|
|
Multi-channel Methods of Speech Enhancement
Zitka, Adam ; Balík, Miroslav (referee) ; Smékal, Zdeněk (advisor)
This thesis deals with multi-channel methods of speech enhancement. Multichannel methods of speech enhancement use a few microphones for recording signals. From mixtures of signals, for example, individual speakers can be separated, noise should be reduced etc. with using neural networks. The task of separating speakers is known as a cocktail-party effect. The main method of solving this problem is called independent component analysis. At first there are described its theoretical foundation and presented conditions and requirements for its application. Methods of ICA try to separate the mixtures with help of searching the minimal gaussian properties of signals. For the analysis of independent components are used different mathematical properties of signals such as kurtosis and entropy. Signals, which were mixed artificially on a computer, can be relatively well separated using, for example, FastICA algorithm or ICA gradient ascent. However, difficult is situation, if we want to separate the signals created in the real recording enviroment, because the separation of speech people speaking at the same time in the real environment affects other various factors such as acoustic properties of the room, noise, delays, reflections from the walls, the position or the type of microphones, etc. Work presents aproach of independent component analysis in the frequency domain, which can successfully separate also recordings made in the real environment.
|
|
|
Comparison of success rate of multi-channel methods of speech signal separation
Přikryl, Petr ; Zezula, Radek (referee) ; Míča, Ivan (advisor)
The separation of independent sources from mixed observed data is a fundamental problem in many practical situations. A typical example is speech recordings made in an acoustic environment in the presence of background noise or other speakers. Problems of signal separation are explored by a group of methods called Blind Source Separation. Blind Source Separation (BSS) consists on estimating a set of N unknown sources from P observations resulting from the mixture of these sources and unknown background. Some existing solutions for instantaneous mixtures are reviewed and in Matlab implemented , i.e Independent Componnent Analysis (ICA) and Time-Frequency Analysis (TF). The acoustic signals recorded in real environment are not instantaneous, but convolutive mixtures. In this case, an ICA algorithm for separation of convolutive mixtures in frequency domain is introduced and in Matlab implemented. This diploma thesis examines the useability and comparisn of proposed separation algorithms.
|
|
|
Implemetation of algorithms for blind source separation in C/C++ language
Funderák, Marcel ; Malý, Jan (referee) ; Míča, Ivan (advisor)
This thesis is describing one of the methods of Blind Source Separation (BSS) which is Independent Component Analysis. There is shown some brief introduction to the theory behind in which there are explained some basic findings. These findings are important for understanding the theory behind algorithms of ICA. These theoretical findings include primarily explanations of basic knowledge of statistics science. In next part there are described methods which are advisable for preprocessing of input signals – mainly Principal Component Analysis (PCA) and whitening of signals. Mainly whitening is very important part of solution of ICA algorithms. Then there are described different ICA algorithm solutions and especially introduction in this problematic. FastICA algorithm description is mainly depicted because it is very good for computer processing since it is strong and it is less computer demanding than other algorithms. After that follows implementation of one of the ICA algorithm in C++ programming language. FastICA algorithm for complex valued signal was chosen.
|
|
|
Assessment of Independent EEG Components Obtained by Different Methods for BCI Based on Motor Imagery
Húsek, Dušan ; Frolov, A. A. ; Kerechanin, J. V. ; Bobrov, P.D.
Eight methods of decomposition of a multichannel EEG signal are compared in terms of their ability to identify the most physiologically significant components. The criterion for the meaningfulness of a method is its ability to reduce mutual information between components; to create components that can be attributed to the activity of dipoles located in the cerebral cortex; find components that are provided by other methods and for this case; and at the same time, these components should most contribute to the accuracy of the BCI based on imaginary movement. Independent component analysis methods AMICA, RUNICA and FASTICA outperform others in the first three criteria and are second only to the Common Spatial Patterns method in the fourth criterion. The components created by all methods for 386 experimental sessions of 27 subjects were combined into more than 100 clusters containing more than 10 elements. Additionally, the components of the 12 largest clusters were analyzed. They have proven to be of great importance in controlling BCI, their origins can be modeled using dipoles in the brain, and they have been detected by several degradation methods. Five of the 12 selected components have been identified and described in our previous articles. Even if the physiological and functional origins of the rest of identified components’ are to be the subject of further research, we have shown that their physiological nature is at least highly probable.\n
|
|
|
Heuristics in blind source separation
Kautský, Václav ; Štěch, Jakub
This paper deals with application of heuristic algorithms (DEBR, MCRS) in blind source separation (BSS). BSS methods focus on a separation of the (source) signal from a linear mixture. The idea of using heuristic algorithms is introduced on the independent component extraction (ICE) model. The motivation for considering heuristics is to obtain an initial guess needed by many ICE algorithms. Moreover, the comparison of this initialization, and other algorithms accuracy is performed.\n
|
|
|
Implemetation of algorithms for blind source separation in C/C++ language
Funderák, Marcel ; Malý, Jan (referee) ; Míča, Ivan (advisor)
This thesis is describing one of the methods of Blind Source Separation (BSS) which is Independent Component Analysis. There is shown some brief introduction to the theory behind in which there are explained some basic findings. These findings are important for understanding the theory behind algorithms of ICA. These theoretical findings include primarily explanations of basic knowledge of statistics science. In next part there are described methods which are advisable for preprocessing of input signals – mainly Principal Component Analysis (PCA) and whitening of signals. Mainly whitening is very important part of solution of ICA algorithms. Then there are described different ICA algorithm solutions and especially introduction in this problematic. FastICA algorithm description is mainly depicted because it is very good for computer processing since it is strong and it is less computer demanding than other algorithms. After that follows implementation of one of the ICA algorithm in C++ programming language. FastICA algorithm for complex valued signal was chosen.
|
|
|
Multi-channel Methods of Speech Enhancement
Zitka, Adam ; Balík, Miroslav (referee) ; Smékal, Zdeněk (advisor)
This thesis deals with multi-channel methods of speech enhancement. Multichannel methods of speech enhancement use a few microphones for recording signals. From mixtures of signals, for example, individual speakers can be separated, noise should be reduced etc. with using neural networks. The task of separating speakers is known as a cocktail-party effect. The main method of solving this problem is called independent component analysis. At first there are described its theoretical foundation and presented conditions and requirements for its application. Methods of ICA try to separate the mixtures with help of searching the minimal gaussian properties of signals. For the analysis of independent components are used different mathematical properties of signals such as kurtosis and entropy. Signals, which were mixed artificially on a computer, can be relatively well separated using, for example, FastICA algorithm or ICA gradient ascent. However, difficult is situation, if we want to separate the signals created in the real recording enviroment, because the separation of speech people speaking at the same time in the real environment affects other various factors such as acoustic properties of the room, noise, delays, reflections from the walls, the position or the type of microphones, etc. Work presents aproach of independent component analysis in the frequency domain, which can successfully separate also recordings made in the real environment.
|
|
|
Comparison of success rate of multi-channel methods of speech signal separation
Přikryl, Petr ; Zezula, Radek (referee) ; Míča, Ivan (advisor)
The separation of independent sources from mixed observed data is a fundamental problem in many practical situations. A typical example is speech recordings made in an acoustic environment in the presence of background noise or other speakers. Problems of signal separation are explored by a group of methods called Blind Source Separation. Blind Source Separation (BSS) consists on estimating a set of N unknown sources from P observations resulting from the mixture of these sources and unknown background. Some existing solutions for instantaneous mixtures are reviewed and in Matlab implemented , i.e Independent Componnent Analysis (ICA) and Time-Frequency Analysis (TF). The acoustic signals recorded in real environment are not instantaneous, but convolutive mixtures. In this case, an ICA algorithm for separation of convolutive mixtures in frequency domain is introduced and in Matlab implemented. This diploma thesis examines the useability and comparisn of proposed separation algorithms.
|
|
|
Blind Separation of Mixtures of Piecewise AR(1) Processes and Model Mismatch
Tichavský, Petr ; Šembera, Ondřej ; Koldovský, Zbyněk
Modeling real-world acoustic signals and namely speech signals as piecewise stationary random processes is a possible approach to blind separation of linear mixtures of such signals. In this paper, the piecewise AR(1) modeling is studied and is compared to the more common piecewise AR(0) modeling, which is known under the names Block Gaussian SEParation (BGSEP) and Block Gaussian Likelihood (BGL). The separation based on the AR(0) modeling uses an approximate joint diagonalization (AJD) of covariance matrices of the mixture with lag 0, computed at epochs (intervals) of stationarity of the separated signals. The separation based on the AR(1) modeling uses the covariances of lag 0 and covariances of lag 1 jointly. For this model, we derive an approximate Cram´er-Rao lower bound on the separation accuracy for estimation based on the full set of the statistics (covariance matrices of lag 0 and lag 1) and covariance matrices with lag 0 only. The bounds show the condition when AR(1) modeling leads to significantly improved separation accuracy.
|
|
|
On Sparsity in Bayesian Blind Source Separation for Dynamic Medical Imaging
Tichý, Ondřej
Dynamic medical imaging is concerned with acquisition and analysis of a sequence of images of the same region of a body during time. In nuclear medicine, each pixel of an image is the sum of particles coming from an applied radioactive tracer from the body in a specific time-interval. Hence, each observed image is a superposition of an unknown number of underlaying organ images. The aim of blind source separation is to separate the images of biologic organs and related time-activity curves from the sequence of images.
|
guest ::
