The Experts below are selected from a list of 29175 Experts worldwide ranked by ideXlab platform
Qiang Huo - One of the best experts on this subject based on the ideXlab platform.
-
a maximum likelihood approach to unsupervised online adaptation of Stochastic Vector mapping function for robust speech recognition
International Conference on Acoustics Speech and Signal Processing, 2007Co-Authors: Donglai Zhu, Qiang HuoAbstract:In the past several years, we've been studying feature transformation approaches for robust automatic speech recognition (ASR) based on the concept of Stochastic Vector mapping (SVM) to compensating for possible "distortions" caused by factors irrelevant to phonetic classification in both training and recognition stages. Although we have demonstrated the usefulness of the SVM-based approaches for several robust ASR applications where diversified yet representative training data are available, the performance improvement of SVM-based approaches is less significant when there is a severe mismatch between training and testing conditions. In this paper, we present a maximum likelihood approach to unsupervised online adaptation (OLA) of SVM function parameters on an utterance-by-utterance basis for achieving further performance improvement. Its effectiveness is confirmed by evaluation experiments on Finnish AuroraS database.
-
ICASSP (4) - A Maximum Likelihood Approach to Unsupervised Online Adaptation of Stochastic Vector Mapping Function for Robust Speech Recognition
2007 IEEE International Conference on Acoustics Speech and Signal Processing - ICASSP '07, 2007Co-Authors: Donglai Zhu, Qiang HuoAbstract:In the past several years, we've been studying feature transformation approaches for robust automatic speech recognition (ASR) based on the concept of Stochastic Vector mapping (SVM) to compensating for possible "distortions" caused by factors irrelevant to phonetic classification in both training and recognition stages. Although we have demonstrated the usefulness of the SVM-based approaches for several robust ASR applications where diversified yet representative training data are available, the performance improvement of SVM-based approaches is less significant when there is a severe mismatch between training and testing conditions. In this paper, we present a maximum likelihood approach to unsupervised online adaptation (OLA) of SVM function parameters on an utterance-by-utterance basis for achieving further performance improvement. Its effectiveness is confirmed by evaluation experiments on Finnish AuroraS database.
-
an environment compensated maximum likelihood training approach based on Stochastic Vector mapping speech recognition applications
International Conference on Acoustics Speech and Signal Processing, 2005Co-Authors: Qiang Huo, Donglai ZhuAbstract:Several recent approaches for robust speech recognition are developed based on the concept of Stochastic Vector mapping (SVM) that perform a frame-dependent bias removal to compensate for environmental variabilities in both training and recognition stages. Some of them require stereo recordings of both clean and noisy speech for the estimation of SVM function parameters. In this paper, we present a detailed formulation of a maximum likelihood training approach for the joint design of SVM function parameters and HMM parameters of a speech recognizer that does not rely on the availability of stereo training data. Its learning behavior and effectiveness is demonstrated by using the experimental results on the Aurora3 Finnish connected digits database recorded by using both close-talking and hands-free microphones in cars.
-
ICASSP (1) - An environment compensated maximum likelihood training approach based on Stochastic Vector mapping [speech recognition applications]
Proceedings. (ICASSP '05). IEEE International Conference on Acoustics Speech and Signal Processing 2005., 1Co-Authors: Qiang Huo, Donglai ZhuAbstract:Several recent approaches for robust speech recognition are developed based on the concept of Stochastic Vector mapping (SVM) that perform a frame-dependent bias removal to compensate for environmental variabilities in both training and recognition stages. Some of them require stereo recordings of both clean and noisy speech for the estimation of SVM function parameters. In this paper, we present a detailed formulation of a maximum likelihood training approach for the joint design of SVM function parameters and HMM parameters of a speech recognizer that does not rely on the availability of stereo training data. Its learning behavior and effectiveness is demonstrated by using the experimental results on the Aurora3 Finnish connected digits database recorded by using both close-talking and hands-free microphones in cars.
Donglai Zhu - One of the best experts on this subject based on the ideXlab platform.
-
a maximum likelihood approach to unsupervised online adaptation of Stochastic Vector mapping function for robust speech recognition
International Conference on Acoustics Speech and Signal Processing, 2007Co-Authors: Donglai Zhu, Qiang HuoAbstract:In the past several years, we've been studying feature transformation approaches for robust automatic speech recognition (ASR) based on the concept of Stochastic Vector mapping (SVM) to compensating for possible "distortions" caused by factors irrelevant to phonetic classification in both training and recognition stages. Although we have demonstrated the usefulness of the SVM-based approaches for several robust ASR applications where diversified yet representative training data are available, the performance improvement of SVM-based approaches is less significant when there is a severe mismatch between training and testing conditions. In this paper, we present a maximum likelihood approach to unsupervised online adaptation (OLA) of SVM function parameters on an utterance-by-utterance basis for achieving further performance improvement. Its effectiveness is confirmed by evaluation experiments on Finnish AuroraS database.
-
ICASSP (4) - A Maximum Likelihood Approach to Unsupervised Online Adaptation of Stochastic Vector Mapping Function for Robust Speech Recognition
2007 IEEE International Conference on Acoustics Speech and Signal Processing - ICASSP '07, 2007Co-Authors: Donglai Zhu, Qiang HuoAbstract:In the past several years, we've been studying feature transformation approaches for robust automatic speech recognition (ASR) based on the concept of Stochastic Vector mapping (SVM) to compensating for possible "distortions" caused by factors irrelevant to phonetic classification in both training and recognition stages. Although we have demonstrated the usefulness of the SVM-based approaches for several robust ASR applications where diversified yet representative training data are available, the performance improvement of SVM-based approaches is less significant when there is a severe mismatch between training and testing conditions. In this paper, we present a maximum likelihood approach to unsupervised online adaptation (OLA) of SVM function parameters on an utterance-by-utterance basis for achieving further performance improvement. Its effectiveness is confirmed by evaluation experiments on Finnish AuroraS database.
-
an environment compensated maximum likelihood training approach based on Stochastic Vector mapping speech recognition applications
International Conference on Acoustics Speech and Signal Processing, 2005Co-Authors: Qiang Huo, Donglai ZhuAbstract:Several recent approaches for robust speech recognition are developed based on the concept of Stochastic Vector mapping (SVM) that perform a frame-dependent bias removal to compensate for environmental variabilities in both training and recognition stages. Some of them require stereo recordings of both clean and noisy speech for the estimation of SVM function parameters. In this paper, we present a detailed formulation of a maximum likelihood training approach for the joint design of SVM function parameters and HMM parameters of a speech recognizer that does not rely on the availability of stereo training data. Its learning behavior and effectiveness is demonstrated by using the experimental results on the Aurora3 Finnish connected digits database recorded by using both close-talking and hands-free microphones in cars.
-
ICASSP (1) - An environment compensated maximum likelihood training approach based on Stochastic Vector mapping [speech recognition applications]
Proceedings. (ICASSP '05). IEEE International Conference on Acoustics Speech and Signal Processing 2005., 1Co-Authors: Qiang Huo, Donglai ZhuAbstract:Several recent approaches for robust speech recognition are developed based on the concept of Stochastic Vector mapping (SVM) that perform a frame-dependent bias removal to compensate for environmental variabilities in both training and recognition stages. Some of them require stereo recordings of both clean and noisy speech for the estimation of SVM function parameters. In this paper, we present a detailed formulation of a maximum likelihood training approach for the joint design of SVM function parameters and HMM parameters of a speech recognizer that does not rely on the availability of stereo training data. Its learning behavior and effectiveness is demonstrated by using the experimental results on the Aurora3 Finnish connected digits database recorded by using both close-talking and hands-free microphones in cars.
David D. Yao - One of the best experts on this subject based on the ideXlab platform.
-
Stochastic Vector difference equations with stationary coefficients
Journal of Applied Probability, 1995Co-Authors: Paul Glasserman, David D. YaoAbstract:We give a unified presentation of stability results for Stochastic Vector difference equations Y n+1 = A n ? Y n ? B n based on various choices of binary operations ? and ?, assuming that {(A n , B n ), n ≥ 0} are stationary and ergodic. In the scalar case, under standard addition and multiplication, the key condition for stability is E[log |A 0 |]<0. In the generalizations, the condition takes the form γ <0, where γ is the limit of a subadditive process associated with {A(n), n≥0}. Under this and mild additional conditions, the process { Y n , n ≥ 0} has a unique finite stationary distribution to which it converges from all initial conditions The variants of standard matrix algebra we consider replace the operations + and x with (max,+), (max,x), (min,+), or (min,x). In each case, the appropriate stability condition parallels that for the standard recursions, involving certain subadditive limits. Since these limits are difficult to evaluate, we provide bounds, thus giving alternative, computable conditions for stability.
-
Stochastic Vector difference equations with stationary coefficients
Journal of Applied Probability, 1995Co-Authors: Paul Glasserman, David D. YaoAbstract:We give a unified presentation of stability results for Stochastic Vector difference equations Y n+1 = A n ? Y n ? B n based on various choices of binary operations ? and ?, assuming that {(A n , B n ), n ≥ 0} are stationary and ergodic. In the scalar case, under standard addition and multiplication, the key condition for stability is E[log |A 0 |]
Paul Glasserman - One of the best experts on this subject based on the ideXlab platform.
-
Stochastic Vector difference equations with stationary coefficients
Journal of Applied Probability, 1995Co-Authors: Paul Glasserman, David D. YaoAbstract:We give a unified presentation of stability results for Stochastic Vector difference equations Y n+1 = A n ? Y n ? B n based on various choices of binary operations ? and ?, assuming that {(A n , B n ), n ≥ 0} are stationary and ergodic. In the scalar case, under standard addition and multiplication, the key condition for stability is E[log |A 0 |]<0. In the generalizations, the condition takes the form γ <0, where γ is the limit of a subadditive process associated with {A(n), n≥0}. Under this and mild additional conditions, the process { Y n , n ≥ 0} has a unique finite stationary distribution to which it converges from all initial conditions The variants of standard matrix algebra we consider replace the operations + and x with (max,+), (max,x), (min,+), or (min,x). In each case, the appropriate stability condition parallels that for the standard recursions, involving certain subadditive limits. Since these limits are difficult to evaluate, we provide bounds, thus giving alternative, computable conditions for stability.
-
Stochastic Vector difference equations with stationary coefficients
Journal of Applied Probability, 1995Co-Authors: Paul Glasserman, David D. YaoAbstract:We give a unified presentation of stability results for Stochastic Vector difference equations Y n+1 = A n ? Y n ? B n based on various choices of binary operations ? and ?, assuming that {(A n , B n ), n ≥ 0} are stationary and ergodic. In the scalar case, under standard addition and multiplication, the key condition for stability is E[log |A 0 |]
Luis Torres - One of the best experts on this subject based on the ideXlab platform.
-
Efficient coding of homogeneous textures using Stochastic Vector quantisation and linear prediction
IEE Proceedings - Vision Image and Signal Processing, 1999Co-Authors: D. Gimeno Gost, Luis TorresAbstract:Vector quantisation (VQ) has been extensively used as an effective image coding technique. One of the most important steps in the whole process is the design of the codebook. The codebook is generally designed using the LBG algorithm which uses a large training set of empirical data that is statistically representative of the images to be encoded. The LBG algorithm, although quite effective for practical applications, is computationally very expensive and the resulting codebook has to be recalculated each time the type of image to be encoded changes. Stochastic Vector quantisation (SVQ) provides an alternative way for the generation of the codebook. In SVQ, a model for the image is computed first, and then the codewords are generated according to this model and not according to some specific training sequence. The SVQ approach presents good coding performance for moderate compression ratios and different type of images. On the other hand, in the context of synthetic and natural hybrid coding (SNHC), there is always need for techniques which may provide very high compression and high quality for homogeneous textures. A new Stochastic Vector quantisation approach using linear prediction which is able to provide very high compression ratios with graceful degradation for homogeneous textures is presented. Owing to the specific construction of the method, there is no block effect in the synthetised image. Results, implementation details, generation of the bit stream and comparisons with the verification model of MPEG-4 are presented which prove the validity of the approach. The technique has been proposed as a still image coding technique in the SNHC standardisation group of MPEG.
-
Stochastic Vector quantization of images
Signal Processing, 1997Co-Authors: Luis Torres, Josep R. Casas, E. AriasAbstract:One of the most important steps in the Vector quantization of images is the design of the codebook. The codebook is generally designed using the LBG algorithm, that is in essence a clustering algorithm which uses a large training set of empirical data that is statistically representative of the image to be quantized. The LBG algorithm, although quite effective for practical applications, is computationally very expensive and the resulting codebook has to be recalculated each time the type of image to be encoded changes. One alternative to the generation of the codebook, called Stochastic Vector quantization, is presented in this paper. Stochastic Vector quantization (SVQ) is based on the generation of the codebook according to some previous model defined for the image to be encoded. The well-known AR model has been used to model the image in the current implementations of the technique, and has shown good performance in the overall scheme. To show the merit of the technique in different contexts, Stochastic Vector quantization is discussed and applied to both pixel-based and segmentation-based image coding schemes.
-
ICASSP (5) - Improvements on Stochastic Vector quantization of images
IEEE International Conference on Acoustics Speech and Signal Processing, 1993Co-Authors: Luis Torres, J. SalillasAbstract:A novel nonadaptive fixed-rate Vector quantizer encoding scheme is presented, and preliminary results are shown. The design of the codebook has been based on a Stochastic approach in order to match a previously defined model for the image to be encoded. Following this approach, the generation of the codebook is made extremely simple in terms of computational load. Good visual results are shown in the range of 0.5-0.8 bit/pixel. Much better performance is expected for adaptive schemes. >
-
ICASSP - Stochastic Vector quantization of images
[Proceedings] ICASSP-92: 1992 IEEE International Conference on Acoustics Speech and Signal Processing, 1992Co-Authors: Luis Torres, E. AriasAbstract:The Vector quantization scheme has proven to be very effective in image coding. One of the most important steps in the whole process is the design of the codebook. The codebook is generally designed using the Linde-Buzo-Gray algorithm, which is in essence a clustering algorithm that uses a large training set of empirical data that are statistically representative of the image to be quantized. The problem addressed is the Stochastic generation of the codebook. The approach is to model the codebook according to some previous model defined for the image to be encoded and then to generate the training set according to the same model and not according to some specific data sequence. The model used is the well-known autoregressive model. Good visual results are shown in the range of 0.5-0.8 b/pixel. >
-
ICIP (1) - A new approach to texture coding using Stochastic Vector quantization
Proceedings of 1st International Conference on Image Processing, 1Co-Authors: D Gimeno, Luis Torres, Josep R. CasasAbstract:A new method for texture coding which combines 2-D linear prediction and Stochastic Vector quantization is presented in this paper. To encode a texture, a linear predictor is computed first. Next, a codebook following the prediction error model is generated and the prediction error is encoded with VQ, using an algorithm which takes into account the pixels surrounding the block being encoded. In the decoder, the error image is decoded first and then filtered as a whole, using the prediction filter. Hence, correlation between pixels is not lost from one block to another and a good reproduction quality can be achieved. >
