No description available.

Σκοπός της παρούσας διπλωματικής εργασίας είναι η δημιουργία ενός συστήματος μίξης ηχητικών σημάτων και προσπάθεια διαχωρισμού τους με βάση τις μεθόδους τυφλού διαχωρισμού σημάτων. Έχοντας ως δεδομένα τα αρχικά σήματα των πηγών γίνεται προσπάθεια, αρχικά μέσω της εφαρμογής της μεθόδου Ανάλυσης Ανεξάρτητων Συνιστωσών (ICA) για την περίπτωση της στιγμιαίας μίξης και στη συνέχεια μέσω της χρήσης αλγορίθμων που στηρίζονται στο μοντέλο παράλληλου παράγοντα (PARAFAC) για την περίπτωση της συνελικτικής μίξης, να προσδιοριστούν τα σήματα των πηγών από τα σήματα μίξης. Επιπλέον, τροποποιώντας τις παραμέτρους του συστήματος που μελετάμε σε κάθε περίπτωση, προσπαθούμε να πετύχουμε τη βέλτιστη απόδοση του διαχωρισμού. / The subject of this diploma thesis is the creation of a mixing system of speech signals and the attempt of their separation using the methods of blind source separation (BSS). Considering the original source signals known, we attempt, firstly by using independent component analysis for instantaneous mixtures and then by using PARAFAC model for convolutive mixtures, to extract the original source signals from the mixing signals. Moreover, by modifying the parameters of the system we make an effort to achieve the best performance of the separation.

Orientador: João Marcos Travassos Romano / Tese (doutorado) - Universidade Estadual de Campinas, Faculdade de Engenharia Eletrica e de Computação / Made available in DSpace on 2018-08-09T16:56:34Z (GMT). No. of bitstreams: 1 Suyama_Ricardo_D.pdf: 28793623 bytes, checksum: cf06bdad425402b4624bbd169bfad249 (MD5) Previous issue date: 2007 / Resumo: O problema de separação cega de fontes (BSS - Blind Source Separation) vem despertando o interesse de um número crescente de pesquisadores. Esse destaque é devido, em grande parte, à formulação abrangente do problema, que torna possível o uso das técnicas desenvolvidas no contexto de BSS nas mais diversas áreas de aplicação. O presente trabalho tem como objetivo propor novos métodos de solução do problema de separação cega de fontes, nos casos de mistura convolutiva e mistura não-linear. Para o primeiro caso propomos um método baseado em predição não-linear, cujo intuito é eliminar o caráter convolutivo da mistura e, dessa forma, separar os sinais utilizando ferramentas bem estabelecidas no contexto de misturas lineares sem memória. No contexto de misturas não-lineares, propomos uma nova metodologia para separação de sinais em um modelo específico de mistura denominado modelo com não-linearidade posterior (PNL - Post Nonlinear ). Com o intuito de minimizar problemas de convergência para mínimos locais no processo de adaptação do sistema separador, o método proposto emprega um algoritmo evolutivo como ferramenta de otimização, e utiliza um estimador de entropia baseado em estatísticas de ordem para avaliar a função custo. A eficácia de ambos os métodos é verificada através de simulações em diferentes cenários / Abstract: The problem of blind source separation (BSS) has attracted the attention of agrowing number of researchers, mostly due to its potential applications in a significant number of different areas. The objective of the present work is to propose new methods to solve the problem of BSS in the cases of convolutive mixtures and nonlinear mixtures. For the first case, we propose a new method based on nonlinear prediction filters. The nonlinear structure is employed to eliminate the convolutive character of the mixture, hence converting the problem into an instantaneous mixture, to which several well established tools may be used to recover the sources. In the context of nonlinear mixtures, we present a new methodology for signal separation in the so-called post-nonlinear mixing models (PNL). In order to avoid convergence to local minima, the proposed method uses an evolutionary algorithm to perform the optimization of the separating system. In addition to that, we employ an entropy estimator based on order-statistics to evaluate the cost function. The effectiveness of both methods is assessed through simulations in different scenarios / Doutorado / Telecomunicações e Telemática / Doutor em Engenharia Elétrica

SAAB has developed an autonomous underwater vehicle that can mimic a conventional submarine for military fleets to exercise anti-submarine warfare. The AUV actively emits amplified versions of received sonar pulses to create the illusion of being a larger object. To prevent acoustic feedback, the AUV must distinguish between the sound to be actively responded to and its emitted signal. This master thesis has examined techniques aimed at preventing the AUV from responding to previously emitted signals to avoid acoustical feedback, without relying on prior knowledge of either the received signal or the signal emitted by the AUV. The two primary types of algorithms explored for this problem include blind source separation and adaptive filtering. The adaptive filters based on Leaky Least Mean Square and Kalman have shown promising results in attenuating the active response from the received signal. The adaptive filters utilize the fact that a certain hydrophone primarily receives the active response. This hydrophone serves as an estimate of the active response since the signal it captures is considered unknown and is to be removed. The techniques based on blind source separation have utilized the recordings of three hydrophones placed at various locations of the AUV to separate and estimate the received signal from the one emitted by the AUV. The results have demonstrated that neither of the reviewed methods is suitable for implementation on the AUV. The hydrophones are situated at a considerable distance from each other, resulting in distinct time delays between the reception of the two signals. This is usually referred to as a convolutive mixture. This is commonly solved using the frequency domain to transform the convolutive mixture to an instantaneous mixture. However, the fact that the signals share the same frequency spectrum and are adjacent in time has proven highly challenging.

Nous considérons le problème de la séparation de mélanges audio réverbérants déterminés et sous-déterminés, c'est-à-dire l'extraction du signal de chaque source dans un mélange multicanal. Nous proposons un cadre général de modélisation gaussienne où la contribution de chaque source aux canaux du mélange dans le domaine temps-fréquence est modélisée par un vecteur aléatoire gaussien de moyenne nulle dont la covariance encode à la fois les caractéristiques spatiales et spectrales de la source. A n de mieux modéliser la réverbération, nous nous aff ranchissons de l'hypothèse classique de bande étroite menant à une covariance spatiale de rang 1 et nous calculons la borne théorique de performance atteignable avec une covariance spatiale de rang plein. Les ré- sultats expérimentaux indiquent une augmentation du rapport Signal-à-Distorsion (SDR) de 6 dB dans un environnement faiblement à très réverbérant, ce qui valide cette généralisation. Nous considérons aussi l'utilisation de représentations temps-fréquence quadratiques et de l'échelle fréquentielle auditive ERB (equivalent rectangular bandwidth) pour accroître la quantité d'information exploitable et décroître le recouvrement entre les sources dans la représentation temps-fréquence. Après cette validation théorique du cadre proposé, nous nous focalisons sur l'estimation des paramètres du modèle à partir d'un signal de mélange donné dans un scénario pratique de séparation aveugle de sources. Nous proposons une famille d'algorithmes Expectation-Maximization (EM) pour estimer les paramètres au sens du maximum de vraisemblance (ML) ou du maximum a posteriori (MAP). Nous proposons une famille d'a priori de position spatiale inspirée par la théorie de l'acoustique des salles ainsi qu'un a priori de continuité spatiale. Nous étudions aussi l'utilisation de deux a priori spectraux précédemment utilisés dans un contexte monocanal ou multicanal de rang 1: un a priori de continuité spatiale et un modèle de factorisation matricielle positive (NMF). Les résultats de séparation de sources obtenus par l'approche proposée sont comparés à plusieurs algorithmes de base et de l'état de l'art sur des mélanges simulés et sur des enregistrements réels dans des scénarios variés.