<p>A few aspects of aircraft noise were evaluated. These were (i) methods of subjective evaluations, (ii) effects of equalization and (iii) the effects of cognitive aspects.</p><p>In the first paper, sorting algorithms were used instead of conventional paired comparison method in order to reduce the number of pairs in the evaluation of subjective judgments. The quick sorting algorithm method revealed more than 99% correlation coefficient with paired comparison method although the method used N*log(N) evaluations instead of N(N-1)/2.</p><p>In the second paper, equalization effects on perception were evaluated in two steps, first with stationary aircraft sounds and second with non-stationary aircraft sounds.</p><p>The first experiment examined the effects of stationary sound segments respect to three different angle positions of the aircrafts relative to the observer (78.7°, 90° & 101.3°), two different SNR conditions (sounds having original broadband plus tonal components versus control broadband sounds having no tonal components) and two different flight conditions (arrival and take-off). Subjects were asked to scale five perceptual attributes (loudness, annoyance, hardness, power and pitch) using Borg CR100 scale. The angle condition showed highly significant effects on annoyance and hardness. Maximal effects were found at an angle of 78.7°. The SNR revealed a significant impact on loudness, power and pitch.</p><p>The second experiment analyzed the effects of tonal components and the problem of appropriate equalization. The spectrum of the signals was modified in two steps (buzz-saw, isolated BPF tone). Further EPNL-equalization, A-, B-, C-, D- and spectral broadband equalizations were applied to the synthesized sounds. Annoyance, loudness, hardness and pitch in the isolated tone conditions showed significantly stronger effects than the buzz-conditions on the perceived judgments. The EPNL-equalization led to a lower degree of differentiation between the spectral conditions compared to B- and C-level equalization.</p><p>In the third paper, the effects of aircraft sounds on children’s cognitive performance were investigated. Impact of aircraft noise on children cognition was found significantly higher in reading comprehensions than in basic mathematics and problem solving tests. It seems children are very sensitive to the modifications in the aircraft noise but further studies are necessary to compliment such a finding.</p>

This thesis introduces a novel situated interactive composition system called the SMuSe (for Situated Music Server) that is grounded on principles of modern cognitive science, provides perceptual control of sound synthesis and includes emotional feedback. It illustrates both a new music composition paradigm and a synthetic psychology approach to the study of music perception, emotion and cognition. The SMuSe is composed of cognitively plausible modules implemented as a hierarchy of musical agents and relies on distributed control, parallelism, emergence and embodiment. By interacting with its environment, which provides feedback via multiple sensors, the system generates complex adaptive affective musical structures. Focusing on the micro-level of sound generation, we present two complementary techniques that give high-level perceptual control over low-level sound synthesis parameters. In a first implicit approach, a support vector machine algorithm learns to automatically map perceptual features such as loudness, pitch and brightness onto additive synthesis parameters. In a second approach, a physically-inspired synthesis model provides explicit access to perceptual and physical parameters such as pitch, loudness, brightness, attack time, inharmonicity and damping. Moving from the study of music generation and control towards the study of the musical experience itself, we then evaluate how the music generated by the SMuSe influences the listeners' emotional responses. A first psychoacoustics experiment shows the significant influence of structural (scale, register, harmony), expressive (velocity, tempo, articulation) and timbre (brightness, attack time, spectral flux) parameters on the emotional scales of valence, arousal and dominance. An additional large scale experiment involving dementia patients (an illness known to induce cognitive and affective deficits) shows that specific sound features (e.g. low loudness, low brightness) provoke specific emotional responses within the patients (e.g. low stress). Moreover, the patients' emotional responses differ from the age-matched control group, and the analysis shows an increased emotional sensitivity to sounds as the severity of the disease increases. These results suggest that sound-based therapy and diagnosis for dementia are possible. Finally, the maturity and flexibility of the SMuSe music system are demonstrated by a series of real-world applications including the sonification of a mixed-reality space, a study on physiologically-based musical interaction, a neurofeedback musical interface, a closed loop system based on reinforcement learning of emotional feedback, and a large scale multimedia performance using brain-computer interfaces. A situated, perceptive, emotive and cognitive approach to the design of musical systems paves the way for new applications for therapy but also for interactive gaming and novel physiologically-based instruments. Our approach provides a well-grounded paradigm to develop advanced synthetic aesthetics system that can inform our understanding of the psychological processes on which they rely. / Esta tesis introduce un nuevo sistema de composición situada e interactiva llamado SMuSe (por Situated Music Server). Dicho sistema está basado en principios extraídos de la ciencia cognitiva moderna, proporciona control perceptual sobre la síntesis sonora e incluye feedback emocional. Por lo tanto, ilustra tanto un paradigma nuevo para la composición musical, como un sintético enfoque psicológico al estudio de la percepción musical, las emociones y la cognición. SMuSe consta de diversos modulos, plausibles desde un punto de vista cognitivo, implementados como una jerarquía de agentes. El funcionamiento de SMuSe explota los principios de control distribuido, paralelismo, emergencia y embodiment. En función del feedback obtenido por la interacción con el entorno, el sistema genera complejas estructuras musicales afectivas. En concreto, a nivel de generación de sonido, presentamos dos técnicas complementarias que proporcionan un control perceptivo de alto nivel sobre parámetros concretos de síntesis sonora. En un primer método implícito, un algoritmo de support vector machine aprende a traducir automáticamente características perceptuales, como volumen, tono y brillo en parámetros de síntesis aditiva. En el segundo método, un modelo físico de síntesis proporciona explícitamente acceso a parámetros perceptivos y físicos, tales como tono, volumen, brillo, tiempo de ataque, inarmonía y factor de amortiguamiento. En lo que respecta al estudio de la experiencia musical en sí misma, evaluamos la influencia de la música generada por SMuSe en las respuestas emocionales del sujeto. Un primer experimento psicoacústico muestra la influencia significativa que tienen parámetros estructurales (escala,registro, armonía), expresivos (velocidad, tempo, articulación) y de timbre (brillo, ataque, flujo espectral) en la escalas emocionales de valencia, activacíon y dominancia. Adicionalmente, un experimento de gran escala realizado con pacientes de demencia, una patología que asociada a déficits afectivos y cognitivos, demuestra que los pacientes responden emocionalmente a rasgos específicos del sonido (e.g. bajo volumen y brillo inducen poca tensión). Además, la respuesta emocional de los pacientes difiere si se compara con la mostrada por un grupo de control con la misma edad media. De tal manera, una mayor respuesta emocional aparece asociada a un mayor grado de demencia. Estos resultados sugieren que sería posible desarrollar técnicas basadas en el uso de música tanto para el tratamiento de la demencia como para su diagnóstico. Para concluir, la madurez y flexibilidad de SMuSe se demuestra con una serie de aplicaciones que incluyen la sonificación de un espacio de realidad mixta, un estudio acerca de la interacción musical mediante datos fisiológicos, un interfaz musical basado en feedback neurológico, un sistema basado en aprendizaje por refuerzo del feedback emocional, y una performance multimedia de gran escala controlada mediante interfaces cerebro-máquina. Este enfoque situado, perceptivo, emocional y cognitivo al diseño de sistemas musicales abre la posibilidad de desarrollar aplicaciones no sólo terapéuticas sino también para los juegos interactivos y nuevos interfaces que empleen fisiología. Nuestra propuesta proporciona un sólido paradigma para el desarrollo de sistemas de síntesis de estéticas avanzadas, que puedan servir para entender los procesos psicológicos subyacentes

The present research sought to investigate the correspondence between auditory and visual feature dimensions and to utilise this knowledge in order to inform the design of audio-visual mappings for visual control of sound synthesis. The first stage of the research involved the design and implementation of Morpheme, a novel interface for interaction with corpus-based concatenative synthesis. Morpheme uses sketching as a model for interaction between the user and the computer. The purpose of the system is to facilitate the expression of sound design ideas by describing the qualities of the sound to be synthesised in visual terms, using a set of perceptually meaningful audio-visual feature associations. The second stage of the research involved the preparation of two multidimensional mappings for the association between auditory and visual dimensions. The third stage of this research involved the evaluation of the Audio-Visual (A/V) mappings and of Morpheme's user interface. The evaluation comprised two controlled experiments, an online study and a user study. Our findings suggest that the strength of the perceived correspondence between the A/V associations prevails over the timbre characteristics of the sounds used to render the complementary polar features. Hence, the empirical evidence gathered by previous research is generalizable/ applicable to different contexts and the overall dimensionality of the sound used to render should not have a very significant effect on the comprehensibility and usability of an A/V mapping. However, the findings of the present research also show that there is a non-linear interaction between the harmonicity of the corpus and the perceived correspondence of the audio-visual associations. For example, strongly correlated cross-modal cues such as size-loudness or vertical position-pitch are affected less by the harmonicity of the audio corpus in comparison to weaker correlated dimensions (e.g. texture granularity-sound dissonance). No significant differences were revealed as a result of musical/audio training. The third study consisted of an evaluation of Morpheme's user interface were participants were asked to use the system to design a sound for a given video footage. The usability of the system was found to be satisfactory. An interface for drawing visual queries was developed for high level control of the retrieval and signal processing algorithms of concatenative sound synthesis. This thesis elaborates on previous research findings and proposes two methods for empirically driven validation of audio-visual mappings for sound synthesis. These methods could be applied to a wide range of contexts in order to inform the design of cognitively useful multi-modal interfaces and representation and rendering of multimodal data. Moreover this research contributes to the broader understanding of multimodal perception by gathering empirical evidence about the correspondence between auditory and visual feature dimensions and by investigating which factors affect the perceived congruency between aural and visual structures.

Ce travail se situe dans le cadre de la recherche de systèmes de notation permettant de transcrire de façon symbolique l’aspect concret et sensoriel, et non seulement abstrait et structurel, des artefacts de la musique par ordinateur. Dans ce but, nous exposons tout d’abord un modèle formel complet et minimal des objets et structures audionumériques, en référence aux critères de la perception ; ce modèle est implémenté sous la forme d’un langage fonctionnel Turing-potent qui permet d’effectuer l’équivalence entre l’expression mathématique d’un signal et sa réalisation informatique. Puis, nous employons ce formalisme afin d’exprimer un ensemble de critères de synthèse sonore, ce qui donne lieu à un logiciel de synthèse dont l’expressivité est considérable. Ces outils sont organisés suivant le schéma des théories Schaefferiennes, par une décomposition catégorielle dans laquelle les paramètres correspondent à des notions morphologiques. Finalement, nous rendons compte d’une série d’expériences visant à évaluer la pertinence de ces critères dans l’audition humaine, avec le concours d’un musicologue, puis sur un ensemble de sujets, et enfin vis-à-vis d’un public aussi large que possible. Ceci nous conduit à remettre en question la méthodologie la plus adéquate pour traiter ce type de problème, qui nous rapproche des sciences humaines et sociales, et suggère une démarche de science participative. / This piece of work is situated in the context of research on notation systems enabled to transcribe symbolically the concrete and sensorial aspect, and not only the abstract and structural aspect, of computer music artefacts. In this perspective, we first expose a complete and minimal formal model for digital audio objects and structures, relatively to the criteria of perception ; this model is implemented as a Turing-potent functional language, that draws the correspondance between the mathematical expression of a signal and its computational realisation. Then, we apply this formal construction to the expression of a number of schemes for sound synthesis, producing a software synthetiser whose expressivity is consequent. These tools are organised following the lines of Schaefferian theories, through a decomposition into categories whose parameters correspond with morphological notions. Finally, we draw the conclusions of a series of experiments aiming to evaluate the relevance of those schemes in human hearing, with the assistance of a musicologist, then with a number of subjects, and eventually by associating a public that is as wide as possible. This leads us to question the methodology most appropriate to tackle this kind of problem, which brings us closer to social science, and suggests a participative science approach.

Muitos pesquisadores têm se interessado em explorar o vasto poder computacional das recentes unidades de processamento gráfico (GPUs) em aplicações fora do domínio gráfico. Essa tendência ao desenvolvimento de propósitos gerais com a GPU (GPGPU) foi intensificada com a produção de APIs não-gráficas, tais como a Compute Unified Device Architecture (CUDA), da NVIDIA. Com elas, estudou-se a solução na GPU de muitos problemas de processamento de sinal 2D e 3D envolvendo álgebra linear e equações diferenciais parciais, mas pouca atenção tem sido dada ao processamento de sinais 1D, que também podem exigir recursos computacionais significativos. Já havia sido demonstrado que a GPU pode ser usada para processamento de sinais em tempo-real, mas alguns processos não se adequavam bem à arquitetura da GPU. Neste trabalho, apresento uma nova técnica para implementar um filtro digital linear recursivo usando a GPU. Até onde eu sei, a solução aqui apresentada é a primeira na literatura. Uma comparação entre esta abordagem e uma implementação equivalente baseada na CPU demonstra que, quando usada em um sistema de processamento de áudio em temporeal, esta técnica permite o processamento de duas a quatro vezes mais coeficientes do que era possível anteriormente. A técnica também elimina a necessidade de processar o filtro na CPU - evitando transferências de memória adicionais entre CPU e GPU - quando se deseja usar o filtro junto a outros processos, tais como síntese de som. A recursividade estabelecida pela equação do filtro torna difícil obter uma implementação eficiente em uma arquitetura paralela como a da GPU. Já que cada amostra de saída é computada em paralelo, os valores necessários de amostras de saída anteriores não estão disponíveis no momento do cômputo. Poder-se-ia forçar a GPU a executar o filtro sequencialmente usando sincronização, mas isso seria um uso ineficiente da GPU. Este problema foi resolvido desdobrando-se a equação e "trocando-se" as dependências de amostras próximas à saída atual por outras precedentes, assim exigindo apenas o armazenamento de um certo número de amostras de saída. A equação resultante contém convoluções que então são eficientemente computadas usando a FFT. A implementação da técnica é geral e funciona para qualquer filtro recursivo linear invariante no tempo. Para demonstrar sua relevância, construímos um filtro LPC para sintetizar em tempo-real sons realísticos de colisões de objetos feitos de diferentes materiais, tais como vidro, plástico e madeira. Os sons podem ser parametrizados por material dos objetos, velocidade e ângulo das colisões. Apesar de flexível, esta abordagem usa pouca memória, exigindo apenas alguns coeficientes para representar a resposta ao impulso do filtro para cada material. Isso torna esta abordagem uma alternativa atraente frente às técnicas tradicionais baseadas em CPU que apenas realizam a reprodução de sons gravados. / Many researchers have been interested in exploring the vast computational power of recent graphics processing units (GPUs) in applications outside the graphics domain. This trend towards General-Purpose GPU (GPGPU) development has been intensified with the release of non-graphics APIs for GPU programming, such as NVIDIA's Compute Unified Device Architecture (CUDA). With them, the GPU has been widely studied for solving many 2D and 3D signal processing problems involving linear algebra and partial differential equations, but little attention has been given to 1D signal processing, which may demand significant computational resources likewise. It has been previously demonstrated that the GPU can be used for real-time signal processing, but several processes did not fit the GPU architecture well. In this work, a new technique for implementing a digital recursive linear filter using the GPU is presented. To the best of my knowledge, the solution presented here is the first in the literature. A comparison between this approach and an equivalent CPU-based implementation demonstrates that, when used in a real-time audio processing system, this technique supports processing of two to four times more coefficients than it was possible previously. The technique also eliminates the necessity of processing the filter on the CPU - avoiding additional memory transfers between CPU and GPU - when one wishes to use the filter in conjunction with other processes, such as sound synthesis. The recursivity established by the filter equation makes it difficult to obtain an efficient implementation on a parallel architecture like the GPU. Since every output sample is computed in parallel, the necessary values of previous output samples are unavailable at the time the computation takes place. One could force the GPU to execute the filter sequentially using synchronization, but this would be a very inefficient use of GPU resources. This problem is solved by unrolling the equation and "trading" dependences on samples close to the current output by other preceding ones, thus requiring only the storage of a limited number of previous output samples. The resulting equation contains convolutions which are then efficiently computed using the FFT. The proposed technique's implementation is general and works for any time-invariant recursive linear filter. To demonstrate its relevance, an LPC filter is designed to synthesize in real-time realistic sounds of collisions between objects made of different materials, such as glass, plastic, and wood. The synthesized sounds can be parameterized by the objects' materials, velocities and collision angles. Despite its flexibility, this approach uses very little memory, requiring only a few coefficients to represent the impulse response for the filter of each material. This turns this approach into an attractive alternative to traditional CPU-based techniques that use playback of pre-recorded sounds.

Muitos pesquisadores têm se interessado em explorar o vasto poder computacional das recentes unidades de processamento gráfico (GPUs) em aplicações fora do domínio gráfico. Essa tendência ao desenvolvimento de propósitos gerais com a GPU (GPGPU) foi intensificada com a produção de APIs não-gráficas, tais como a Compute Unified Device Architecture (CUDA), da NVIDIA. Com elas, estudou-se a solução na GPU de muitos problemas de processamento de sinal 2D e 3D envolvendo álgebra linear e equações diferenciais parciais, mas pouca atenção tem sido dada ao processamento de sinais 1D, que também podem exigir recursos computacionais significativos. Já havia sido demonstrado que a GPU pode ser usada para processamento de sinais em tempo-real, mas alguns processos não se adequavam bem à arquitetura da GPU. Neste trabalho, apresento uma nova técnica para implementar um filtro digital linear recursivo usando a GPU. Até onde eu sei, a solução aqui apresentada é a primeira na literatura. Uma comparação entre esta abordagem e uma implementação equivalente baseada na CPU demonstra que, quando usada em um sistema de processamento de áudio em temporeal, esta técnica permite o processamento de duas a quatro vezes mais coeficientes do que era possível anteriormente. A técnica também elimina a necessidade de processar o filtro na CPU - evitando transferências de memória adicionais entre CPU e GPU - quando se deseja usar o filtro junto a outros processos, tais como síntese de som. A recursividade estabelecida pela equação do filtro torna difícil obter uma implementação eficiente em uma arquitetura paralela como a da GPU. Já que cada amostra de saída é computada em paralelo, os valores necessários de amostras de saída anteriores não estão disponíveis no momento do cômputo. Poder-se-ia forçar a GPU a executar o filtro sequencialmente usando sincronização, mas isso seria um uso ineficiente da GPU. Este problema foi resolvido desdobrando-se a equação e "trocando-se" as dependências de amostras próximas à saída atual por outras precedentes, assim exigindo apenas o armazenamento de um certo número de amostras de saída. A equação resultante contém convoluções que então são eficientemente computadas usando a FFT. A implementação da técnica é geral e funciona para qualquer filtro recursivo linear invariante no tempo. Para demonstrar sua relevância, construímos um filtro LPC para sintetizar em tempo-real sons realísticos de colisões de objetos feitos de diferentes materiais, tais como vidro, plástico e madeira. Os sons podem ser parametrizados por material dos objetos, velocidade e ângulo das colisões. Apesar de flexível, esta abordagem usa pouca memória, exigindo apenas alguns coeficientes para representar a resposta ao impulso do filtro para cada material. Isso torna esta abordagem uma alternativa atraente frente às técnicas tradicionais baseadas em CPU que apenas realizam a reprodução de sons gravados. / Many researchers have been interested in exploring the vast computational power of recent graphics processing units (GPUs) in applications outside the graphics domain. This trend towards General-Purpose GPU (GPGPU) development has been intensified with the release of non-graphics APIs for GPU programming, such as NVIDIA's Compute Unified Device Architecture (CUDA). With them, the GPU has been widely studied for solving many 2D and 3D signal processing problems involving linear algebra and partial differential equations, but little attention has been given to 1D signal processing, which may demand significant computational resources likewise. It has been previously demonstrated that the GPU can be used for real-time signal processing, but several processes did not fit the GPU architecture well. In this work, a new technique for implementing a digital recursive linear filter using the GPU is presented. To the best of my knowledge, the solution presented here is the first in the literature. A comparison between this approach and an equivalent CPU-based implementation demonstrates that, when used in a real-time audio processing system, this technique supports processing of two to four times more coefficients than it was possible previously. The technique also eliminates the necessity of processing the filter on the CPU - avoiding additional memory transfers between CPU and GPU - when one wishes to use the filter in conjunction with other processes, such as sound synthesis. The recursivity established by the filter equation makes it difficult to obtain an efficient implementation on a parallel architecture like the GPU. Since every output sample is computed in parallel, the necessary values of previous output samples are unavailable at the time the computation takes place. One could force the GPU to execute the filter sequentially using synchronization, but this would be a very inefficient use of GPU resources. This problem is solved by unrolling the equation and "trading" dependences on samples close to the current output by other preceding ones, thus requiring only the storage of a limited number of previous output samples. The resulting equation contains convolutions which are then efficiently computed using the FFT. The proposed technique's implementation is general and works for any time-invariant recursive linear filter. To demonstrate its relevance, an LPC filter is designed to synthesize in real-time realistic sounds of collisions between objects made of different materials, such as glass, plastic, and wood. The synthesized sounds can be parameterized by the objects' materials, velocities and collision angles. Despite its flexibility, this approach uses very little memory, requiring only a few coefficients to represent the impulse response for the filter of each material. This turns this approach into an attractive alternative to traditional CPU-based techniques that use playback of pre-recorded sounds.

Cette thèse questionne l'analyse des musiques d'informatique. Ces musiques sont particulières dans la mesure où elles existent avec un medium tout à fait singulier, un medium numérique, que nous désignons tout au long de nos travaux par artefact. Notre réflexion part du constat que les moyens d’appréhension de ces artefacts sont nouveaux pour la théorie de l’analyse musicale. Nous posons alors en liminaire de nos travaux la question suivante : «  comment analyser ces musiques à partir de leur artefact ? ». Notre thèse est une réponse à cette question en utilisant une pièce pionnière des années 1980, Jupiter, de Philippe Manoury, pour flûte et électronique, composée à l’Ircam avec Miller Puckette. Notre réponse se fait en trois temps. Dans un premier nous éclairons les artefacts en relevant cinq points de singularité qui leur sont propres et nous mettons en évidence un système clé dans l’étude des artefacts, le système Homme-Artefact-Son. Dans un deuxième temps nous apportons des réponses théoriques à la question. Nous montrons que les artefacts sont difficiles à saisir dans la mesure où ils se structurent en différents niveaux d’abstraction. Toute saisie doit alors se faire au sein d’un processus hautement dynamique, où des significations computationnelles, sonores et musicales se recouvrent autant que nécessaire. Un cadre conceptuel est ensuite proposé pour guider la description des artefacts. Enfin le dernier temps de notre thèse est dédié à l’analyse de Jupiter. Nous mettons en pratique les éléments de réponses que nous avons avancés. L’artefact de la pièce est entièrement démantelé et ensuite une analyse est effectuée, section par section, en confrontant l’écoute et la connaissance que nous avons acquise de l’artefact. / This thesis question the musical analysis of computer music works. These works are particular insofar as they exist with a wholly singular medium, a digital medium, which we name artifact. Our reflection starts from the ascertainment that the means needed to understand the artifact are new for the theory of musical analysis. Thereby our work begins with the following question : « how does computer music works can be analyzed with the help of their artifact ? ». Our thesis intends to answer to this question through a pioneer work from the 1980s, Jupiter, composed by Philippe Manoury at Ircam with Miller Puckette. Our answer is built in three parts. At first, we enlighten artifact by defining five points of singularity which characterize them. In the same line we identify a key system in the study of artifact, the Man Artifact Sound system. The second part is dedicated to our answer. One of the main difficulties in studying artifact is that they are structured in different levels of abstraction, and each one has its own structure. Consequently, the understanding of artifact had to be done inside a highly dynamic process, where the computational, sound and musical significations overlap as much as necessary. A conceptual framework is then suggested by which artifacts can be described, taking into account the different levels of abstraction. In the last part we put into practice those last propositions in analyzing Jupiter. His artifact is entirely dismantled and then an analysis is carried out, section by section, in confronting the listening and the knowledge we get from the artifact.

Muitos pesquisadores têm se interessado em explorar o vasto poder computacional das recentes unidades de processamento gráfico (GPUs) em aplicações fora do domínio gráfico. Essa tendência ao desenvolvimento de propósitos gerais com a GPU (GPGPU) foi intensificada com a produção de APIs não-gráficas, tais como a Compute Unified Device Architecture (CUDA), da NVIDIA. Com elas, estudou-se a solução na GPU de muitos problemas de processamento de sinal 2D e 3D envolvendo álgebra linear e equações diferenciais parciais, mas pouca atenção tem sido dada ao processamento de sinais 1D, que também podem exigir recursos computacionais significativos. Já havia sido demonstrado que a GPU pode ser usada para processamento de sinais em tempo-real, mas alguns processos não se adequavam bem à arquitetura da GPU. Neste trabalho, apresento uma nova técnica para implementar um filtro digital linear recursivo usando a GPU. Até onde eu sei, a solução aqui apresentada é a primeira na literatura. Uma comparação entre esta abordagem e uma implementação equivalente baseada na CPU demonstra que, quando usada em um sistema de processamento de áudio em temporeal, esta técnica permite o processamento de duas a quatro vezes mais coeficientes do que era possível anteriormente. A técnica também elimina a necessidade de processar o filtro na CPU - evitando transferências de memória adicionais entre CPU e GPU - quando se deseja usar o filtro junto a outros processos, tais como síntese de som. A recursividade estabelecida pela equação do filtro torna difícil obter uma implementação eficiente em uma arquitetura paralela como a da GPU. Já que cada amostra de saída é computada em paralelo, os valores necessários de amostras de saída anteriores não estão disponíveis no momento do cômputo. Poder-se-ia forçar a GPU a executar o filtro sequencialmente usando sincronização, mas isso seria um uso ineficiente da GPU. Este problema foi resolvido desdobrando-se a equação e "trocando-se" as dependências de amostras próximas à saída atual por outras precedentes, assim exigindo apenas o armazenamento de um certo número de amostras de saída. A equação resultante contém convoluções que então são eficientemente computadas usando a FFT. A implementação da técnica é geral e funciona para qualquer filtro recursivo linear invariante no tempo. Para demonstrar sua relevância, construímos um filtro LPC para sintetizar em tempo-real sons realísticos de colisões de objetos feitos de diferentes materiais, tais como vidro, plástico e madeira. Os sons podem ser parametrizados por material dos objetos, velocidade e ângulo das colisões. Apesar de flexível, esta abordagem usa pouca memória, exigindo apenas alguns coeficientes para representar a resposta ao impulso do filtro para cada material. Isso torna esta abordagem uma alternativa atraente frente às técnicas tradicionais baseadas em CPU que apenas realizam a reprodução de sons gravados. / Many researchers have been interested in exploring the vast computational power of recent graphics processing units (GPUs) in applications outside the graphics domain. This trend towards General-Purpose GPU (GPGPU) development has been intensified with the release of non-graphics APIs for GPU programming, such as NVIDIA's Compute Unified Device Architecture (CUDA). With them, the GPU has been widely studied for solving many 2D and 3D signal processing problems involving linear algebra and partial differential equations, but little attention has been given to 1D signal processing, which may demand significant computational resources likewise. It has been previously demonstrated that the GPU can be used for real-time signal processing, but several processes did not fit the GPU architecture well. In this work, a new technique for implementing a digital recursive linear filter using the GPU is presented. To the best of my knowledge, the solution presented here is the first in the literature. A comparison between this approach and an equivalent CPU-based implementation demonstrates that, when used in a real-time audio processing system, this technique supports processing of two to four times more coefficients than it was possible previously. The technique also eliminates the necessity of processing the filter on the CPU - avoiding additional memory transfers between CPU and GPU - when one wishes to use the filter in conjunction with other processes, such as sound synthesis. The recursivity established by the filter equation makes it difficult to obtain an efficient implementation on a parallel architecture like the GPU. Since every output sample is computed in parallel, the necessary values of previous output samples are unavailable at the time the computation takes place. One could force the GPU to execute the filter sequentially using synchronization, but this would be a very inefficient use of GPU resources. This problem is solved by unrolling the equation and "trading" dependences on samples close to the current output by other preceding ones, thus requiring only the storage of a limited number of previous output samples. The resulting equation contains convolutions which are then efficiently computed using the FFT. The proposed technique's implementation is general and works for any time-invariant recursive linear filter. To demonstrate its relevance, an LPC filter is designed to synthesize in real-time realistic sounds of collisions between objects made of different materials, such as glass, plastic, and wood. The synthesized sounds can be parameterized by the objects' materials, velocities and collision angles. Despite its flexibility, this approach uses very little memory, requiring only a few coefficients to represent the impulse response for the filter of each material. This turns this approach into an attractive alternative to traditional CPU-based techniques that use playback of pre-recorded sounds.

La synthèse pour par modèles physiques propose une approche à de la création musicale alternative à de celui celle plus habituelle du traitement du signal. En prenant en considération le phénomène musical comme un unicum émergeant de l'interaction entre le musicien et les instruments à sa disposition, on expérimente et donne corps phénoménologique et sensible aux actions créatrices.En ne considérant pas comme entités distinctes la matière sonore et la structure musicale, on oriente les potentialités de l'ordinateur et crée une dialectique originale et féconde entre le formel (structurel) et le perceptif (cognitif).Le champ d'action envisagé dans cette thèse concerne le développement, la formalisation et la catégorisation des modèles physiques réalisés à l’aide du formalisme CORDIS-ANIMA de structures - créées par modélisation physique - utiles pour la composition musicale, dans la perspective de mettre en évidence la fonction de formalisation musicale associée portée au par ce paradigme. de simulation par modèle physique CORDIS-ANIMA.Le point de départ est ma pratique personnelle de près de 20 années, en tant que compositeur, avec le logiciel de création sonore GENESIS du laboratoire ACROE-ICA. Cette pratique, à travers des travaux à la fois scientifiques (en modélisation) et artistiques (en composition musicale), m'a conduit à considérer cet environnement non pas comme un synthétiseur, mais comme un instrument " organique " permettant de créer une composition musicale complète couvrant les trois échelles de catégorisation usuelle de l'acoustique et de la musique : micro-formelle (le timbre, l'harmonie, l'orchestration), mezzo-formelle (le rythme, la mélodie et les séquences / structures harmoniques de premier niveau) et macro-formelle (la structure harmonique de niveau supérieur, le schéma formel de la composition complète).L'objectif ne consiste pas à proposer le cadre d'une "nouvelle musique" ou d'une nouvelle esthétique, mais à "bien tempérer" les instruments pour une nouvelle pratique de la création musicale explorant et exploitant au mieux les potentialités de l'ordinateur et des technologies numériques, aussi dans la direction d'un élargissement de la dialectique instrumentalité - écriture musicale vers une "supra instrumentalité" [Cadoz6] et vers des perspectives "post-scriptiques" de la création musicale [Cadoz7]. / Physical model synthesis offers an approach to composition alternative to the more usual signal processing. Considering the musical phenomenon as a "unicum" emerging from the interaction between the musicians and their instruments the physical model synthesis realizes and gives phenomenological and sensitive corporeality to the creative actions. By conceiving not as separate entities sound material and musical structure directs the potentialities of the computer and creates an original and fruitful dialectic between the formal (structural) and perceptual (cognitive).The sphere of action for my PhD concerns the development, formalisation and categorisation of structural models – created by physical modelling – useful for musical composition in the perspective to highlight the musical formalisation function associated with the CORDIS-ANIMA physical model simulation paradigm.The starting point for this work is 20 years of personal use as a composer, of the GENESIS physical model musical creation software developed by the ACROE-ICA laboratory.This experience has led me, through works both scientific (modelling) and artistic (music composition), to consider this environment not as a synthesis tool, but as a complex instrument, which allows to create a complete musical composition covering all three usual categories of acoustics and music: micro-formal (the tone, harmony, orchestration), mezzo-formal (the rhythm, melody, and the basic sequences/harmonic structures) and macro-formal (the higher level harmonic structure, the formal outline of the entire composition).My goal is not to propose the framework of a new music or a new aesthetic, but develop "well-tempered" instruments for a new practice of music creativity that explores and accomplishes better the potentialities of computer and digital technology. This also leads in the direction of broadening the dialectic instrumentality - writing music to a "supra instrumentality" [Cadoz6] and to post-scriptic outlook on musical creation [Cadoz7].

Since the heyday of analogue synthesizers in the 70's, they have largely been replaced by digital hardware and software synthesizers. However, in recent years, there has been a revival in analogue designs, possibly due to its ``warmer" sound. This projects aims to take part of this renewal by building a simple analogue synth design with the most basic modules (e.g. oscillators, filters, mixers, amplifier), accompanied by a step sequencer for programming melodies. This will be done by designing circuits and implementing them on breadboards. The circuits were designed with inspiration from various online resources, along with theoretical analysis and simulation software for complex circuitry. The result is a fully functional synthesizer with four sawtooth oscillators. The only modules missing from the initial design are battery support and a line out output for recording the output of the synthesizer. The pitch specification was met as the oscillator did not differ from the expected frequency by more than $\pm$15 cents (hundredths of a semitone), for a range of five octaves. Some possible improvements include better step sequencer user friendliness by installing a display to indicate the notes, more robustness by implementing the synth on a circuit board instead of breadboard. Some improvements can be made for the synth. For example, a display for the step sequencer would facilitate melody programming. Moreover, implementing the synth on a circuit board instead of breadboards would greatly improve robustness and reduce the risk of sound disruptions.