Finalmente, en cuanto a interacción entre habla y gestos, se realiza un análisis de sincronía y énfasis orientado a controlar un avatar cuentacuentos en 3D. Al tal efecto, se definen indicadores de fuerza tanto para gestos como para habla. Después de validarlos con tests perceptivos, una regla de intensidad se obtiene de su correlación. Además, una regla de sincronía se deriva para determinar correspondencias temporales entre los gestos y el habla. Estos análisis se han llevado a cabo sobre interpretaciones neutras y agresivas por parte de un actor para cubrir un gran rango de niveles de énfasis, como primer paso para evaluar la integración de un avatar parlante después del sistema de texto a habla.<br>For ages, the oral interpretation of tales and stories has been a worldwide tradition tied to entertainment, education, and perpetuation of culture. During the last decades, some works have focused on the analysis of this particular speaking style rich in subtle expressive nuances represented by specific acoustic cues. In line with this fact, there has also been a growing interest in the development of storytelling applications, such as those related to interactive storytelling. This thesis deals with one of the key aspects of audiovisual storytellers: improving the naturalness of the expressive synthetic speech by analysing the storytelling speech in detail, together with providing better non-verbal language to a speaking avatar by synchronizing that speech with its gestures. To that effect, it is necessary to understand in detail the acoustic characteristics of this particular speaking style and the interaction between speech and gestures. Regarding the acoustic characteristics of storytelling speech, the related literature has dealt with the acoustic analysis of storytelling speech in terms of prosody, being only suggested that voice quality may play an important role for the modelling of its subtleties. In this thesis, the role of both prosody and voice quality in indirect storytelling speech is analysed across languages to identify the main expressive categories it is composed of together with the acoustic parameters that characterize them. To do so, an analysis methodology is proposed to annotate this particular speaking style at the sentence level based on storytelling discourse modes (narrative, descriptive, and dialogue), besides introducing narrative sub-modes. Considering this annotation methodology, the indirect speech of a story oriented to a young audience (covering the Spanish, English, French, and German versions) is analysed in terms of prosody and voice quality through statistical and discriminant analyses, after classifying the sentence-level utterances of the story in their corresponding expressive categories. The results confirm the existence of storytelling categories containing subtle expressive nuances across the considered languages beyond narrators' personal styles. In this sense, evidences are presented suggesting that such storytelling expressive categories are conveyed with subtler speech nuances than basic emotions by comparing their acoustic patterns to the ones obtained from emotional speech data. The analyses also show that both prosody and voice quality contribute almost equally to the discrimination among storytelling expressive categories, being conveyed with similar acoustic patterns across languages. It is also worth noting the strong relationship observed in the selection of the expressive category per utterance across the narrators even when, up to our knowledge, no previous indications were given to them. In order to translate all these expressive categories to a corpus-based Text-To-Speech system, the recording of a speech corpus for each category would be required. However, building ad-hoc speech corpora for each and every specific expressive style becomes a very daunting task. In this work, we introduce an alternative based on an analysis-oriented-to-synthesis methodology designed to derive rule-based models from a small but representative set of utterances, which can be used to generate storytelling speech from neutral speech. The experiments conducted on increasing suspense as a proof of concept show the viability of the proposal in terms of naturalness and storytelling resemblance. Finally, in what concerns the interaction between speech and gestures, an analysis is performed in terms of time and emphasis oriented to drive a 3D storytelling avatar. To that effect, strength indicators are defined for speech and gestures. After validating them through perceptual tests, an intensity rule is obtained from their correlation. Moreover, a synchrony rule is derived to determine temporal correspondences between speech and gestures. These analyses have been conducted on aggressive and neutral performances to cover a broad range of emphatic levels as a first step to evaluate the integration of a speaking avatar after the expressive Text-To-Speech system.

Children who are identified with slow expressive language development (SELD) around the age of two are producing less than fifty intelligible words or no two word phrases. Current research suggests that some children with SELD outgrow their delay while others continue to develop long term language difficulties. The literature shows varied findings of short term recovery but long term deficits, and shifts in the specific expressive language deficits ~s the child with SELD matures and encounters increased language demands. Suggestions are found for a mix of monitoring and early intervention, in step with signs of readiness and dynamic assessments, to facilitate improved performance and hasten development, particularly in the areas of metalinguistics and narratives. This study attempted to support the recommendation of early intervention, particularly for those children with an initial greater severity levels of expressive communication delay at the age of two. The 24 male and seven female SELD subjects were part of the Portland Language Development Project, a longitudinal study. Intake was at two years, and placement in the Intervention (Rx) or No Intervention (No Rx) group was a result of follow-up information gathered from parents regarding enrollment in any early intervention services before the age of four: Using mean Developmental Sentence Scores (DSS) for four outcome points, 1-tests determined that no significant differences existed in the improvement of language production between the Rx and No Rx groups. Secondly, 1-tests showed no significant differences in the two group's initial severity levels, using the Expressive Communication sub-domain of the Vine~and Adaptive Behavior Scale (V ABS), as the measure of severity at intake. A non-significant trend of consistently higher actual mean DSS scores across all outcome points, and an actual lower mean Expressive Communication score on the V ABS at intake was noted for the Rx group. A significant difference was found in the mean intake ages of the two groups, with older toddlers falling into the Rx group. Research and clinical_ implications are discussed, including attention to the length, type and content of very early intervention services, effective initial and follow up assessments, and factors that favor recommending early intervention

Language delay and phonological delay have been shown to coexist. Because they so often co-occur, it is possible that they may interact, sharing a relationship during the child's development. A group of children who were "late talkers" as toddlers, achieved normal development in their syntactic ability by the preschool period. Because their language abilities are known to have increased rapidly, data on their phonological development could provide information on the relationship between phonological and syntactic development. The purpose of this study was to compare the percentage of phonological process usage of the eight most commonly used simplification processes in four-year-old expressive language delayed (ELD) children, children with a history of slow expressive language development (HX), and normally developing (ND) children. The questions this study sought to answer were: do ELD children exhibit a higher percentage of phonological process usage than ND children, and are HX children significantly different in their percentage of phonological process usage than ND and/or ELD children.

De plus en plus, l'interaction entre personne et machine se rapproche du naturel afin de ressembler à l'interaction entre humains, incluant l'expressivité (en particulier les émotions et les attitudes). Dans la communication parlée, les attitudes, et plus généralement les affects sociaux, sont véhiculés principalement par la prosodie. Pour les langues tonales, la prosodie est utilisée aussi pour coder l'information sémantique dans les variations de tons. Ce travail de thèse présente une étude des affects sociaux du vietnamien, une langue à tons et une langue peu dotée, afin d'appliquer les résultats obtenus à un système de synthèse de haute qualité capable de produire la parole " expressive " pour le vietnamien. Le premier travail de cette thèse consiste en la construction du premier corpus audio-visuel des attitudes vietnamiennes, qui contient seize attitudes. Ce corpus est ensuite utilisé pour étudier la perception audio-visuelle et interculturelle des attitudes vietnamiennes. Pour cela, une série de tests perceptifs a été effectuée avec des auditeurs natifs et non-natifs (des auditeurs francophones pour les non-natifs). Les résultats de ces tests montrent que les facteurs influant sur la perception des attitudes sont l'expression de l'attitude elle-même et la modalité de présentation (audio, visuelle et audio-visuelle). Ces résultats nous ont ainsi permis de trouver des affects sociaux communs ou interculturels entre le vietnamien et le français. Puis, un autre test de perception a été réalisé sur des phrases avec tons afin d'explorer l'effet du système tonal du vietnamien sur la perception des attitudes. Les résultats montrent que les juges non-natifs peuvent traiter et séparer les indices tonals locaux et les traits saillants prosodiques de portée globale. Après une présentation de nos études sur les affects sociaux en vietnamien, nous décrivons notre modélisation de la prosodie des attitudes en vue de la synthèse de la parole expressive en vietnamien. En nous basant sur le modèle de superposition des contours fonctionnels, nous proposons une méthode pour modéliser et générer de la prosodie expressive en vietnamien. Cette méthode est ensuite appliquée pour générer de la parole expressive en vietnamien, puis évaluée par des tests de perception sur les énoncés synthétiques. Les résultats de perception valident bien la performance de notre modèle et confirment que l'approche de superposition de contours fonctionnels peut être utilisée pour modéliser une prosodie complexe comme dans le cas de la parole expressive d'une langue à tons.

Researchers have long been concerned with the effects of otitis media on speech and language acquisition because of the high correlation of a mild to moderate hearing loss during the time period that fluid (effusion) may be in the middle ear. Middle-ear effusion would prevent many of the auditory messages from accurately reaching the nervous system (Zinkus, 1986). Deprived of the ability to discern the subtle acoustic differences that provide information for phonetic contrasts, a child's speech acquisition may differ from children who do not experience such losses.

This thesis defends the claim that reason-giving is a discrete type of speech act (of an expressive kind) that has a distinctive value. It further argues that this value is best understood in terms of recognition, rather than justification, and that it is intrinsic to reason-giving. Its main aim is to argue against the commonly-held view that the main, sometimes the only, value to reason-giving lies in its capacity to provide justification (and in the related claim that if reasons cannot justify, then reason-giving has no value). The argument presented is intended to support that recognition (of a certain type) is a value that reason-giving has independently from any other value that it might or might not have – including justification; and hence, that reason-giving has a certain distinctive value that is not predicated upon a capacity for actually achieving justification. In particular, this thesis argues, based on speech act theory and on the concept of recognition, that this value is best understood as consisting in the expression of a particular type of recognition for the other. To establish this claim, in chapter one, it begins by setting out the standard view: that the value of reason-giving lies in its capacity to justify, and analyses some of the moves that have been made in the literature when the connection between reason-giving and justification breaks down. In chapters two to four, it uses speech act theory to analyse the acts of arguing and reason-giving, and to argue that reason-giving is a discrete speech act that has features in common with but is not reducible to arguing. Finally, in chapter five, it defends the claim that reason-giving has an intrinsic value, and that this value is best understood as an expressive value: namely, the expression of recognition for the other as a rational being (which is a valuable feature of the other’s humanity); and that it has this value regardless of whether the reasons in question are “good” from a justificatory standpoint.

<p> The study was a pre-experimental pre- and post-treatment single case study which focused on evaluating the effects of augmentative and alternative communication (AAC) modeling during shared storybook readings. Aided AAC modeling interventions provide verbal and visual language models to support language comprehension and use for children with complex communication needs (CCN). The study measured four aspects of change before and after the AAC modeling phase including a) the number of communicative turns by the AAC user; b) the complexity and length of the initiations and responses made by the AAC user; c) the accuracy in responses to comprehension questions following the story; d) perceptions of the parent participant in regards to the intervention. The results indicated that when aided AAC modeling was implemented the child participant demonstrated an increase in: the number of communicative turns, accuracy in answering comprehension questions, comprehension of story grammar terminology, and production of story retells.</p>

The purpose of this study was to compare expressive communication, receptive communication, and socialization achievement in 18- to 34-month-old ELD toddlers to the same skills in normally-speaking children. The questions this study sought to answer were, how do the three skill areas in ELD toddlers compare with the same skills in normal toddlers?, will ELD subjects evidence specific profiles of deficits involving not only expressive but receptive and social skills as well? and, within the ELD subjects will two subgroups emerge, one group having poor expressive skills only, and a second group having deficits in addition to expression.

The purpose of this investigation was to examine the construct validity of a new test, which purports to measure morphology, entitled Test for Examining Expressive Morphology (TEEM) (Shipley, Stone and Sue, 1983). Additional tests of expressive morphology, the Bankson Language Screening Test (BLST) (Bankson, 1977) and the Test of Language Development-Primary (TOLD-P) (Newcomer and Hammill, 1982) were utilized to determine the association of the TEEM with two highly-researched instruments.

Children with severe physical disabilities often do not have the capabilities for oral communication. Professionals are frequently faced with selecting vocabulary for children who are unable to use vocal output because of severe motor impairments. A child who is nonambulatory may have additional reasons for communicating and sees the world from a different viewpoint than his ambulatory peers. Selecting appropriate words for an initial lexicon that are useful to nonspeaking disabled children that also meet normal language acquisition standards is a concern. This study specifically addresses this concern by looking at the vocabulary differences of ambulatory and nonambulatory preschool children. The purpose of this research project was to compare expressive vocabulary produced by nonambulatory, speaking children with the expressive vocabulary produced by ambulatory, speaking children. It is suggested that the vocabulary of nonambulatory, speaking children might be more appropriate for selecting a lexicon for AAC systems if indeed, they are different from words produced by ambulatory, speaking children.

L'analyse des signaux de parole permet de comprendre le fonctionnement de l'appareil vocal, mais aussi de décrire de nouveaux paramètres permettant de qualifier et quantifier la perception de la voix. Dans le cas de la parole expressive, l'intérêt se porte sur des variations importantes de qualité vocales et sur leurs liens avec l'expressivité et l'intention du sujet. Afin de décrire ces liens, il convient de pouvoir estimer les paramètres du modèle de production mais aussi de décomposer le signal vocal en chacune des parties qui contribuent à ce modèle. Le travail réalisé au cours de cette thèse s'axe donc autour de la segmentation et la décomposition des signaux vocaux et de l'estimation des paramètres du modèle de production vocale : Tout d'abord, la décomposition multi-échelles des signaux vocaux est abordée. En reprenant la méthode LoMA qui trace des lignes suivant les amplitudes maximum sur les réponses temporelles au banc de filtre en ondelettes, il est possible d'y détecter un certain nombre de caractéristiques du signal vocal : les instants de fermeture glottique, l'énergie associée à chaque cycle ainsi que sa distribution spectrale, le quotient ouvert du cycle glottique (par l'observation du retard de phase du premier harmonique). Cette méthode est ensuite testée sur des signaux synthétiques et réels. Puis, la décomposition harmonique + bruit des signaux vocaux est abordée. Une méthode existante (PAPD - Périodic/APériodic Décomposition) est adaptée aux variations de fréquence fondamentale par le biais de la variation dynamique de la taille de la fenêtre d'analyse et est appelée PAP-A. Cette nouvelle méthode est ensuite testée sur une base de signaux synthétiques. La sensibilité à la précision d'estimation de la fréquence fondamentale est notamment abordée. Les résultats montrent des décompositions de meilleures qualité pour PAP-A par rapport à PAPD. Ensuite, le problème de la déconvolution source/filtre est abordé. La séparation source/filtre par ZZT (zéros de la transformée en Z) est comparée aux méthodes usuelles à base de prédiction linéaire. La ZZT est utilisée pour estimer les paramètres du modèle de la source glottique via une méthode simple mais robuste qui permet une estimation conjointe de deux paramètres du débit glottique : le quotient ouvert et l'asymétrie. La méthode ainsi développée est testée et combinée à l'estimation du quotient ouvert par ondelettes. Finalement, ces trois méthodes d'estimations sont appliquées à un grand nombre de fichiers d'une base de données comportant différents styles d'élocution. Les résultats de cette analyse sont discutés afin de caractériser le lien entre style, valeur des paramètres de la production vocale et qualité vocale. On constate notamment l'émergence très nette de groupes de styles.

The purpose of this study was to examine the relationship between verbal and nonverbal indices of intelligence from an infant development scale given at approximately two years of age with scores on an expressive language test administered to the same children two years later. The questions this study sought to answer were: (1) is there a difference between the number of verbal and nonverbal items passed at 18-34 months by a group of normal children and a group of late talkers, (2) is there a significant relationship between correct verbal and nonverbal responses on an inf ant development test and scores on an expressive language test two years later in normal children and late talkers, (3) is there a significant relationship between correct verbal receptive items on an infant development test and scores on an expressive language test two years later in normal children and late talkers, and (4) is there a significant relationship between correct expressive responses on an infant development test and expressive scores two years later in the normal children and the group of later talkers.

L'objectif de la thèse est de proposer des méthodes permettant d'améliorer l'expressivité des systèmes de synthèse de la parole. Une des propositions centrales de ce travail est de définir, utiliser et mesurer l'impact de structures linguistiques opérant au delà du niveau de la phrase, par opposition aux approches opérant sur des phrases isolées de leur contexte. Le cadre de l'étude est restreint au cas de la lecture de contes pour enfants. Les contes ont la particularité d'avoir été l'objet d'un certain nombre d'études visant à en dégager une structure narrative et de faire intervenir une certain nombre de stéréotypes de personnages (héros, méchant, fée) dont le discours est souvent rapporté. Ces caractéristiques particulières sont exploitées pour modéliser les propriétés prosodiques des contes au delà du niveau de la phrase. La transmission orale des contes a souvent été associée à une pratique musicale (chants, instruments) et leur lecture reste associée à des propriétés mélodiques très riches, dont la reproduction reste un défi pour les synthétiseurs de parole modernes. Pour répondre à ces problématiques, un premier corpus de contes écrits est collecté et annoté avec des informations relatives à la structure narrative des contes, l'identification et l'attribution des citations directes, le référencement des mentions des personnages ainsi que des entités nommées et des énumérations étendues. Le corpus analysé est décrit en terme de couverture et d'accord inter-annotateurs. Il est utilisé pour modéliser des systèmes de segmentation des contes en épisode, de détection des citations directes, des actes de dialogue et des modes de communication. Un deuxième corpus de contes lus par un locuteur professionnel est présenté. La parole est alignée avec les transcriptions lexicale et phonétique, les annotations du corpus texte et des méta-informations décrivant les caractéristiques des personnages intervenant dans le conte. Les relations entre les annotations linguistiques et les propriétés prosodiques observées dans le corpus de parole sont décrites et modélisées. Finalement, un prototype de contrôle des paramètres expressifs du synthétiseur par sélection d'unités Acapela est réalisé. Le prototype génère des instructions prosodiques opérant au delà du niveau de la phrase, notamment en utilisant les informations liées à la structure du conte et à la distinction entre discours direct et discours rapporté. La validation du prototype de contrôle est réalisée dans le cadre d'une expérience perceptive, qui montre une amélioration significative de la qualité de la synthèse<br>The aim of this thesis is to propose ways to improve the expressiveness of speech synthesis systems. One of the central propositions of this work is to define, use and measure the impact of linguistic structures operating beyond the sentence level, as opposed to approaches operating on sentences out of their context. The scope of the study is restricted to the case of storytelling for children. The stories have the distinction of having been the subject of a number of studies in order to highlight a narrative structure and involve a number of stereotypical characters (hero, villain, fairy) whose speech is often reported. These special features are used to model the prosodic properties tales beyond the sentence level. The oral transmission of tales was often associated with musical practice (vocals, instruments) and their reading is associated with rich melodic properties including reproduction remains a challenge for modern speech synthesizers. To address these issues, a first corpus of written tales is collected and annotated with information about the narrative structure of stories, identification and allocation of direct quotations, referencing references to characters as well as named entities and enumerations areas. The corpus analyzed is described in terms of coverage and inter-annotator agreement. It is used to model systems segmentation tales episode, detection of direct quotes, dialogue acts and modes of communication. A second corpus of stories read by a professional speaker is presented. The word is aligned with the lexical and phonetic transcriptions, annotations of the corpus text and meta-information describing the characteristics of the characters involved in the story. The relationship between linguistic annotations and prosodic properties observed in the speech corpus are described and modeled. Finally, a prototype control expressive synthesizer parameters by Acapela unit selection is made. The prototype generates prosodic operating instructions beyond the sentence level, including using the information related to the structure of the story and the distinction between direct speech and reported speech. Prototype validation control is performed through a perceptual experience, which shows a significant improvement in the quality of the synthesis

Interest in children who are diagnosed with expressive language delay has increased over the years. This has resulted in follow-up studies which have suggested that these children would have difficulties in academics during their elementary school years (Hall & Tomblin, 1978; Weiner, 1974) The current study sought to determine if children with a history of and children with continued expressive language delay would have problems with academics once they reached the second grade. The Peabody Individual Achievement Test (PIAT) was used to measure academic performance. It was chosen for it's reliable standardization and use of five different subtest areas to determine overall academic achievement. The purpose of this study was to determine if there are significant differences in academic performance on PIAT between three groups of second grade children with different language histories. The three groups are: 1) children with normal language history 2) children with a history of expressive language delay (HELD) who were identified as late to talk between 20 and 34 months of age, but who received a score at or above the tenth percentile in the second grade on the DSS (Developmental Sentence Scoring, Lee 1974), and 3) children with chronic expressive language delay (ELD) who were identified as late to talkers between 20 and 34 months of age, and received a score below the tenth percentile in the second grade on the DSS. Significant differences were found between the ELD group and the Normal group in the areas of Math and General Information as well as the Total Test Score. The ELD group also performed significantly lower than the HELD group in the areas of Math and the Total Test Score. There were no significant differences found between the HELD group and the Normals or between the ELD and HELD groups on the General Information subtest. These results were consistent with the most recent research article by Whitehurst and Fischel (1994) which looked at three longitudinal studies and found that by five years of age most children diagnosed with specific expressive language delay were performing within the normal range in ·various areas of language development.

Children with slow expressive language development often catch up to their normally developing peers in expressive language, but may still exhibit difficulties with metalinguistic skills. Research shows that children who have difficulty with phonemic awareness also have difficulty with reading, which is important for success in school. Speech-language pathologists assist children who have difficulty with expressive oral language and facilitate language development in children who have difficulties with learning metalinguistic skills, such as phonemic awareness. The purpose of the present study was to compare the phoneme segmentation skills in three groups of children: (a) children with a history of oral expressive language delay (HELD) (n= 22) who were identified as toddlers with slow developing expressive language, but caught up to their normally developing peers by first grade; (b) children identified as toddlers with slow developing oral expressive language and by first grade still maintained the expressive language delays (ELD) (n= 7); and (c) children who were identified at age two as developing normal oral expressive language and maintained normal oral expressive language development (NL) (n= 23) in first grade. The children participated in a phonological segmentation test. The study answered four questions: Is there a significant difference among the three groups of children in the number of correct responses on a phonological segmentation test at ( 1 ) the one phoneme level, (2) the two phoneme level, (3) the three phoneme level, and ( 4) the total number of correct responses. Utilizing an ANOVA test, a significant difference was found among the groups at the two phoneme level, with a trend toward a significant difference at the one phoneme level. Other significant differences were not found. The difference at the two phoneme level was between the ELD group and the normal group, as well as between the ELD group and the HELD group.

L’objectif de cette thèse est l’étude et la conception d’une plateforme de synthèse de parole expressive.Le système de synthèse — LIPS3, développé dans le cadre de ce travail, incorpore deux éléments : un module linguistique et un module de synthèse paramétrique par apprentissage statistique (construit à l’aide de HTS et de STRAIGHT). Le système s’appuie sur un corpus monolocuteur conçu, enregistréet étiqueté à cette occasion.Une première étude, sur l’influence de la qualité de l’étiquetage du corpus d’apprentissage, indique que la synthèse paramétrique statistique est robuste aux erreurs de labels et d’alignement. Cela répond au problème de la variation des réalisations phonétiques en parole expressive.Une seconde étude, sur l’analyse acoustico-phonétique du corpus permet la caractérisation de l’espace expressif utilisé par la locutrice pour réaliser les consignes expressives qui lui ont été fournies. Les paramètres de source et les paramètres articulatoires sont analysés suivant les classes phonétiques, ce qui permet une caractérisation fine des phonostyles.Une troisième étude porte sur l’intonation et le rythme. Calliphony 2.0 est une interface de contrôlechironomique temps-réel permettant la modification de paramètres prosodiques (f0 et tempo) des signaux de synthèse sans perte de qualité, via une manipulation directe de ces paramètres. Une étude sur la stylisation de l’intonation et du rythme par contrôle gestuel montre que cette interface permet l’amélioration, non-seulement de la qualité expressive de la parole synthétisée, mais aussi de la qualité globale perçue en comparaison avec la modélisation statistique de la prosodie.Ces études montrent que la synthèse paramétrique, combinée à une interface chironomique, offre une solution performante pour la synthèse de la parole expressive, ainsi qu’un outil d’expérimentation puissant pour l’étude de la prosodie<br>The subject of this thesis was the study and conception of a platform for expressive speech synthesis.The LIPS3 Text-to-Speech system — developed in the context of this thesis — includes a linguistic module and a parametric statistical module (built upon HTS and STRAIGHT). The system was based on a new single-speaker corpus, designed, recorded and annotated.The first study analyzed the influence of the precision of the training corpus phonetic labeling on the synthesis quality. It showed that statistical parametric synthesis is robust to labeling and alignment errors. This addresses the issue of variation in phonetic realizations for expressive speech.The second study presents an acoustico-phonetic analysis of the corpus, characterizing the expressive space used by the speaker to instantiate the instructions that described the different expressive conditions. Voice source parameters and articulatory settings were analyzed according to their phonetic classes, which allowed for a fine phonostylistic characterization.The third study focused on intonation and rhythm. Calliphony 2.0 is a real-time chironomic interface that controls the f0 and rhythmic parameters of prosody, using drawing/writing hand gestures with a stylus and a graphic tablet. These hand-controlled modulations are used to enhance the TTS output, producing speech that is more realistic, without degradation as it is directly applied to the vocoder parameters. Intonation and rhythm stylization using this interface brings significant improvement to the prototypicality of expressivity, as well as to the general quality of synthetic speech.These studies show that parametric statistical synthesis, combined with a chironomic interface, offers an efficient solution for expressive speech synthesis, as well as a powerful tool for the study of prosody

Language is the way in which individuals are able to express ideas, feelings, needs, expectations, and form relationships with others in their surrounding environment. A disruption in language development may negatively impact a child's social development. Research shows that children with language delays or disorders tend to have increased social and behavioral difficulties (Cantwell & Baker, 1977; Caulfield, Fischel, DeBaryshe, & Whitehurst, 1989). However, research has not examined temperament differences in young children with language delays or disorders. The question this study sought to answer was: Is there a significant difference in the dimensions of temperament between children with a history of slow expressive language development and their peers with normal language development? The subjects in this study ranged in age from 64 to 74 months. They included 33 children with a history of slow expressive language development (SELD) and 27 children with normal language development. The temperamental characteristics of each of the subjects was assessed by ratings provided by their parents, utilizing the Parent Temperament Questionnaire for Children (Thomas, Chess, & Korn, 1977), a questionnaire to assess the way a child behaves during everyday situations. The question was analyzed by calculating the means and standard deviations for the nine temperament dimensions for· the two groups. To determine if there were significant differences among the two groups, two tailed t-tests were computed at the .05 level of significance. The Parent Temperament Questionnaire for Children (Thomas et al., 1977) showed the children with a history of slow expressive language development to have significantly different scores in the dimensions of approach-withdrawal and intensity of reaction than the subjects with normal language development at 5 years of age.

Cette thèse étudie les rapports entre l’expression d’attitudes agressives contrôlées et la perception de la dominance, à partir d’extraits de séances télévisées du conseil Municipal de Montreuil (93100) durant l’année 2013, période marquée par un climat politique vif et hostile. Un corpus a été constitué à partir d’extraits de parole spontanée de la Maire, Dominique Voynet, et de quatre de ses opposants. Les cinq locuteurs ont participé à l’enregistrement d’une relecture neutre du corpus de leurs propres extraits de parole (25 stimuli par locuteur) ainsi qu’à l’auto-évaluation perceptive de leurs stimuli (profils émotionnels), dont les résultats ont été comparés à l’évaluation perceptive des extraits par des auditeurs naïfs. Tous les locuteurs ont été perçus comme légèrement hostiles, légèrement mécontents et légèrement dominants malgré des profils émotionnels auto-évalués variables. Les extraits originaux et relus ont été comparés au niveau de leur structuration prosodico-syntaxique et de leurs caractéristiques temporelles et mélodiques. Les résultats montrent que 1) certains locuteurs semblent plus s’appuyer sur des paramètres mélodiques et d’autres sur des paramètres temporels ; 2) on peut néanmoins dégager les tendances générales concernant les corrélats dans la parole de l’hostilité et de la dominance dans notre corpus : a) des écarts entre structuration syntaxique et prosodique des extraits, b) la réduction ou l’absence d’allongements syllabiques finaux pré-pausaux, c) de fortes variations de plage de variation de F0 de part et d’autre des pauses silencieuses<br>This dissertation studies the relationship between the expression of controlled aggressive attitudes and the perception of dominance, based on extracts from televised sessions of the Municipal Council of Montreuil during 2013; a period marked by a lively and hostile political climate. We constituted a corpus of spontaneous speech extracts from the Mayor, Dominique Voynet, and four of her opponents. During subsequent recording sessions, the five speakers were asked to read transcriptions of their own speech extracts in a neutral tone (25 stimuli per speaker). They also participated in a self-evaluation questionnaire that focused on the perception of emotional profiles in their own stimuli. The results of the speakers’ perceptive evaluations were compared to those of a control group of naïve listeners. All speakers were perceived as slightly hostile, discontent and dominant, despite the variability observed in the self-evaluated profiles. The original and re-read extracts were compared in their prosodic- syntactic structure as well as their temporal and melodic characteristics. We show that: 1) some speakers seem to rely mostly on melodic parameters whereas others primarily use temporal parameters, 2) nevertheless, general trends emerge regarding the speech correlates of hostility and dominance in our corpus, notably: a) discrepancies between the syntactic and the prosodic structure of the extracts, b) reduction or absence of pre-pausal final syllabic lengthening, c) large variations in F0 range on both sides of silent pauses

Dans ces travaux de thèse nous abordons l'expressivité de la parole lue avec un type de données particulier qui sont les livres audio. Les livres audio sont des enregistrements audio d’œuvres littéraires fait par des professionnels (des acteurs, des chanteurs, des narrateurs professionnels) ou par des amateurs. Ces enregistrements peuvent être destinés à un public particulier (aveugles ou personnes mal voyantes). La disponibilité de ce genre de données en grande quantité avec une assez bonne qualité a attiré l'attention de la communauté scientifique en traitement automatique du langage et de la parole en général, ainsi que des chercheurs spécialisés dans la synthèse de parole expressive. Pour explorer ce vaste champ d'investigation qui est l'expressivité, nous proposons dans cette thèse d'étudier trois entités élémentaires de l'expressivité qui sont véhiculées par les livres audio: l'émotion, les variations liées aux changements discursifs et les propriétés du locuteur. Nous traitons ces patrons d'un point de vue prosodique. Les principales contributions de cette thèse sont la construction d'un corpus de livres audio comportant un nombre important d'enregistrements partiellement annotés par un expert, une étude quantitative caractérisant les émotions dans ce type de données, la construction de modèles basés sur des techniques d'apprentissage automatique pour l'annotation automatique de types de discours et enfin nous proposons une représentation vectorielle de l'identité prosodique d'un locuteur dans le cadre de la synthèse statistique paramétrique de la parole<br>In this thesis, we study the expressivity of read speech with a particular type of data, which are audiobooks. Audiobooks are audio recordings of literary works made by professionals (actors, singers, professional narrators) or by amateurs. These recordings may be intended for a particular audience (blind or visually impaired people). The availability of this kind of data in large quantities with a good enough quality has attracted the attention of the research community in automatic speech and language processing in general and of researchers specialized in expressive speech synthesis systems. We propose in this thesis to study three elementary entities of expressivity that are conveyed by audiobooks: emotion, variations related to discursive changes, and speaker properties. We treat these patterns from a prosodic point of view. The main contributions of this thesis are: the construction of a corpus of audiobooks with a large number of recordings partially annotated by an expert, a quantitative study characterizing the emotions in this type of data, the construction of a model based on automatic learning techniques for the automatic annotation of discourse types and finally we propose a vector representation of the prosodic identity of a speaker in the framework of parametric statistical speech synthesis

U disertaciji su opisani postupci sinteze ekspresivnog govorakorišćenjem parametarskih pristupa. Pokazano je da se korišćenjemdubokih neuronskih mreža dobijaju bolji rezultati nego korišćenjemskrivenix Markovljevih modela. Predložene su tri nove metode zasintezu ekspresivnog govora korišćenjem dubokih neuronskih mreža:metoda kodova stila, metoda dodatne obuke mreže i arhitekturazasnovana na deljenim skrivenim slojevima. Pokazano je da se najboljirezultati dobijaju korišćenjem metode kodova stila. Takođe jepredložana i nova metoda za transplantaciju emocija/stilovabazirana na deljenim skrivenim slojevima. Predložena metodaocenjena je bolje od referentne metode iz literature.<br>In this thesis methods for expressive speech synthesis using parametricapproaches are presented. It is shown that better results are achived withusage of deep neural networks compared to synthesis based on hiddenMarkov models. Three new methods for synthesis of expresive speech usingdeep neural networks are presented: style codes, model re-training andshared hidden layer architecture. It is shown that best results are achived byusing style code method. The new method for style transplantation based onshared hidden layer architecture is also proposed. It is shown that thismethod outperforms referent method from literature.

Notre objectif est de simuler des comportements multimodaux expressifs pour les agents conversationnels animés ACA. Ceux-ci sont des entités dotées de capacités affectives et communicationnelles; ils ont souvent une apparence humaine. Quand un ACA parle ou rit, il est capable de montrer de façon autonome des comportements multimodaux pour enrichir et compléter son discours prononcé et transmettre des informations qualitatives telles que ses émotions. Notre recherche utilise les modèles d’apprentissage à partir données. Un modèle de génération de comportements multimodaux pour un personnage virtuel parlant avec des émotions différentes a été proposé ainsi qu’un modèle de simulation du comportement de rire sur un ACA. Notre objectif est d'étudier et de développer des générateurs d'animation pour simuler la parole expressive et le rire d’un ACA. En partant de la relation liant prosodie de la parole et comportements multimodaux, notre générateur d'animation prend en entrée les signaux audio prononcés et fournit en sortie des comportements multimodaux. Notre travail vise à utiliser un modèle statistique pour saisir la relation entre les signaux donnés en entrée et les signaux de sortie; puis cette relation est transformée en modèle d’animation 3D. Durant l'étape d’apprentissage, le modèle statistique est entrainé à partir de paramètres communs qui sont composés de paramètres d'entrée et de sortie. La relation entre les signaux d'entrée et de sortie peut être capturée et caractérisée par les paramètres du modèle statistique. Dans l'étape de synthèse, le modèle entrainé est utilisé pour produire des signaux de sortie (expressions faciale, mouvement de tête et du torse) à partir des signaux d'entrée (F0, énergie de la parole ou pseudo-phonème du rire). La relation apprise durant la phase d'apprentissage peut être rendue dans les signaux de sortie. Notre module proposé est basé sur des variantes des modèles de Markov cachés (HMM), appelées HMM contextuels. Ce modèle est capable de capturer la relation entre les mouvements multimodaux et de la parole (ou rire); puis cette relation est rendue par l’animation de l’ACA<br>Our aim is to render expressive multimodal behaviors for Embodied conversational agents, ECAs. ECAs are entities endowed with communicative and emotional capabilities; they have human-like appearance. When an ECA is speaking or laughing, it is capable of displaying autonomously behaviors to enrich and complement the uttered speech and to convey qualitative information such as emotion. Our research lies in the data-driven approach. It focuses on generating the multimodal behaviors for a virtual character speaking with different emotions. It is also concerned with simulating laughing behavior on an ECA. Our aim is to study and to develop human-like animation generators for speaking and laughing ECA. On the basis of the relationship linking speech prosody and multimodal behaviors, our animation generator takes as input human uttered audio signals and output multimodal behaviors. Our work focuses on using statistical framework to capture the relationship between the input and the output signals; then this relationship is rendered into synthesized animation. In the training step, the statistical framework is trained based on joint features, which are composed of input and of output features. The relation between input and output signals can be captured and characterized by the parameters of the statistical framework. In the synthesis step, the trained framework is used to produce output signals (facial expression, head and torso movements) from input signals (F0, energy for speech or pseudo-phoneme of laughter). The relation captured in the training phase can be rendered into the output signals. Our proposed module is based on variants of Hidden Markov Model (HMM), called Contextual HMM. This model is capable of capturing the relationship between human motions and speech (or laughter); then such relationship is rendered into the synthesized animations

Les travaux de cette thèse portent sur la modélisation des émotions pour la synthèse audiovisuelle expressive de la parole à partir du texte. Aujourd’hui, les résultats des systèmes de synthèse de la parole à partir du texte sont de bonne qualité, toutefois la synthèse audiovisuelle reste encore une problématique ouverte et la synthèse expressive l’est encore d’avantage. Nous proposons dans le cadre de cette thèse une méthode de modélisation des émotions malléable et flexible, permettant de mélanger les émotions comme on mélange les teintes sur une palette de couleurs. Dans une première partie, nous présentons et étudions deux corpus expressifs que nous avons construits. La stratégie d’acquisition ainsi que le contenu expressif de ces corpus sont analysés pour valider leur utilisation à des fins de synthèse audiovisuelle de la parole. Dans une seconde partie, nous proposons deux architectures neuronales pour la synthèse de la parole. Nous avons utilisé ces deux architectures pour modéliser trois aspects de la parole : 1) les durées des sons, 2) la modalité acoustique et 3) la modalité visuelle. Dans un premier temps, nous avons adopté une architecture entièrement connectée. Cette dernière nous a permis d’étudier le comportement des réseaux de neurones face à différents descripteurs contextuels et linguistiques. Nous avons aussi pu analyser, via des mesures objectives, la capacité du réseau à modéliser les émotions. La deuxième architecture neuronale proposée est celle d’un auto-encodeur variationnel. Cette architecture est capable d’apprendre une représentation latente des émotions sans utiliser les étiquettes des émotions. Après analyse de l’espace latent des émotions, nous avons proposé une procédure de structuration de ce dernier pour pouvoir passer d’une représentation par catégorie vers une représentation continue des émotions. Nous avons pu valider, via des expériences perceptives, la capacité de notre système à générer des émotions, des nuances d’émotions et des mélanges d’émotions, et cela pour la synthèse audiovisuelle expressive de la parole à partir du texte<br>: The work of this thesis concerns the modeling of emotions for expressive audiovisual textto-speech synthesis. Today, the results of text-to-speech synthesis systems are of good quality, however audiovisual synthesis remains an open issue and expressive synthesis is even less studied. As part of this thesis, we present an emotions modeling method which is malleable and flexible, and allows us to mix emotions as we mix shades on a palette of colors. In the first part, we present and study two expressive corpora that we have built. The recording strategy and the expressive content of these corpora are analyzed to validate their use for the purpose of audiovisual speech synthesis. In the second part, we present two neural architectures for speech synthesis. We used these two architectures to model three aspects of speech : 1) the duration of sounds, 2) the acoustic modality and 3) the visual modality. First, we use a fully connected architecture. This architecture allowed us to study the behavior of neural networks when dealing with different contextual and linguistic descriptors. We were also able to analyze, with objective measures, the network’s ability to model emotions. The second neural architecture proposed is a variational auto-encoder. This architecture is able to learn a latent representation of emotions without using emotion labels. After analyzing the latent space of emotions, we presented a procedure for structuring it in order to move from a discrete representation of emotions to a continuous one. We were able to validate, through perceptual experiments, the ability of our system to generate emotions, nuances of emotions and mixtures of emotions, and this for expressive audiovisual text-to-speech synthesis

Notre objectif est de simuler des comportements multimodaux expressifs pour les agents conversationnels animés ACA. Ceux-ci sont des entités dotées de capacités affectives et communicationnelles; ils ont souvent une apparence humaine. Quand un ACA parle ou rit, il est capable de montrer de façon autonome des comportements multimodaux pour enrichir et compléter son discours prononcé et transmettre des informations qualitatives telles que ses émotions. Notre recherche utilise les modèles d’apprentissage à partir données. Un modèle de génération de comportements multimodaux pour un personnage virtuel parlant avec des émotions différentes a été proposé ainsi qu’un modèle de simulation du comportement de rire sur un ACA. Notre objectif est d'étudier et de développer des générateurs d'animation pour simuler la parole expressive et le rire d’un ACA. En partant de la relation liant prosodie de la parole et comportements multimodaux, notre générateur d'animation prend en entrée les signaux audio prononcés et fournit en sortie des comportements multimodaux. Notre travail vise à utiliser un modèle statistique pour saisir la relation entre les signaux donnés en entrée et les signaux de sortie; puis cette relation est transformée en modèle d’animation 3D. Durant l'étape d’apprentissage, le modèle statistique est entrainé à partir de paramètres communs qui sont composés de paramètres d'entrée et de sortie. La relation entre les signaux d'entrée et de sortie peut être capturée et caractérisée par les paramètres du modèle statistique. Dans l'étape de synthèse, le modèle entrainé est utilisé pour produire des signaux de sortie (expressions faciale, mouvement de tête et du torse) à partir des signaux d'entrée (F0, énergie de la parole ou pseudo-phonème du rire). La relation apprise durant la phase d'apprentissage peut être rendue dans les signaux de sortie. Notre module proposé est basé sur des variantes des modèles de Markov cachés (HMM), appelées HMM contextuels. Ce modèle est capable de capturer la relation entre les mouvements multimodaux et de la parole (ou rire); puis cette relation est rendue par l’animation de l’ACA<br>Our aim is to render expressive multimodal behaviors for Embodied conversational agents, ECAs. ECAs are entities endowed with communicative and emotional capabilities; they have human-like appearance. When an ECA is speaking or laughing, it is capable of displaying autonomously behaviors to enrich and complement the uttered speech and to convey qualitative information such as emotion. Our research lies in the data-driven approach. It focuses on generating the multimodal behaviors for a virtual character speaking with different emotions. It is also concerned with simulating laughing behavior on an ECA. Our aim is to study and to develop human-like animation generators for speaking and laughing ECA. On the basis of the relationship linking speech prosody and multimodal behaviors, our animation generator takes as input human uttered audio signals and output multimodal behaviors. Our work focuses on using statistical framework to capture the relationship between the input and the output signals; then this relationship is rendered into synthesized animation. In the training step, the statistical framework is trained based on joint features, which are composed of input and of output features. The relation between input and output signals can be captured and characterized by the parameters of the statistical framework. In the synthesis step, the trained framework is used to produce output signals (facial expression, head and torso movements) from input signals (F0, energy for speech or pseudo-phoneme of laughter). The relation captured in the training phase can be rendered into the output signals. Our proposed module is based on variants of Hidden Markov Model (HMM), called Contextual HMM. This model is capable of capturing the relationship between human motions and speech (or laughter); then such relationship is rendered into the synthesized animations