The Experts below are selected from a list of 33405 Experts worldwide ranked by ideXlab platform
Abderrahmane Amrouche - One of the best experts on this subject based on the ideXlab platform.
-
fusion strategies for distributed speaker recognition using Residual Signal based g729 resynthesized speech
International Conference on Information Fusion, 2013Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:With the development of VoIP (Voice over IP) service, there is an emerging need to speech compression, particularly for digital speech communication and biometric speaker recognition (SR) systems. This paper presents results issued from Universal Background Gaussian Mixture Model (GMM UBM) based SR system, that is trained and tested on clean and G729 resynthesized speech. To overcome the performance loss due to the G729 codec, Residual Signal extracted from clean and G729 resynthesized database is used. To get better the performance, we investigated score fusion strategies based on Logistic Regression (LR). The first fusion based on GMM UBM score using LFCC (Linear Frequency Cepstrum Coefficients) and LFCC extracted from LP (Linear Prediction) Residual Signal. The second used the LFCC extracted from G729 resynthesized speech and its LP Residual Signal. The best performance is obtained by Logistic Regression (LR) fusion. The correct rate in the first case is 95% based baseline system and 83% based G729 resynthesized speech in the second case. The obtained results, using TIMIT database, have proven the efficiency of data fusion techniques for automatic speaker recognition.
-
FUSION - Fusion strategies for distributed speaker recognition using Residual Signal based G729 resynthesized speech
2013Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:With the development of VoIP (Voice over IP) service, there is an emerging need to speech compression, particularly for digital speech communication and biometric speaker recognition (SR) systems. This paper presents results issued from Universal Background Gaussian Mixture Model (GMM UBM) based SR system, that is trained and tested on clean and G729 resynthesized speech. To overcome the performance loss due to the G729 codec, Residual Signal extracted from clean and G729 resynthesized database is used. To get better the performance, we investigated score fusion strategies based on Logistic Regression (LR). The first fusion based on GMM UBM score using LFCC (Linear Frequency Cepstrum Coefficients) and LFCC extracted from LP (Linear Prediction) Residual Signal. The second used the LFCC extracted from G729 resynthesized speech and its LP Residual Signal. The best performance is obtained by Logistic Regression (LR) fusion. The correct rate in the first case is 95% based baseline system and 83% based G729 resynthesized speech in the second case. The obtained results, using TIMIT database, have proven the efficiency of data fusion techniques for automatic speaker recognition.
-
ICISP - SVM based GMM supervector speaker recognition using LP Residual Signal
Lecture Notes in Computer Science, 2012Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:Feature extraction is an important step for speaker recognition systems. In this paper, we generated MFCC (Mel Frequency Cepstral Coefficients) and LPCC (Linear Predictive Cepstral Coefficients) from LP Residual of speech Signal, instead their calculation directly from speech samples. These features represent complementary vocal cord information's. In this work, Universal Background Gaussian Mixture Models (GMM-UBM) and Gaussian Supervector (GMM-SVM) based speaker modeling have been used. Experimental results, using, ARADIGITS data-base, show the efficiency of the GMM-SVM based approach associated with feature vectors issued from LP Residual Signal.
Dalila Yessad - One of the best experts on this subject based on the ideXlab platform.
-
fusion strategies for distributed speaker recognition using Residual Signal based g729 resynthesized speech
International Conference on Information Fusion, 2013Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:With the development of VoIP (Voice over IP) service, there is an emerging need to speech compression, particularly for digital speech communication and biometric speaker recognition (SR) systems. This paper presents results issued from Universal Background Gaussian Mixture Model (GMM UBM) based SR system, that is trained and tested on clean and G729 resynthesized speech. To overcome the performance loss due to the G729 codec, Residual Signal extracted from clean and G729 resynthesized database is used. To get better the performance, we investigated score fusion strategies based on Logistic Regression (LR). The first fusion based on GMM UBM score using LFCC (Linear Frequency Cepstrum Coefficients) and LFCC extracted from LP (Linear Prediction) Residual Signal. The second used the LFCC extracted from G729 resynthesized speech and its LP Residual Signal. The best performance is obtained by Logistic Regression (LR) fusion. The correct rate in the first case is 95% based baseline system and 83% based G729 resynthesized speech in the second case. The obtained results, using TIMIT database, have proven the efficiency of data fusion techniques for automatic speaker recognition.
-
FUSION - Fusion strategies for distributed speaker recognition using Residual Signal based G729 resynthesized speech
2013Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:With the development of VoIP (Voice over IP) service, there is an emerging need to speech compression, particularly for digital speech communication and biometric speaker recognition (SR) systems. This paper presents results issued from Universal Background Gaussian Mixture Model (GMM UBM) based SR system, that is trained and tested on clean and G729 resynthesized speech. To overcome the performance loss due to the G729 codec, Residual Signal extracted from clean and G729 resynthesized database is used. To get better the performance, we investigated score fusion strategies based on Logistic Regression (LR). The first fusion based on GMM UBM score using LFCC (Linear Frequency Cepstrum Coefficients) and LFCC extracted from LP (Linear Prediction) Residual Signal. The second used the LFCC extracted from G729 resynthesized speech and its LP Residual Signal. The best performance is obtained by Logistic Regression (LR) fusion. The correct rate in the first case is 95% based baseline system and 83% based G729 resynthesized speech in the second case. The obtained results, using TIMIT database, have proven the efficiency of data fusion techniques for automatic speaker recognition.
-
ICISP - SVM based GMM supervector speaker recognition using LP Residual Signal
Lecture Notes in Computer Science, 2012Co-Authors: Dalila Yessad, Abderrahmane AmroucheAbstract:Feature extraction is an important step for speaker recognition systems. In this paper, we generated MFCC (Mel Frequency Cepstral Coefficients) and LPCC (Linear Predictive Cepstral Coefficients) from LP Residual of speech Signal, instead their calculation directly from speech samples. These features represent complementary vocal cord information's. In this work, Universal Background Gaussian Mixture Models (GMM-UBM) and Gaussian Supervector (GMM-SVM) based speaker modeling have been used. Experimental results, using, ARADIGITS data-base, show the efficiency of the GMM-SVM based approach associated with feature vectors issued from LP Residual Signal.
Didier Georges - One of the best experts on this subject based on the ideXlab platform.
-
Robust Fault Tolerant Control for Descriptor Systems
IEEE Transactions on Automatic Control, 2004Co-Authors: Benoît Marx, Damien Koenig, Didier GeorgesAbstract:A new architecture for fault tolerant controllers is proposed for the generic class of descriptor systems. It is based on coprime factorization of nonproper systems and on the Youla parameterization of stabilizing controllers. Noticing that the Youla controllers include a so called Residual Signal, fault tolerant control is achieved. Nominal control and robust fault tolerance are addressed separately. Moreover, fault tolerant control can be improved with a scheme integrating fault diagnosis. The design of the diagnosis and fault tolerant control filters reduce to a standard H-control problem of usual state-space system.
Hong Wang - One of the best experts on this subject based on the ideXlab platform.
-
Robust descriptor observer-based fault detection for stochastic distributions using output probability density functions
2004Co-Authors: Hong WangAbstract:The problem of robust observer-based fault detection is investigated for systems with bounded stochastic distributions. By constructing an auxiliary augmented stochastic descriptor system, a proportional and derivative descriptor observer is developed to solve the fault detection problem, where the system input and the output probability density function are used in this observer design. The derivative gain is chosen to attenuate the output uncertainties, and the free parameters of the proportional gain are then selected to generate an optimally robust Residual Signal for fault detection. It is shown that this Residual Signal can be made insensitive to the model uncertainties, input disturbances and output noises, but sensitive to system faults.
-
A Heuristic Approach to Fault Tolerant Control of Unknown Nonlinear Systems Using Neural Networks
IFAC Proceedings Volumes, 1998Co-Authors: J. R. Noriega, Hong WangAbstract:Abstract A heuristic approach for the problem of fault tolerant control of unknown nonlinear systems is discussed in this paper. The method uses a heuristically determined feedback function for the compensation of the system response and a neural network model. It is assumed that changes in the system parameters can lead to an increase in the magnitude of the Residual Signal. In this case, the Residual is formulated as the difference of the system output and the neural network model output. A fault is normally associated with an unexpected increase in the Residual Signal. Therefore, the Residual is constantly monitored to detect the fault and to start the compensation algorithm. In addition, the Residual is fed back through a compensation block. Thus fault tolerance is achieved by adjusting the control Signal of the failed system such that the Residual Signal approaches its original faultless magnitude. The mathematical form of the compensation block is defined by a combination of experimentation and heuristic knowledge of the response of the system.
-
On the Generation of an Optimally Robust Residual Signal for Systems with Structured Model Uncertainty
1992 American Control Conference, 1992Co-Authors: Samantha Daley, Hong WangAbstract:In this paper, the recently developed. parametric design for observer gain matrices is used to formulate an optimally robust Residual Signal for fault diagnosis in systems with structured model uncertainty. Using the available free parameters inside observer gain matrix and with a proper choice of performance function, it is shown that a Residual Signal can be obtained which is insensitive to the model uncertainty and sensitive to the faults of the system. A simulation for a fifth order system is carried out and good results are obtained.
Benoît Marx - One of the best experts on this subject based on the ideXlab platform.
-
Robust Fault Tolerant Control for Descriptor Systems
IEEE Transactions on Automatic Control, 2004Co-Authors: Benoît Marx, Damien Koenig, Didier GeorgesAbstract:A new architecture for fault tolerant controllers is proposed for the generic class of descriptor systems. It is based on coprime factorization of nonproper systems and on the Youla parameterization of stabilizing controllers. Noticing that the Youla controllers include a so called Residual Signal, fault tolerant control is achieved. Nominal control and robust fault tolerance are addressed separately. Moreover, fault tolerant control can be improved with a scheme integrating fault diagnosis. The design of the diagnosis and fault tolerant control filters reduce to a standard H-control problem of usual state-space system.
