Face-to-face dialogue is the most natural mode of communication between humans. The combination of human visual perception of expression and perception in changes in intonation provides semantic information that communicates idea, feelings and concepts. The realistic modelling of speech movements, through automatic facial animation, and maintaining audio-visual coherence is still a challenge in both the computer graphics and film industry.

Cette thèse s'inscrit dans une étude sur l’élaboration d'une tête parlante. Nous nous intéressons tout particulièrement à la prédiction du mouvement de coarticulation des lèvres et de la mâchoire. Après avoir analysé les variations intra et interlocuteur des paramètres labiaux de deux corpora audiovisuels, nous avons conçu un algorithme de prédiction de la coarticulation basé sur des règles phonétiques et prenant en considération l'interaction entre les articulateurs. Nous avons ensuite proposé une technique pour synthétiser les mouvements articulatoires des lèvres et de la mâchoire en utilisant un corpus monolocuteur. Le principe de base est la concaténation de séquences élémentaires de type VC...CV qui ont été jugées pertinentes par notre algorithme de prédiction phonétique, et qui sont soit extraites du corpus, soit obtenues par complétion. Nous avons modélisé les mouvements articulatoires par des sigmoïdes qui offrent l'avantage de réduire considérablement la taille du modèle construit et permettent de s'adapter facilement à des vitesses d'élocution ou des stratégies articulatoires particulières tout en conservant les contrastes distinctifs entre les sons successifs et leurs caractéristiques intrinsèques. Afin d'estimer la qualité de notre synthèse, nous avons mesuré les différences entre les signaux réels et synthétisés sur l'ensemble des phrases du corpus et nous avons comparé notre solution avec l’algorithme de Cohen et Massaro. Nous avons montré que notre synthèse est meilleure pour certaines séquences spécifiques de type VCCV où l'anticipation est plus complexe. / This thesis comes within the scope of talking heads. We are particularly interested in the prediction of labial and jaw coarticulation movements. After analyzing intra and inter speaker variability using two corpora, we defined a prediction algorithm for anticipatory coarticulation based on phonetic rules which takes into account interactions between articulators. We then proposed a solution to estimate labial and jaw movements using a one speaker corpus. It consists in concatenating elementary VC...CV sequences selected by our prediction algorithm and either extracted from the corpus or rebuilt by completion. We modeled articulatory movements using sigmoids which offer the advantage of considerably reducing the model size and which are adaptable to speaking rate or articulatory strategies. Additionally, sigmoids are able to keep distinctive contrasts between neighboring segments as well as intrinsic characteristics of the sounds. With the aim of estimating the quality of our synthesis process, we measured differences between real and predicted data for all the sentences of the corpus et we compared our solution with Cohen and Massaro 's algorithm. It turns out that our solution is better for specific VCCV sequences in which anticipation is more complex.

Cette étude permet d’explorer les variations spectrales provoquées par deux types de changements temporels qui résultent respectivement de l’opposition de longueur vocalique (variations intrinsèques) et des variations du débit de parole (variations extrinsèques). Trois protocoles expérimentaux ont été conçus, deux en arabe jordanien et un en français, pour examiner ce phénomène. Un intérêt particulier a été porté aux occlusives produites dans des séquences CVC dans le but d’évaluer la coarticulation de la consonne en position initiale et finale. La durée des voyelles et la fréquence des trois premiers formants au début ou à la fin et au milieu de chaque séquence ont été mesurées dans chaque condition de longueur et de débit. Les équations de locus ont été utilisées afin de décrire la relation CV/VC quand elle subit ces deux types de variations. Selon les résultats, la qualité de la voyelle et de la consonne est influencée par l’opposition de durée et par le débit de parole de manière distincte. Le changement généré par les variations temporelles est évalué à l’aide des équations de locus et à travers des analyses de dispersion. Ces outils révèlent qu’il existe un chevauchement coarticulatoire plus important quand la durée de la voyelle est réduite, peu importe l’origine des variations temporelles dans les deux langues (intrinsèque et extrinsèque). Par contre, les variations temporelles impactent la classification des occlusives de manière distincte selon leur origine. / The aim of this study is to examine the different spectral variations that are impacted by two types of temporal modifications: phonological vowel length contrast (intrinsic variations) and varying speaking rates (extrinsic variations). Three experiments, two on Jordanian Arabic and one on French, were conducted to investigate this phenomenon. We were interested in stop consonants produced in the initial and the final positions in CVC sequences. Vowel duration and the frequency of the first three formants were measured for each vowel length and speaking rate. In addition, locus equations were computed to measure the impact of time variations on CV / VC coarticulation. According to the results, the vowel and consonant qualities are influenced both by vowel duration contrasts and by speaking rate modifications in different ways. Such modifications are evidenced by locus equations and by dispersion analyses as well. These two quantification tools revealed that there is greater coarticulatory overlap when the duration of the vowel is reduced, regardless of the origin of temporal variations in the two languages (intrinsic and extrinsic). On the other hand, temporal variations affect the classification of the stop consonant differently according to their origin.

Cette thèse concerne la comparaison des stratégies de coarticulation nasale sur les voyelles dans deux langues possédant des voyelles nasales phonologiques (français et ourdou) et trois langues ne possédant pas de voyelles nasales phonologiques (l’anglais américain, le bosnien et le norvégien). Pour évaluer le degré de nasalité produit, nous avons utilisé un accéléromètre piézoélectrique (AP) (non invasive et transportable). Cinq locuteurs natifs par langue ont enregistré un corpus de logatomes (CVNC et un ensemble de mots et de phrases. Cette thèse montre que (1) une différence inter-langues de stratégies existe, confirmant ainsi un fait déjà connu, (2) l’existence d’un contraste phonologique entre voyelles nasales et voyelles orales n’a pas d’influence prévisible sur la propagation de la nasalité, confirmant ainsi un fait déjà connu, (3) un modèle d’anticipation de nasalité peut être plus adéquat pour une langue que pour une autre, (4) les auditeurs francophones ne distinguent pas entre plusieurs degrés de nasalisation contextuelle, mais il existe une corrélation entre la quantité de vibrations issue de l’AP et la perception du degré de nasalisation pour les auditeurs de l’ourdou. L’intérêt de cette étude est que la comparaison de phénomènes de coarticulation nasale des voyelles dans les cinq langues a été faite avec la même instrumentation, dans les mêmes conditions. Un autre intérêt est de montrer l’intérêt de la méthode choisie. / This thesis concerns the comparison of the phenomena of nasal coarticulation of vowels in two languages that have phonological nasal vowels (French and Urdu) and three languages that don’t have phonological nasal vowels (American English, Bosnian and Norwegian). To evaluate the degree of nasality, we used a piezoelectric accelerometer (PA) (non invasive and transportable). Five native speakers of each language recorded the corpus of logatomes CVNC, words and sentences. This thesis shows that : (1) an inter-language difference exists with respect to nasal coarticulation strategies, confirming an already known fact, (2) the existence or absence of a phonological contrast between nasal vowels and oral vowels in a language does not necessarily have a predictable influence on the propagation of nasality, confirming an already known fact, (3) one model of anticipation of nasality may be more suitable for one language, while for another language another model may be more suitable. (4) the french auditors don’t disinguish between several degrees of contextual nasalization but there is a correlation between the quantity of vibration from the AP and the perception of the degree of nasality for the auditors of urdu. The interest of this study is that the comparison of the nasal coarticulation phenomena of vowels in the five languages was made with the same instrumentations under the same conditions. Another interest is to show the interest of the chosen method.

Le bégaiement est un trouble complexe, qu'il est encore difficile de définir de manière satisfaisante et complète. Outre les symptômes secondaires, et comportements accompagnateurs, des facteurs langagiers peuvent interférer avec ce trouble. Des difficultés dans les compétence phonologiques pourraient être concomitantes, voire en interaction avec le bégaiement, avec notamment dans la parole adulte, un impact non négligeable de la complexité phonologique. Par ailleurs, des difficultés coarticulatoires seraient au cœur du bégaiement. Mais les résultats des différentes études sont très disparates, tant dans la parole des enfants que dans celle des adultes qui bégaient. Enfin, l'influence du feedback auditif est surprenante puisque toute modification de ce type de feedback a un pouvoir améliorant chez un certain nombre de personnes. Ce constat questionne la définition du bégaiement en tant que trouble de la production et l'oriente plutôt vers un trouble perceptivo-moteur. Cette thèse se propose d'étudier la parole fluente et disfluente de personnes bègues françaises et italiennes. Etant donné que les différences entre personnes qui bégaient et personnes fluentes, apparaissent essentiellement quand le système moteur est soumis à un facteur déstabilisant, nous avons choisi d'étudier l'adaptation du comportement coarticulatoire lorsque la complexité phonologique augmente. Nous avons également voulu analyser l'impact de la modification du feedback auditif sur le comportement coarticulatoire. Enfin, nous avons étudié le rôle de la complexité phonologique sur les disfluences notamment, dans une situation de dialogue autour d'une image. Des adultes et des enfants, italiens, et français, bègues et fluents ont été enregistrés dans 4 situations de parole : lectures, discours spontané, tâche de répétition, et tâche de description d'image. Toutes ces tâches ont été réalisées dans deux conditions perceptives : une condition normale, et une condition avec feedback auditif modifié. En condition perceptive normale, la coarticulation des personnes qui bégaient est plus faible que celle des personnes fluentes. La langue semble jouer également un rôle important puisque les personnes qui bégaient des deux langues ne se comportent pas toujours de la même manière vis-à-vis de leurs homologues fluents. Le comportement coarticulatoire des personnes bègues semble également sensible à l'augmentation de la complexité phonologique. Les résultats sous feedback auditifs modifiés apparaissent contradictoires. Par ailleurs, les disfluences pourraient être influencées par la complexité phonologique, mais cette influence semble dépendre de la sévérité du bégaiement et de l'âge du sujet. Des perspectives cliniques sont évoquées. / Stuttering is a complex (complicated-intricate) disorder, and it is difficult to give a satisfactory and complete definition of it. As well as secondary symptoms and entailed comportments, language factors can interfere in this disorder. Difficulties in phonological competence might be concomitant if not interactive in stuttering, with, mainly in adult language, an inconsiderable impact of the phonological complexity Furthermore, the coarticulatory difficulties might be at the core of stuttering. But the results of different studies diverge on the subject when both children and adults' stuttering are considered. And let us note that the influence of the auditory-feedback is surprising as every change in this feedback seems to have an improving power in an appreciable number of patients. So the definition of stuttering, being a production disorder, can be questioned. Stuttering rather appears as a perceptual motor disorder. The aim of this thesis is to study the fluent and disfluent language in French and Italian stutterers knowing that the differences between stutterers and fluent persons mainly appear when the motor system is conditional to a destabilizing factor. We chose to study how the coarticulatory behavior adapts when the phonological complexity increases. We also tried to analyze how the modification of the auditory-feedback influences the coarticulatory behavior. And we studied the part played by the phonological complexity on disfluencies mainly in a situation of dialogue over a picture. Italian and French adults and children stutterers and their fluent counterparts were registered placed in four different situations of language: reading, spontaneous speech, repetitive task and picture description and in two perceptive conditions: a normal one and one with modified auditory-feedback. When placed in normal perceptive condition, the stutterers' coarticulation is weaker than that of their fluent counterparts. Besides, the language seems to play an important role since Italian and French stutterers have not the same response towards their fluent counterparts. Let us note that this stutterers' coarticulatory comportment is also sensitive to the increase of the phonological complexity. The results we have when the auditory-feedback is altered are contradictory. Disfluency could also be influenced by phonological complexity but this influence seems to be dependent on the acuteness of the stuttering as well as on the stutterer's age. A significant clinical breakthrough is evoked

Esta tesis examina la acústica de secuencias vocálicas que ocurren entre dos palabras y experimentan contracción silábica en el español hablado en Tucson, Arizona. La tradición lingüística hispánica define este proceso con el término "sinalefa". Numerosas fuentes notan que la sinalefa hablada en el suroeste de los Estados Unidos tiene características diferentes de las documentadas para otros dialectos en España. Ocho mujeres tucsonenses realizaron una tarea de repetición del habla en que pronunciaron frases detres palabras en las cuales, en habla rápida, se espararía la aplicación de un proceso de sinalefa. El resultado del análisis de las grabaciones muestra una forma de habla que no ha sido descrita previamente en la cual aparece un proceso que llamamos "coarticulación asimilatoria progresiva".

Phonemic perception exhibits coarticulation sensitivity, phonotactic sensitivity and lexical sensitivity. Three kinds of models of speech perception are found in the literature, which embody different answers to the question of how the three kinds of sensitivity are related to each other: two-step models, one-step models and lexicalist models. In two-step models (Church, 1987), phonemes are first extracted, and phonotactic repairs are subsequently made on the obtained phoneme string; both phonemic categorization and phonotactic repair are sublexical, and coarticulation sensitivity should only affect initial (prephonotactic) phonemic categorization. In one-step models (Dehaene-Lambertz et al., 2000; Dupoux et al., 2011; Mehler et al., 1990), phonemic categorization and phonotactic repair are sublexical and simultaneous; phonotactic repairs themselves depend on coarticulation cues. Such models can be implemented in two different versions: suprasegmental matching, according to which a speech signal is matched against phonotactics-respecting suprasegmental units (such as syllables), rather than phonemes, and slot filling, according to which a speech signal is matched against phonemes as fillers for slots in phonotactics-respecting suprasegmental units. In lexicalist models (Cutler et al., 2009; McClelland & Elman, 1986), coarticulation sensitivity and/or phonotactic sensitivity reduce to lexical sensitivity. McClelland & Elman (1986) claim a lexicalist reduction of phonotactic sensitivity; Cutler et al.’s (2009) make a claim implying lexicalist reductions both of phonotactic sensitivity and of coarticulation sensitivity. This thesis attempts to distinguish among those models. Since different perceptual processes are assumed in these three models (whether sublexical units are perceived, or how many stages are involved in perceptual processing), our understanding of how speech perception works crucially depends on the relative superiority of those three kinds of models. Based on the results available in the past literature on the one hand, and on the results of perceptual experiments with Japanese listeners testing their coarticulation sensitivity in different settings on the other, this thesis argues for the superiority of the slot filling version of one-step models over the others. According to this conclusion, phonemic parsing (categorization) and phonotactic parsing (repair) are separate but parallel sublexical processes.

This study investigates phonetic and phonological factors which influence the degree of vowel-to-vowel coarticulation in Korean and English, especially around palatalization rules. Two phonetic factors and two phonological factors were examined in investigating the degree of anticipatory or carryover coarticulation in VCV sequences. The phonetic factors were the intervening consonants (alveolar stop vs. (alveo)palatals), and the second vowels (/i/ vs. /a/); the phonological factors were the effect of palatalization, and the lexical status of palatalization (lexical vs. postlexical palatalization). Ultrasound imaging techniques and F2 measurements are employed to see how much further front the articulation of V1 in V1CV2 sequences is due to influence of V2 across the consonants. Ultrasound images of vowels and their F2 values were quantified and statistically analyzed with ANOVA.First, it was found that V1 in V1CV2 sequences in Korean was articulated further front when intervening consonants were palatals than when they were alveolars, while there was no difference in frontness of V1 between two consonantal conditions in English. This indicates that Korean palatals are a stronger barrier to vowel-to-vowel coarticulation, while English alveopalatals are not. Next, V1 in both languages was articulated further front when V2 was /i/ than when V2 was /a/. Third, we had striking findings that palatalization rules caused stronger vowel-to-vowel coarticulation than in nonpalatalized words. Results showed that V1s were articulated further front across derived palatals than across underlying palatals in V1Ci sequences. Last, it was determined that V2 was articulated further front in words which underwent postlexical palatalization than in words which underwent lexical palatalization. Such findings indicate that postlexical palatalization causes greater gestural overlap than lexical palatalization, showing stronger degree of coarticulation.Based on the experimental results that phonetic details such as the degree of vowel-to-vowel coarticulation are highly conditioned by the lexical status of palatalization as well as the application of palatalization, I suggest a unified model of phonology and phonetics, using feature-and-gesture based OT frameworks. Second, I follow the proposal that abstract intergestural timing relations should be incorporated into phonological representations either in the input or output (Cho 1998, Gafos 2002, Yun 2005b).

Ce travail de recherche porte sur un aspect phonétique qui s’inscrit dans trois domaines, la phonétique, la dialectologie et la sociophonétique arabes. Notre démarche, nos outils et nos analyses sont phonétiques par essence. Nous avons analysé la fréquence des trois premiers formants [F1, F2, F3] des voyelles cardinales brèves /i, u, a/ de l’arabe libyen de Tripoli (ALT) et nous avons alterné le contexte phonétique consonantique pharyngalisé / tˁ, sˁ, dˁ/ vs non pharyngalisé / t, s, d/ afin de vérifier l’impact de celui-ci sur la fréquence centrale des formants. Cependant, les résultats ainsi obtenus nous ont permis de comparer l’ALT à d’autres variétés populaires arabes modernes. Comme ils nous ont permis d’opérer des distinctions sociales fondamentales, comme celle du gender. Notre problématique articule la question de la réalisation de l’espace acoustique des voyelles en ALT avec le contraste consonantique de pharyngalisation, les patrons coarticulatoires qui en résultent ainsi que l’outil ‘équation de locus’ pour les révéler, tout cela dans une dimension de stratification sociale par le gender. Trois hypothèses de travail ont été présentées, la première sur la variation de l’espace vocalique et ses motivations, la deuxième sur la pertinence de l’utilisation de l’équation de locus et la troisième sur les distinctions liées au gender. Nos résultats, fondés sur l’analyse d’un corpus de mots trisyllabiques [C1V1- C2V2- C3V3] où C était soit une consonne pharyngalisée /s ˁ, t ˁ, d ˁ/, soit une consonne non pharyngalisée /s, t, d/, V étant une des trois voyelles brèves cardinales /i, u, a/, lu par dix locuteurs (6 hommes et 5 femmes) permettent de valider nos trois hypothèses : la variation des valeurs formantiques des voyelles, de l’espace acoustique et de la distance entre les deux premiers formants en fonction des trois facteurs : 1) le contexte consonantique (pharyngalisé vs non pharyngalisé) ; 2) la position prosodique (accentué vs inaccentué) ; et 3) la distinction sociale (homme vs femme). Notre travail a pu répondre positivement aux objectifs qui lui ont été assignés au départ : 1) sur le plan phonétique, donner un aperçu du système vocalique de l’ALT et de sa variation en fonction de la pharyngalisation ; 2) sur le plan dialectologique, répondre aux questions de la typologie dialectale arabe et le classement de l’arabe libyen, dialecte oriental vs dialecte maghrébin ; et 3) sur le plan sociophonétique, vérifier la profonde distinction sociale, parole de femme vs parole d’homme / This research focuses on a phonetic aspect that falls into three areas, phonetics, dialectology and Arabic sociophonetics. Our approach, tools and analysis are phonetic in nature. We analyzed the frequency of the first three formants [F1, F2, F3] of short cardinal vowels / i, u, a /of the Libyan Arabic of Tripoli (ALT) and we alternated the phonetic context of pharyngealized consonants / t ˁ, s ˁ , ˁ d / versus non pharyngealized / t, s, d / to verify the impact of the latter on the central frequency of formants. However, the results obtained have allowed us to compare the ALT to other popular varieties in modern Arabic. As they allowed us to make fundamental social distinctions, such as the gender. Our research problem deals with the issue of the realization of the acoustic space of vowels in ALT, in contrast with the consonant pharyngealization, the resulting co-articulatory patterns as well as the tool of 'locus equation' to reveal, all of this in a dimension of social stratification by gender. Three hypotheses were presented; the first was on the variation of the vowel space and its motivations, the second was on the appropriateness of using the locus equation and the third was related to gender distinctions. Our results, based on the analysis of a corpus of trisyllabic words [C1V1 C2V2-C3V3-] where C is either a pharyngalized consonant / s ˁ, ˁ t, d ˁ / or a non- pharyngalized consonant / s, t, d /, V is one of the three cardinal short vowels / i, u, a /, read by ten speakers (6 males and 5 females) to validate our three hypotheses. Changes in formant values of vowels, the acoustic space and the distance between the first two formants were based on three factors: 1) the consonantal context (pharyngalized vs. non- pharyngalized), 2) the prosodic position (stressed vs. unstressed), and 3) social distinction (male vs. female). Our study has been able to respond positively to the objectives assigned to it at the start: 1) on the phonetic level, it gives an overview of the vowel system of ALT and its variation with the pharyngealization 2) on the dialectological level, it answers questions from Arabic dialect typology and classification of the Eastern Libyan Arabic dialect vs Maghreb dialect and 3) on the sociophonetic level, it verifies the profound social distinction, woman's speech vs man's speech

This study examines anticipatory pharyngealisation (i.e., emphasis) in Libyan Arabic, across a hierarchy of prosodic boundary levels (syllable vs. word vs. phonological phrase vs. intonation phrase ‘IP’) in order to quantify the magnitude, and identify the planned domain of anticipatory pharyngealisation. The acoustic manifestation of pharyngealisation is lowering in the second formant (F2) in pharyngealised contexts compared to their plain cognates. To investigate speech production models of how pharyngealisation is anticipated in advance, F2 measurements were taken at onset, mid and offset points of both vowels (V) in a word-final VCV sequence, in the context [VbV # Emphatic trigger]. The strength of [#], a prosodic boundary, was varied syntactically to manipulate the presumed hierarchical strength of that boundary from zero (where the VbV and the trigger are in the same word) up to an intonational phrase boundary. We expect that the stronger the boundary, the greater the resistance to the spread of pharyngealisation. The duration of the final vowel (i.e., the pre-trigger vowel) was also measured to assess if pharyngealisation magnitude on it and on the first vowel is influenced by the temporal proximity to the emphatic trigger. Results show (1) that within word boundaries pharyngealisation effects are present on both vowels, and (2) there are effects of pharyngealisation on the final vowel, i.e. the pre-trigger across word and phrase boundaries, and (3) there is no evidence of pharyngealisation across an IP boundary. An examination of the pre-trigger vowel + pause duration suggests that the lack of coarticulatory effects on the final vowel, i.e., pre-trigger vowel, across an IP boundary may be due to the temporal distance from the trigger: all tokens in this condition had a pre-trigger pause. For word and phrase boundary conditions, F2 was higher the greater the temporal distance from the pharyngealised trigger. These results suggest that anticipatory pharyngealisation is qualitatively different within the word as compared to across word boundaries. More clearly, the magnitude of pharyngealisation is categorical within word boundaries, and gradient across prosodic boundaries higher than the word. These findings suggest that pharyngealisation within the word is phonological, whereas across word boundaries it is primarily a phonetic process, conditioned by the temporal proximity to the pharyngealised trigger. Results also show that the planned domain of [pharyngealisation] is the word. However, additional phonetic pharyngealisation effects can extend across word boundaries as a result of coarticulation.