No description available.

No description available.

Emotion adds an important element to the discussion of how information is conveyed and processed by humans; indeed, it plays an important role in the contextual understanding of messages. This research is centered on investigating relevant features for affect classification, along with modeling the multimodal and multitemporal nature of emotion. The use of formant-based features for affect classification is explored. Since linear predictive coding (LPC) based formant estimators often encounter problems with modeling speech elements, such as nasalized phonemes and give inconsistent results for bandwidth estimation, a robust formant-tracking algorithm was introduced to better model the formant and spectral properties of speech. The algorithm utilizes Gaussian mixtures to estimate spectral parameters and refines the estimates using maximum a posteriori (MAP) adaptation. When the method was used for features extraction applied to emotion classification, the results indicate that an improved formant-tracking method will also provide improved emotion classification accuracy. Spectral features contain rich information about expressivity and emotion. However, most of the recent work in affective computing has not progressed beyond analyzing the mel-frequency cepstral coefficients (MFCC’s) and their derivatives. A novel method for characterizing spectral peaks was introduced. The method uses a multi-resolution sinusoidal transform coding (MRSTC). Because of MRSTC’s high precision in representing spectral features, including preservation of high frequency content not present in the MFCC’s, additional resolving power was demonstrated. Facial expressions were analyzed using 53 motion capture (MoCap) markers. Statistical and regression measures of these markers were used for emotion classification along the voice features. Since different modalities use different sampling frequencies and analysis window lengths, a novel classifier fusion algorithm was introduced. This algorithm is intended to integrate classifiers trained at various analysis lengths, as well as those obtained from other modalities. Classification accuracy was statistically significantly improved using a multimodal-multitemporal approach with the introduced classifier fusion method. A practical application of the techniques for emotion classification was explored using social dyadic plays between a child and an adult. The Multimodal Dyadic Behavior (MMDB) dataset was used to automatically predict young children’s levels of engagement using linguistic and non-linguistic vocal cues along with visual cues, such as direction of a child’s gaze or a child’s gestures. Although this and similar research is limited by inconsistent subjective boundaries, and differing theoretical definitions of emotion, a significant step toward successful emotion classification has been demonstrated; key to the progress has been via novel voice and visual features and a newly developed multimodal-multitemporal approach.

Thesis (MSL and HT (Interdisciplinary Health Sciences. Speech-Language and Hearing Therapy))--Stellenbosch University, 2008. / This study aims to describe the speech processing skills of three children ages 6;0, 6;10 and 8; 10, with cochlear implants. A psycholinguistic framework was used to profile each child’s strengths and weaknesses, using a single case study approach. Each child’s speech processing skills are described based on detailed psycholinguistically-orientated assessments. In addition, retrospective data from 1-2 years post-implantation were examined in the light of the psycholinguistic framework in order to describe each child’s development over time and in relation to time of implantation. Results showed each child to have a unique profile of strengths and weaknesses, and widely varying outcomes in terms of speech processing even though all three children had the same initial difficulty (congenital bilateral hearing loss). Links between speech processing and other aspects of development as well as contextual factors are discussed in relation to outcomes for each child. The case studies contribute to knowledge of speech processing skills in children with cochlear implants, and have clinical implications for those who work with children with cochlear implants and their families.

The productions of adult second language (L2) learners are often detected as having a foreign accent by native speakers of the target language. However, there is no clear answer for what kind of problems contribute to L2 learners’ foreign accent. This thesis aims to investigate potential factors behind foreign accent. We intend to achieve this goal by examining cross-linguistic empirical evidence of the L2 acquisition of Japanese lexical pitch accent by English learners. L2 prosody has been found to significantly influence native speakers’ auditory impression of foreign accent. L2 prosody also allows us to test crosslinguistic differences in the function of the key acoustic correlates of L2 contrasts. In this thesis we examine F0, which signals both lexical pitch accent and phrasal distinctions in Japanese, but which signals only phrasal distinctions, not lexical distinctions, in English. For adult L2 learners to achieve target-like productions, the literature suggests that three abilities are the key factors: 1) learners’ ability to differentiate the acoustic correlate of the target L2 contrasts, 2) ability to articulate the acoustic correlate of the target L2 contrasts and 3) ability to categorize the target L2 contrasts. This thesis evaluates all three of these potential factors. The main contribution of this thesis is to provide a comprehensive view of foreign accent, by investigating possible interactions between the factors and by examining the different abilities of the same learners. Another contribution is to provide empirical evidence for the nature of learners’ problems with foreign accent during L2 acquisition, by testing two groups of English learners of Japanese (experienced and inexperienced) in comparison with Japanese native speakers. The first experiment used intelligibility scores and overall F0 patterns to quantify the degree of foreign accent in the learners’ productions of Japanese lexical pitch accent. The second experiment showed that the learners’ ability to differentiate F0 contours in a nonspeech context was equal to that of the native speakers. The third experiment showed that the learners’ ability to articulate the F0 contours in a non-speech context differed from that of the native speakers. The fourth experiment showed that although learners were able to hear the phonetic differences between the target L2 contrasts, due to poor formation of the target L2 categories and poor lexical assignment ability, the inexperienced learners seem to have greater difficulty than experienced learners both in categorizing boundary items into the target L2 categories and in assigning the L2 categories to lexical items. Overall, the foreign accent of adult L2 learners’ productions is explained through a combination of articulation and categorization factors. Importantly, this cross-sectional study has indicated how learners’ problems with foreign accent change as they gain L2 experience. Whereas experienced learners seem to have problems mainly in the articulation and phonetic realization of the L2 contrasts, the inexperienced learners seem to have mainly problems in phonetic and lexical-phonological representations of the target L2 categories in addition to articulation and phonetic realization. This study offers both theoretical insights for the field of L2 speech acquisition research and also practical insights for the L2 classroom.

How do bi/multilinguals switch between languages so effectively that there is no obvious intrusion from the alternatives? One can examine this by comparing language selection with task selection, or language switching with task switching. This is the approach adopted in the first of two strands of research presented in this thesis. In task switching, providing advance warning of the task typically leads to a reduction in the performance ‘switch cost’, suggesting top-down biasing of task selection. It is not clear whether the language switch cost also reduces with preparation, partly because there have been very few attempts to examine preparation for a language switch, and partly because these attempts suffered from non-trivial methodological drawbacks. In Experiments 1-3 I used an optimised picture naming paradigm in which language changed unpredictably and was specified by a language cue presented at different intervals before the picture. Experiment 1, conducted on ‘unbalanced’ bilinguals, revealed some evidence of reduction in the language switch cost for naming times with preparation, but only when cue duration was short. In an attempt to further optimise the paradigm, in Experiment 2 the cue-stimulus interval (which was varied from trial to trial in Experiment 1), was varied over blocks instead. Visual cues were replaced with auditory cues – the latter also enabled a comparison between semantically transparent word cues (the spoken names of the languages) and less transparent cues (fragments of national anthems). Experiment 2 revealed a reduction in switch cost with preparation for naming latencies, but only in the second language; the first language showed the reverse. To examine whether the increase in switch cost with preparation in the first language could be due to unbalanced bilinguals biasing processing towards L2, balanced bilinguals were tested in Experiment 3. This revealed a robust reduction in switch cost in naming latencies for both languages, which was driven primarily by the trials with the anthem cues. However, in the error rates the switch cost increased with preparation interval, thus complicating the interpretation of the reduction observed for response times. Experiment 4 investigated whether preparation for a language switch elicits the electrophysiological patterns commonly found during preparation for a task switch – a switch-induced positive polarity Event-Related Potential (ERP) with a posterior scalp distribution. Contrary to a recent report of the absence of the posterior positivity in language switching, it was clearly present in the present EEG data. As in task switching, the amplitude of the posterior positivity predicted performance. The electrophysiological data suggest that preparation for a language switch and preparation for a task switch rely on highly overlapping control mechanisms. The behavioural data suggest that advance control can be effective in language switching, but perhaps not as effective as in task switching. Experiments 1-3 also examined the effect of stimulus associative history – whether the language used on the previous encounter with a given stimulus influenced performance on the current trial). Having previously named a given picture in the same language benefited overall performance, but did not do so more for switches than repeats. Thus, stimulus associative history does not seem to contribute to the language switch cost. The second strand of my research asked whether bilinguals can set themselves independently for speech vs. comprehension. Previous research has examined the cost of switching the language in output tasks and in input tasks. But, it is not clear whether one can apply separate control settings for input and output selection. To investigate this, I used a paradigm that combined switching languages for speech production and comprehension. My reasoning was that, if there is cross-talk between the control settings for input vs. output, performance in one pathway should benefit if the language selected for the other pathway is the same relative to when it is different: a ‘language match effect’. Conversely, if there is no cross-talk, there should not be a language match effect. In Experiment 5 bilinguals alternated predictably between naming numbers in their first and second language (in runs of 3 trials), whilst also having to semantically categorise spoken words which occasionally (and unpredictably) replaced the numbers. The language of the categorisation ‘probes’ varied over blocks of ~17 naming runs, but was constant within a block. The results showed a clear match effect in the input task (categorisation), but not the output task (naming). To examine the potential role of proficiency, Experiment 6 used the same paradigm to test unbalanced and balanced bilinguals. The pattern of results was qualitatively similar in both groups to that observed in Experiment 5: a language match effect confined to the input task. These results suggest ‘leakage’ from the output control settings into the input control settings.

O presente trabalho foi realizado em dois estudos. O primeiro deles (Estudo 1) teve como objetivo descrever e comparar as características acústicas das líquidas /l/; /4/ e /"do português brasileiro enquanto o segundo (Estudo 2), procurou descrever e comparar a percepção auditiva das líquidas apresentadas a juízes. Para isto, foi analisada acusticamente a produção das líquidas /l/; /4/ e /"realizada por 10 crianças sem alterações de fala e linguagem (GC), com idade compreendida entre 5:11 e 9:0 anos, e 10 com transtorno fonológico (GTF) e idade entre 5:0 e 12:0 anos. Após a avaliação fonológica, as crianças dos dois grupos foram solicitadas a repetir três vezes as sílabas /Na, Ni, Nu/; /4a, 4i, 4u/ e /"a, "i, "u/ e as palavras /se\'boNa/, /\'Nãma/, /\'mi"W, /<aka\'4\'/, /<i\'4afa/, /pa\'"asu/. A amostra de fala foi coletada e armazenada no CSL-Computarized Speech Laboratory 4300B/4500. Os parâmetros acústicos selecionados para a análise acústica do Estudo 1 foram: valores das freqüências formantes F1, F2 e F3, duração do som-alvo, duração da sílaba contendo o som-alvo, duração do som anterior (quando presente) e do subseqüente, duração da porção estável do som-alvo, valores da transição do som-alvo para a vogal subseqüente (/l/) e duração da transição do som-alvo para a vogal subseqüente (/l/), transição do som-alvo para o som inserido (/4/ e /") e do sominserido para a vogal subseqüente (/4/ e /"), duração da transição do som-alvo para o som inserido (/4/ e /"), slope (medida de velocidade de movimentação dos articuladores). Em seguida à análise dos dados, uma amostra de fala foi selecionada pela pesquisadora para o Estudo 2. Esta amostra foi apresentada a dois grupos de juízes da FMUSP, um composto por alunos do curso de graduação e outro por alunos de pós-graduação, todos fonoaudiólogos, que deveriam realizar um julgamento perceptivo-auditivo das sílabas e palavras. Os dados foram analisados estatisticamente por meio dos testes ANOVA, Teste-t e qui-quadrado (nível de significância adotado de 5%). Em relação às medidas acústicas, os resultados indicaram que as crianças do GC, de modo geral, foram capazes de produzir as sílabas e palavras solicitadas de maneira mais estável e precisa do que as crianças do GTF. A análise do slope mostrou que o GC apresentou movimentos mais rápidos e precisos, do ponto de vista articulatório, quando comparado ao GTF. Há evidencias que o GC se utilizou dos parâmetros acústicos de duração e das freqüências formantes para diferenciar uma líquida da outra, enquanto o GTF utilizou predominantemente a duração como diferenciador. Observando o comportamento acústico-articulatório do GC verificou-se que a língua apresenta uma maior movimentação ascendente e descendente durante a produção do /N/ e uma mudança ântero-posterior mais acentuada na produção do /4/. Em relação à análise perceptiva foi possível notar que, apesar do maior número de erros ter sido observado no grupo de juízes da pós-graduação, ambos tiveram mais dificuldade de julgar perceptivamente as sílabas e palavras com a líquida /4/. / The present research was divided in two different studies. The purpose of the first one (Study 1) was to describe and to compare the acoustic characteristics of the three Brazilian Portuguese liquid sounds /N/; /4/ and/"/. The aim of the second study (Study 2) was to describe and to compare the auditory perception of these liquids when presented to judges. To accomplish these two purposes, acoustic analyses were carried out of the production of /N/; /4/ and /"/ by 10 children without phonological disorders aged from 5:11 to 9:0 years old (CG) and 10 children with phonological disorders (PDG) aged between 5:0 and 12:0 years old. After a complete phonological evaluation the children in of both groups were asked to repeat three times the syllables /Na, Ni, Nu/; /4a, 4i, 4u/ e /"a, "i, "u/ and the words /se\'boNa/-onion, /\'Nãma/-mud, /\'mi"W-corn, /<aka\'4\'/-alligator, /<i\'4afa/-giraffe, /pa\'"asu/-clown. The samples were collected and stored in the CSL-Computerized Speech Laboratory 4300B/4500. The acoustic parameters selected to be analyzed for Study 1 were: formant frequencies F1, F2 and F3, duration of the target sound, duration of the syllable containing the liquid target sound, duration of the precedent (when present) and subsequent sound to the liquid, duration of the steady state from the target sound, transition of the target sound to the subsequent vowel (for /l/) and duration of transition of the target sound to the subsequent vowel (for /l/), transition of the target sound to the inserted sound (for /4/ e /") and from the target sound to the subsequent vowel (for /4/ e /"), duration of the transition from the target sound to the inserted sound (for /4/ e /") and slope analysis (measurement of rate of formantfrequency change ). After data analysis the researcher selected a speech sample for Study 2. The sample was presented to two groups of judges from FMUSP. One group was composed of graduate students and the other one by post-graduate students, all speech-pathologists. All of them were requested to perceptually judge syllables and words. Data were statistically analyzed using ANOVA, T test and chisquare (significance level adopted 5%). In general terms, results of Study 1 indicated that CG produced solicited syllables and words with more stability and precision than PDG. From the articulatory point of view, slope analysis showed that CG had steeper formant-frequency changes, and therefore presumably faster and more precise movements than PDG. Some evidence shows that CG used duration and formant frequencies to differentiate their liquid productions while PDG used duration prior to other acoustic parameters. Observing acoustic-articulatory parameters it was observed that CG presented ample ascendant and descendent movements during /N/ production while anterior and posterior movements were more visible during /4/ production. Study 2 results indicated that, even though the number of errors was greater for the post-graduation group of judges than for the graduate group, both of them presented more difficulty in perceptually judging the sound /4/ presented in syllables and words.

A fluência de fala varia de indivíduo para indivíduo, fluente ou gago, dependendo de diversos fatores. O objetivo desta Tese foi verificar a variação da fluência da fala em falantes do Português Brasileiro. Participaram deste estudo 594 indivíduos fluentes, 336 do gênero feminino e 258 do gênero masculino com idades entre 2:0 e 99:11 anos, residentes na cidade de São Paulo. As variáveis da fluência analisadas foram: tipologia das rupturas, velocidade de fala e porcentagem de descontinuidade de fala. Esta Tese foi dividida em quatro estudos. O primeiro estudo teve como objetivo verificar o perfil evolutivo da fluência da fala. De acordo com os resultados o padrão de rupturas de fala não sofre grande variabilidade entre os grupos etários, enquanto que a velocidade de fala varia ao longo das fases da vida, podendo indicar aquisição, desenvolvimento, estabilização e degeneração. O segundo estudo teve como objetivo conhecer a variação da fluência segundo o gênero e as fases da adolescência (adolescência inicial - 12 a 14 anos; e adolescência final - 5 a 17 anos). Os gêneros não se diferenciam para nenhuma das variáveis analisadas, enquanto que as fases da adolescência se diferenciam quanto a todas as variáveis. Ocorre uma diminuição nas rupturas de fala e um aumento na velocidade de fala com o aumento da idade. O terceiro estudo teve como objetivo verificar a influência do gênero e do nível de escolaridade (fundamental, médio e superior) na fluência da fala de adultos. De acordo com os resultados o gênero e o nível de escolaridade não são fatores determinantes para a fluência da fala em adultos fluentes. Observa-se variação isolada para interjeição, revisão e prolongamento. Os indivíduos de Nível Superior apresentam menos interjeições e revisões e mais prolongamentos. O quarto estudo teve como objetivo conhecer as especificidades do Perfil da Fluência dos idosos em relação a todos os parâmetros avaliados. Os idosos foram comparados quanto às décadas de vida e pelo agrupamento em idosos e idosos com mais de 80 anos. De acordo com os resultados observa-se um aumento estatisticamente significativo nas taxas de ruptura e uma diminuição na velocidade de fala para os idosos com mais de 80 anos. De maneira geral percebe-se que o fator determinante na fluência da fala é a idade, principalmente quanto à velocidade de fala. O gênero e o nível de escolaridade parecem irrelevantes. / Speech fluency varies from one individual to the next, fluent or stutterer, depending on several factors. The purpose of the present Thesis was to verify variations in speech fluency of Brazilian Portuguese speakers. Participants of this study were 594 fluent individuals, 336 female and 258 male, with ages between 2:0 and 99:11 years, residents in the city of São Paulo. The analyzed fluency variables were: typology of speech disruption, speech rate and percentage of speech discontinuity. This Thesis was divided in four studies. The first study had as a purpose to verify the development profile of speech fluency. According to the results, the speech typology pattern does not present great variability in the different age groups. Speech rate, however, varies along the stages of life suggesting a pattern of acquisition, development, stabilization and deterioration. The second study had as a purpose to investigate fluency variations according to gender and phases of adolescence (early adolescence - 12 to 14 years; and late adolescence - 15 to 17 years). Fluency variables did not present variations according to gender, but did present differences - all variables - regarding the phases of adolescence. A decrease in speech disruptions is observed as well as an increase in speech rate with the increase in age. The third study has as a purpose to verify the influence of gender and literacy (fundamental, average and higher education) in the speech fluency of adults. According to the results, gender and literacy are not determinant factors for the speech fluency of fluent adults. An isolated variation is observed for interjection, revision and prolongation. Individuals with higher education present fewer interjections and revisions, and more prolongations. The fourth study had as a purpose to verify the Fluency Profile of elderlies regarding all of the assessed parameters. Elderlies were compared between themselves regarding the decades of life and by the division elderlies and elderlies with more than 80 years of age. According to the results, a statistically significant increase in speech disruptions is observed as well as a decrease in speech rate for elderlies more than 80 years old. Overall, age seems to be a determinant factor for speech fluency, especially in terms of speech rate. Gender and literacy seem to be irrelevant.

OBJETIVO: Verificar se ocorrem modificações nos parâmetros vocais perceptivos, acústicos e no Voice Onset Time das consoantes plosivas em adultos implantados com deficiência auditiva pré-lingual, após reabilitação vocal específica. MÉTODO: Participaram 20 adultos alfabetizados com deficiência auditiva sensorioneural de severa a profunda bilateral pré-lingual implantados tardiamente, com uso fluente de linguagem oral para a comunicação, com idades entre 17 e 48 anos. Todos apresentavam exame laringológico normal e limiares auditivos com implante coclear melhores que 40dBHL. Os indivíduos foram aleatoriamente distribuídos em dois grupos: Grupo 1 (estudo) e Grupo 2 (controle), com 10 pacientes em cada um deles, sendo 5 pacientes do gênero masculino e 5 do gênero feminino, semelhantes na média de idade e no tempo de privação auditiva anterior ao implante coclear. Os pacientes do Grupo 1 foram submetidos a um protocolo de terapia de voz e fala composto por 12 sessões individuais, com a mesma terapeuta, utilizando exercícios vocais e treinamento da produção de consoantes e do Voice Onset Time (VOT) com auxílio da espectrografia. O Grupo 2 realizou apenas gravações de voz. Foram realizadas gravações da voz antes e após a participação no protocolo de terapia para o Grupo 1, e após o mesmo período, 3 meses depois, sem qualquer intervenção, para o Grupo 2. Para as sessões de gravações da voz, utilizamos leitura das frases do protocolo CAPE-V, emissão da vogal sustentada /a/ e leitura de frases veículo com as seis plosivas do português brasileiro em posição inicial da palavra. Foi realizada avaliação perceptivo-auditiva das vozes por três juízes, e análise acústica por meio do programa PRAAT. RESULTADOS: Observou-se redução estatisticamente significante no grau geral da voz, na instabilidade vocal e no grau de alteração da ressonância após a reabilitação vocal no Grupo 1. Foi possível comprovar estatisticamente que os indivíduos do sexo feminino do Grupo 1 se diferenciaram do Grupo 2 nos valores da variabilidade da frequência fundamental (F0). Em relação à análise do VOT, o Grupo 1 apresentou modificação estatisticamente significante na melhora da produção das consoantes sonoras [b] e [d]. O Grupo 2 não apresentou mudanças significantes em quaisquer dos parâmetros analisados. CONCLUSÃO: Após a intervenção vocal, os implantados adultos submetidos à reabilitação vocal apresentaram modificações nos parâmetros perceptivo-auditivos e acústicos, com redução do grau geral da voz, da instabilidade vocal, do grau da alteração da ressonância, dos valores da variabilidade da F0 na emissão da vogal sustentada e do VOT das consoantes sonoras [b] e [d] / OBJECTIVE: To verify changes in the perceptual and acoustic vocal parameters and in the Voice Onset Time of plosive consonants in pre-lingual deaf adults with cochlear implants, after specific vocal rehabilitation. METHODS: 20 literate adults with bilateral severe to profound pre-lingual sensorineural hearing loss participated in the study; individuals were implanted late and were fluent users of oral language. Ages ranged from 17 to 48 years. All individuals presented normal results in laryngoscopy, and hearing thresholds with the cochlear implant better than 40dBHL. Individuals were randomly distributed in two groups: Group 1 (study) and Group 2 (control) both with ten patients, five men and five women, matching mean age and hearing deprivation time before the cochlear implantation. Patients from Group 1 underwent a protocol of vocal and speech therapy including 12 individual sessions with the same therapist, comprising vocal exercises and plosive consonants production training, and the Voice Onset Time (VOT) with the support of the acoustic spectrogram. Group 2 only underwent vocal recordings. The vocal recordings occurred before and after the participation in the therapy protocol for Group 1 and after the same period, three months later without any intervention, for Group 2. The recording sessions used the CAPE-V protocol sentence reading, emission of sustained vowel /a/ and reading of carrier phrases with the six plosive consonants of the Brazilian Portuguese language in the initial position of the word. Auditory-perceptual evaluation of voices was performed by three judges and the acoustical analysis used the PRATT program. RESULTS: A statistically significant reduction in the overall vocal degree, in the vocal instability and in the degree of resonance change was observed after vocal rehabilitation in Group 1. It was possible to statistically evidence that female individuals from Group 1 differed from the ones in Group 2 regarding the modification of variability values of fundamental frequency (F0). Concerning the VOT analysis, Group 1 presented statistically significant improvement in the production of consonants [b] and [d]. Group 2 did not present significant changes in any of the analyzed parameters. CONCLUSIONS: The implanted adults submitted to vocal rehabilitation presented changes in the auditory-perceptual and acoustic parameters, with reduction of the overall vocal degree, vocal instability, degree of resonance change, in variability values of F0 in the emission of sustained vowel and in VOT of consonants [b] and [d] after vocal intervention

Un modèle tridimensionnel du visage a été élaboré, dans la perspective de contribuer à l'étude de questions importantes sur le contrôle moteur de la production de la parole. Ce modèle est construit sur un maillage constitué d'éléments hexahédraux et de clavettes, qui comporte 3 couches distinctes et est symétrique par rapport au plan medio-sagittal. Les muscles faciaux sont représentés dans le maillage par un sous-ensemble d'éléments contigus. Les propriétés élastiques des éléments du maillage sont décrites par une loi de comportement de type isotrope quasi incompressible et hyperélastique. Dans une première phase de ce travail, pour étudier les conséquences globales de l'activation des muscles oro-faciaux sur les mimiques faciales et les gestes labiaux, et plus particulièrement sur les gestes labiaux en parole, un modèle linéaire de muscle a été élaboré. L'influence des variations de la raideur des tissus mous sur les gestes faciaux a été étudiée. En effet, l'activation des muscles entraîne un raidissement des tissus mous musculaires concernés. Cet effet est pris en compte dans le modèle de muscle par un changement de la loi de comportement hyperélastique avec l'activation musculaire. Une attention particulière a été portée dans cette étude à la production du geste de protrusion/arrondissement des lèvres qui est un geste fondamental dans la production des voyelles arrondies, en particulier en Français. Nous montrons que le raidissement des tissus mous musculaires facilite la production précise de ce geste grâce à l'existence d'un effet de saturation dans la relation entre les activations musculaires et les paramètres géométriques des lèvres qui sont pertinents acoustiquement. Ce résultat souligne l'importance des propriétés dynamiques des articulateurs dans la production des gestes de la parole, et il nous a incités à améliorer encore la modélisation de la source principale de force en production de la parole, c'est-à-dire les muscles. C'est pourquoi, un modèle de muscles plus réaliste a été élaboré qui se fonde sur une loi de comportement transversalement isotrope quasi incompressible et hyperélastique et sur un modèle de muscle de type Hill. Ce modèle a été implémenté dans le logiciel éléments finis ANSYS® grâce à sa fonction de programmation USERMAT. La prise en compte supplémentaire d'une loi caractéristique force-vitesse a permis la modélisation complète d'un modèle de muscle de type Hill. Ceci a été fait sous ANSYS® grâce à sa fonction de programmation USERELEM. Cette loi caractéristique force-vitesse introduit un effet d'amortissement dans le mouvement du muscle du fait d'une atténuation croissante de la force musculaire lorsque la vitesse de compression du muscle augmente. Ce nouvel élément de type muscle a été conçu de manière telle qu'il est possible d'implémenter d'autres modèles de muscles que le modèle de type Hill. C'est pourquoi nous avons aussi implémenté le modèle de Feldman, qui a été utilisé de manière importante à Gipsa-lab dans les dernières années. L'intégration du modèle de Feldman dans une structure à éléments finis a nécessité une reformulation de façon à le rendre compatible avec une modélisation distribuée. Les modèles de Hill et de Feldman ont ensuite été incorporés dans le modèle de visage pour remplacer le modèle linéaire initial. Dans ces conditions les premières simulations du geste de protrusion/arrondissement labial ont donné des résultats réalistes. Finalement une comparaison des résultats obtenus avec le modèle de Hill avec ceux qui génère le modèle de Feldman montrent que les formes labiales finales sont très similaires pour les deux modèles. / Un modèle tridimensionnel du visage a été élaboré, dans la perspective de contribuer à l'étude de questions importantes sur le contrôle moteur de la production de la parole. Ce modèle est construit sur un maillage constitué d'éléments hexahédraux et de clavettes, qui comporte 3 couches distinctes et est symétrique par rapport au plan medio-sagittal. Les muscles faciaux sont représentés dans le maillage par un sous-ensemble d'éléments contigus. Les propriétés élastiques des éléments du maillage sont décrites par une loi de comportement de type isotrope quasi incompressible et hyperélastique. Dans une première phase de ce travail, pour étudier les conséquences globales de l'activation des muscles oro-faciaux sur les mimiques faciales et les gestes labiaux, et plus particulièrement sur les gestes labiaux en parole, un modèle linéaire de muscle a été élaboré. L'influence des variations de la raideur des tissus mous sur les gestes faciaux a été étudiée. En effet, l'activation des muscles entraîne un raidissement des tissus mous musculaires concernés. Cet effet est pris en compte dans le modèle de muscle par un changement de la loi de comportement hyperélastique avec l'activation musculaire. Une attention particulière a été portée dans cette étude à la production du geste de protrusion/arrondissement des lèvres qui est un geste fondamental dans la production des voyelles arrondies, en particulier en Français. Nous montrons que le raidissement des tissus mous musculaires facilite la production précise de ce geste grâce à l'existence d'un effet de saturation dans la relation entre les activations musculaires et les paramètres géométriques des lèvres qui sont pertinents acoustiquement. Ce résultat souligne l'importance des propriétés dynamiques des articulateurs dans la production des gestes de la parole, et il nous a incités à améliorer encore la modélisation de la source principale de force en production de la parole, c'est-à-dire les muscles. C'est pourquoi, un modèle de muscles plus réaliste a été élaboré qui se fonde sur une loi de comportement transversalement isotrope quasi incompressible et hyperélastique et sur un modèle de muscle de type Hill. Ce modèle a été implémenté dans le logiciel éléments finis ANSYS® grâce à sa fonction de programmation USERMAT. La prise en compte supplémentaire d'une loi caractéristique force-vitesse a permis la modélisation complète d'un modèle de muscle de type Hill. Ceci a été fait sous ANSYS® grâce à sa fonction de programmation USERELEM. Cette loi caractéristique force-vitesse introduit un effet d'amortissement dans le mouvement du muscle du fait d'une atténuation croissante de la force musculaire lorsque la vitesse de compression du muscle augmente. Ce nouvel élément de type muscle a été conçu de manière telle qu'il est possible d'implémenter d'autres modèles de muscles que le modèle de type Hill. C'est pourquoi nous avons aussi implémenté le modèle de Feldman, qui a été utilisé de manière importante à Gipsa-lab dans les dernières années. L'intégration du modèle de Feldman dans une structure à éléments finis a nécessité une reformulation de façon à le rendre compatible avec une modélisation distribuée. Les modèles de Hill et de Feldman ont ensuite été incorporés dans le modèle de visage pour remplacer le modèle linéaire initial. Dans ces conditions les premières simulations du geste de protrusion/arrondissement labial ont donné des résultats réalistes. Finalement une comparaison des résultats obtenus avec le modèle de Hill avec ceux qui génère le modèle de Feldman montrent que les formes labiales finales sont très similaires pour les deux modèles.