No description available.

Cette thèse propose des algorithmes de séparation aveugle de sources audio en utilisant un réseau de capteurs. L'application finale de ces algorithmes est l'audition des robots dans le cadre du projet ROMEO. Dans cette thèse, nous avons développé des algorithmes de séparation aveugle de sources audio basés sur des critères de parcimonie. Nous montrons que la minimisation de la norme l1 avec une technique d'optimisation du gradient naturel permet d’élaborer un algorithme se situant au niveau de l’état de l’art. Nous montrons qu’un critère basé sur la paramétrisation de la pseudo-norme lp, avec 0<p<1 en améliore les performances. Ceci revient à rendre la contrainte de parcimonie plus dure au fur et à mesure que l'algorithme avance dans ses itérations. Pour exploiter l'aspect multicapteurs de notre application (16 capteurs sont fixés autour de la tête de l'humanoïde), nous avons proposé des algorithmes de séparation avec une étape de prétraitement de formation de voies fixe. Dans le cas de l'audition des robots, les capteurs sont souvent placés sur la tête de l'humanoïde. Afin de tenir compte de l'influence de la tête sur le champ sonore proche, nous avons construit les filtres de formation en utilisant les fonctions de transfert de tête (HRTF) du robot. L'étape de formation de voies améliore les résultats de séparation par rapport à l'utilisation d'un algorithme de séparation seule de minimum 10dB. Cette thèse propose aussi les versions adaptatives de ces algorithmes. Dans le scénario réel où le nombre de sources présentes dans l'environnement du robot est inconnu et change au cours du temps, nous montrons comment détecter et suivre le nombre de sources. / This thesis proposes blind audio source separation algorithms using a microphone array. The final application of these algorithms is robot audition through the ROMEO project. In this thesis, we developed blind source separation algorithms based on a sparcity criterion. We show that l1 minimization using the natural gradient optimization technique has the same performance that the state of the art. We show that a criterion based on the parametrization of the quazi-norm lp, with 0<p<1, improves the previous results: the sparsity criterion gets harder through the iterations of the algorithm. Then, we exploited the multisensor aspect of our application (16 sensors are fixed in the head of the humanoid) and we proposed a separation algorithms with a fixed beamforming preprocessing step. In the robot audition case, the sensors are often placed on the head of the humanoid. To take into account the influence of the head in the near sound manifold, we built the beamforming filters using the head related transfer functions (HRTF) of the robot. The beamforming step improves the separation results compared to the use of a blind source separation only. This thesis also proposes the adaptive versions of those algorithms. In the real scenario where the number of sources is unknown and changes, we show how to detect and follow the number of sources.

Este trabalho tem por objetivo desenvolver ferramentas que permitam gerar em fones-de-ouvido o efeito psicoacústico de fontes sonoras locomovendo-se no espaço, por meio da auralização do sinal monofônico original. Embora a auralização binaural possa ser feita empregando variações de atraso (chamadas ITD interaural time difference, ou diferença de tempo interaural) e de intensidade (chamadas ILD interaural level difference, ou diferença de nível interaural) entre os canais, melhores resultados psicoacústicos podem ser obtidos ao se utilizar filtros digitais conhecidos como HRTFs (head related transfer functions, ou funções de transferência relativas à cabeça). Uma HRTF insere no sinal monofônico informações que possibilitam ao sistema auditivo identificá-lo como proveniente de uma direção específica, direção esta que é única para cada HRTF. Para posicionar uma fonte estática em uma direção específica, bastaria, então, filtrar o sinal original pela HRTF da direção desejada. Se, no entanto, for desejável que a fonte se locomova em uma trajetória contínua, um número infinitamente grande de filtros seria necessário. Como eles são, normalmente, obtidos empiricamente, um número arbitrariamente alto deles não está disponível. Disso surge a necessidade de técnicas de interpolação de HRTFs, que possibilitem gerar os filtros intermediários não disponíveis. Este trabalho apresenta três novas técnicas de interpolação de HRTFs, para assim alcançar o objetivo de auralizar fontes sonoras móveis: a interpolação triangular, que é uma técnica de interpolação linear baseada na técnica de panorama sonoro VBAP (vector-based amplitude panning, ou panorama sonoro baseado em vetores); o método das movimentações discretas, que busca explorar o limiar de percepção do nosso sistema auditivo para, com isso, gerar uma técnica extremamente barata computacionalmente; e a interpolação espectral, que altera continuamente as estruturas das HRTFs para gerar filtros interpolados. São apresentadas também as implementações feitas dessas novas técnicas desenvolvidas, bem como os testes numéricos realizados para medir sua eficácia. / The goal of this work is the development of tools that allow simulating through headphones the psychoacoustic effect of sound sources moving in space, by the auralisation of the original monophonic signals. Although binaural auralisation can be implemented using variations in delays (called ITD interaural time difference) and in intensities (called ILD interaural level difference) among channels, better psychoacoustic results can be achieved using digital filters known as HRTFs (head related transfer functions). A HRTF inserts in the monophonic signal information that allow the auditory system to perceive this signal to be as if coming from a specific direction, which is unique for each single HRTF. Thus, to position a static sound source at a specific direction, filtering the original signal with the HRTF from the desired direction would be enough. Nevertheless, if it is desired that the sound source moves in a continuous trajectory, an infinitely large amount of filters would be necessary. Since they are usually obtained by measurements, such an arbitrarily large amount of them is not available. In this case, HRTF interpolation techniques that generate intermediary filters must be used. This work presents three new HRTF interpolation techniques in order to auralise moving sound sources: the triangular interpolation, a linear interpolation technique based on the VBAP amplitude panning technique; the discrete movements method, an extremely efficient technique that exploits the auditory systems limitations in perceiving very small changes in direction; and the spectral interpolation, that alters continuously the structures of the HRTFs to generate interpolated filters. Implementations of these techniques are discussed and numerical tests are also presented.

Este trabalho tem por objetivo desenvolver ferramentas que permitam gerar em fones-de-ouvido o efeito psicoacústico de fontes sonoras locomovendo-se no espaço, por meio da auralização do sinal monofônico original. Embora a auralização binaural possa ser feita empregando variações de atraso (chamadas ITD interaural time difference, ou diferença de tempo interaural) e de intensidade (chamadas ILD interaural level difference, ou diferença de nível interaural) entre os canais, melhores resultados psicoacústicos podem ser obtidos ao se utilizar filtros digitais conhecidos como HRTFs (head related transfer functions, ou funções de transferência relativas à cabeça). Uma HRTF insere no sinal monofônico informações que possibilitam ao sistema auditivo identificá-lo como proveniente de uma direção específica, direção esta que é única para cada HRTF. Para posicionar uma fonte estática em uma direção específica, bastaria, então, filtrar o sinal original pela HRTF da direção desejada. Se, no entanto, for desejável que a fonte se locomova em uma trajetória contínua, um número infinitamente grande de filtros seria necessário. Como eles são, normalmente, obtidos empiricamente, um número arbitrariamente alto deles não está disponível. Disso surge a necessidade de técnicas de interpolação de HRTFs, que possibilitem gerar os filtros intermediários não disponíveis. Este trabalho apresenta três novas técnicas de interpolação de HRTFs, para assim alcançar o objetivo de auralizar fontes sonoras móveis: a interpolação triangular, que é uma técnica de interpolação linear baseada na técnica de panorama sonoro VBAP (vector-based amplitude panning, ou panorama sonoro baseado em vetores); o método das movimentações discretas, que busca explorar o limiar de percepção do nosso sistema auditivo para, com isso, gerar uma técnica extremamente barata computacionalmente; e a interpolação espectral, que altera continuamente as estruturas das HRTFs para gerar filtros interpolados. São apresentadas também as implementações feitas dessas novas técnicas desenvolvidas, bem como os testes numéricos realizados para medir sua eficácia. / The goal of this work is the development of tools that allow simulating through headphones the psychoacoustic effect of sound sources moving in space, by the auralisation of the original monophonic signals. Although binaural auralisation can be implemented using variations in delays (called ITD interaural time difference) and in intensities (called ILD interaural level difference) among channels, better psychoacoustic results can be achieved using digital filters known as HRTFs (head related transfer functions). A HRTF inserts in the monophonic signal information that allow the auditory system to perceive this signal to be as if coming from a specific direction, which is unique for each single HRTF. Thus, to position a static sound source at a specific direction, filtering the original signal with the HRTF from the desired direction would be enough. Nevertheless, if it is desired that the sound source moves in a continuous trajectory, an infinitely large amount of filters would be necessary. Since they are usually obtained by measurements, such an arbitrarily large amount of them is not available. In this case, HRTF interpolation techniques that generate intermediary filters must be used. This work presents three new HRTF interpolation techniques in order to auralise moving sound sources: the triangular interpolation, a linear interpolation technique based on the VBAP amplitude panning technique; the discrete movements method, an extremely efficient technique that exploits the auditory systems limitations in perceiving very small changes in direction; and the spectral interpolation, that alters continuously the structures of the HRTFs to generate interpolated filters. Implementations of these techniques are discussed and numerical tests are also presented.

Avec la généralisation de l'écoute au casque, la technique binaurale apparaît comme une solution privilégiée pour démocratiser l'accès à des contenus sonores spatialisés. La synthèse binaurale repose sur l'utilisation de filtres appelés HRTFs qui restituent à l'auditeur l'ensemble des indices acoustiques de localisation. Ceux-ci possèdent une composante individuelle forte et l'utilisation de filtres non-individuels entraîne des défauts de localisation ou de timbre à la restitution. L’acquisition individuelle des HRTFs requiert un dispositif de mesure complexe. Les méthodes d’individualisation visent à offrir des solutions alternatives à cette mesure individuelle et reposent généralement sur l’exploitation de bases de données de HRTFs. A cet effet, une nouvelle base de données de HRTFs à haute résolution spatiale et fréquentielle a été constituée. Le développement d’un modèle visant à prédire les directions perçues de sources virtuelles synthétisées avec des HRTFs non-individuelles constitue le cœur du travail de thèse. La métrique utilisée pour quantifier la similarité entre HRTFs fait l’objet d’une attention particulière. L’objectif ultime est d’évaluer dans quelle mesure ce modèle peut être utilisé pour sélectionner le jeu de HRTFs optimal pour un individu à partir de ses réponses dans un test de localisation de sources sonores virtuelles synthétisées avec des HRTFs non-individuelles. La mise en place de tels tests implique l’utilisation d’une méthode de report, responsable d’une part d’erreur dans les réponses. La thèse comprend une étude comparative de 3 méthodes de report et la préconisation d’une méthode mieux adaptée au contexte de l’écoute binaurale au casque. / With the spread of headphone listening, binaural technology appears as the most appropriate solution to democratize the access to spatialized audio contents. Binaural synthesis of virtual sound sources is based on the use of filters called HRTFs, which provide the listener with accurate localization cues. These cues are however highly listener-dependent and the use of non-individual HRTFs may lead to localization and timbre artefacts.Individual acquisition of HRTFs requires a complex measurement setup installed in an anechoic chamber which is incompatible with large scale deployment. Therefore, individualization methods have been devised in order to offer alternatives to this individual measurement. They are often based on the exploitation of large HRTFs databases. To this end, a new HRTFs database with high spatial and frequency resolution has been created. The development of a model that predicts the perceived direction of a virtual source synthesized with non-individual HRTFs is the core of the thesis work. The choice of the metric used for quantifying the similarity between HRTFs receives a particular attention. The ultimate goal is to evaluate how such a model can be used to select automatically the optimal HRTFs set for an individual, from the observation of his responses in a localization test of virtual sound sources synthesized with non-individual HRTFs. The implementation of such a test implies the use of a reporting method, which may introduce some bias in the responses. This thesis includes a comparative study of 3 reporting methods and the recommendation of a method more suitable in the context of binaural listening through headphones.

Cette thèse propose des algorithmes de séparation aveugle de sources audio en utilisant un réseau de capteurs. L'application finale de ces algorithmes est l'audition des robots dans le cadre du projet ROMEO. Dans cette thèse, nous avons développé des algorithmes de séparation aveugle de sources audio basés sur des critères de parcimonie. Nous montrons que la minimisation de la norme l1 avec une technique d'optimisation du gradient naturel permet d'élaborer un algorithme se situant au niveau de l'état de l'art. Nous montrons qu'un critère basé sur la paramétrisation de la pseudo-norme lp, avec 0

Binaural simulation is the recreation of a three-dimensional audio environment around a listener's head. The binaural simulation of structural acoustic data would open new opportunities in virtual prototyping and simulation. By modeling the structure as an array of vibrating monopoles and applying Head Related Transfer Functions (HRTFs) to each of the sources, a binaural simulation of this type can be created. Unfortunately, this simulation method requires an extensive amount of computer power and speed for real-time simulation, more so than is available with current technology. The objective of this research is to reduce the number of computations required in the binaural simulation of structural acoustic data. This thesis details two methods of reducing the number of real-time calculations required in this binaural analysis: singular value decomposition (SVD), and equivalent source reduction (ESR). The SVD method reduces the complexity of the HRTF computations by breaking the HRTFs into dominant singular values and vectors. The ESR method reduces the number of sources to be analyzed in real-time by replacing sources on the scale of a structural wavelength with sources on the scale of an acoustic wavelength. The ESR and SVD reduction methods can be combined to provide an estimated computation time reduction of 99.4%. In addition, preliminary tests show that there is a 97% correlation between the results of the combined reduction methods and the results found with current binaural simulation techniques. / Master of Science

Cette thèse se concentre sur l'individualisation des HRTF pour la synthèse binaurale. Les systèmes permettant de mesurer des HRTF sont difficiles d'accès au grand public, ce qui pose un problème pour la démocratisation de l'audio binaurale de haute qualité. C'est pour cette raison que nous recherchons une méthode permettant de calculer les HRTF d'une personne qu'à partir de la connaissance de sa morphologie. Ces travaux de thèse avancent sur les investigations visant à avoir une technique de personnalisation de l'audio binaurale. La technique de personnalisation proposée part du principe qu'il existe un lien direct et systématique liant des paramètres de la morphologie d'un individu avec les indices spectraux de ses HRTF. Elle est basée sur la détermination d'une fonction qui prend en entrée des paramètres morphologiques et qui donne en sortie des paramètres de transformation à appliquer à un jeu de HRTF existant pour obtenir un jeu de HRTF personnalisé. Cette fonction est estimée à partir d'analyses statistiques faits sur une base de données contenant des modèles numériques des individus ainsi que leurs HRTF. Pour atteindre cet objectif, il faut construire dans un premier temps les données à partir desquelles nous allons estimer notre fonction de personnalisation. Ces travaux expliquent les systèmes que nous avons conçus pour mesurer les HRTF des individus ainsi que pour obtenir leur modèle numérique tridimensionnel. De plus, les protocoles de mesure associés à chaque système sont expliqués. Enfin, nous expliquons comment les données obtenues peuvent être exploitées pour développer un procédé de personnalisation des HRTF. / This thesis focuses on the HRTF individualization problem in the context of binaural synthesis for general applications. HRTF strongly depend on morphological features of a person and, in order to provide compelling auditory spaces, binaural synthesis requires the use of individualized HRTF. Measuring or calculating the HRTF of a listener are common but lengthy and costly methods that are not feasible for general public applications. That is the reason why we aim to develop an alternative technique to obtain customized HRTF. The technique proposed relies on estimating the spectral cues of the HRTF, corresponding to the colorations induced by pinna filtering, from a person's morphology. The spectral cues represent the most complex and individual part of HRTF. The work presented in this thesis is based on the existence of a direct and systematic link between the morphology of a person and the spectral cues of their HRTF. The goal is to start from individualization techniques of HRTF and make studies for better understanding the relationship between the morphology of a person and their HRTF. The ultimate goal is then to find a proper morphological matching personalization technique. The first step to achieve this goal is to build the measuring systems who will help us create the databases that we will use for our analyses. This thesis explains the two measuring systems that were created and the measuring protocols that were used to create two related databases containing the 3D models and the measured HRTFs of a collection of people. We then explain how these databases can be used to fit the HRTF individualization technique we have proposed.

Ces travaux de thèse concernent la conception d’un dispositif de restitution sonore léger en salle usuelle, permettant la diffusion de signaux binauraux. La priorité du travail est la précision dans la reproduction du niveau et du timbre, avant la spatialisation du son. Afin d’assurer la compatibilité avec les signaux binauraux existants et pour proposer un système à faible nombre de canaux, la technologie transaurale est prise comme point de départ. Pour limiter la coloration introduite par la salle d’écoute, particulièrement gênante, il est proposé de placer les sources du système à proximité de l’auditeur afin de maximiser le rapport champ direct sur champ diffus. Ce placement inhabituel a plusieurs effets, parmi lesquels quatre ont été étudiés séparément : l’influence des filtres transauraux sur les sources électro-acoustiques, les variations inter-individuelles de morphologies, l’effet de salle et le placement de l’auditeur. Des tests d’écoute ont été réalisés pour une sélection de configurations, et les résultats ont permis d’implémenter des indicateurs objectifs représentatifs des réponses des auditeurs. La synthèse de ces indicateurs a permis de proposer trois configurations considérées comme optimales, et dont la combinaison pourrait être envisagée. / This work deals with the design of a 3D sound system involving a few number of loudspeaker and able to work inside any usual room, for reproducing binaural sounds. This system focuses on an accurate reproduction of perceived level and timbre, even before the sound spatialization. To ensure compatibility with binaural recordings and to achieve a system with a low number of loudspeakers, this work started from a transaural system. To avoid tone coloration induced by the listening room, the sound sources are placed close to the listener, thus maximizing the energy ratio between direct and diffuse fields. This has consequences on other aspects, four of which are considered separately: inter-individual morphological variations, demands on the electro-acoustic sources, room effect and misalignment of the listener. Some configurations have been evaluated in listening tests, and objective indicators are deduced from these results. The generalization allows to propose three configurations considered as optimal, and which might be combined.

Les technologies nomades (smartphones, tablettes, . . . ) étant actuellement très répandues,nous avons souhaité, dans le cadre de cette thèse, les utiliser comme vecteur pour proposer au grand public des outils de spatialisation sonore. La taille et le nombre de transducteurs utilisés pour la captation et la restitution sonore spatialisée sont à ce jour la limitation principale pour une utilisation nomade. Dans une première étape, la captation d’un opéra pour une restitution sur des tablettes tactiles nous a permis d’évaluer les technologies audio 3D disponibles aujourd’hui. Les résultats de cette évaluation ont révélé que l’utilisation des quatre capteurs du microphone Soundfield donne de bons résultats à condition d’effectuer un décodage binaural adapté pour une restitution sur casque. Selon une approche inspirée des méthodes de localisation de source et le concept de format « objet », un prototype de prise de son 3D léger et compact a été développé. Le dispositif microphonique proposé se compose de trois capsules microphoniques cardioïdes. A partir des signaux microphoniques, un algorithme de post-traitement spatial est capable, d’une part, de déterminer la direction des sources et, d’autre part, d’extraire un signal sonore représentatif de la scène spatiale. Ces deux informations permettent ainsi de caractérisercomplètement la scène sonore 3D en fournissant un encodage spatial offrant le double avantage d’une compression de l’information audio et d’une flexibilité pour le choix du système de reproduction. En effet, la scène sonore ainsi encodée peut être restituée en utilisant un décodage adapté sur n’importe quel type de dispositif.Plusieurs méthodes de localisation et différentes configurations microphoniques (géométrie et directivité) ont été étudiées.Dans une seconde étape, l’algorithme d’extraction de l’information spatiale a été modifié pour prendre en compte les caractéristiques réelles in situ des microphones.Des méthodes pour compléter la chaîne acoustique sont proposées permettant la restitution binaurale ainsi que sur tout autre dispositif de restitution. Elles proposent l’utilisation de capteurs de localisation présents sur les terminaux mobiles afin d’exploiter les capacités qu’ils offrent aujourd’hui. / Mobile technologies (such as smartphones and tablets) are now common devices of the consumer market. In this PhD we want to use those technologies as the way to introduce tools of sound spatialization into the mass market. Today the size and the number of traducers used to pick-up and to render a spatial sound scene are the main factors which limit the portability of those devices. As a first step, a listening test, based on a spatial audio recording of an opera, let us to evaluate the 3D audio technologies available today for headphone rendering. The results of this test show that, using the appropriate binaural decoding, it is possible to achieve a good binaural rendering using only the four sensors of the Soundfield microphone.Then, the steps of the development of a 3D sound pick-up system are described. Several configurations are evaluated and compared. The device, composed of 3 cardioid microphones, was developed following an approach inspired by the sound source localization and by the concept of the "object format encoding". Using the microphone signals and an adapted post-processing it is possible to determine the directions of the sources and to extract a sound signal which is representative of the sound scene. In this way, it is possible to completely describe the sound scene and to compress the audio information.This method offer the advantage of being cross platform compatible. In fact, the sound scene encoded with this method can be rendered over any reproduction system.A second method to extract the spatial information is proposed. It uses the real in situ characteristics of the microphone array to perform the sound scene analysis.Some propositions are made to complement the 3D audio chain allowing to render the result of the sound scene encoding over a binaural system or any king of speaker array using all capabilities of the mobile devices.