• ETRI Journal
• /
• v.32 no.5
• /
• pp.795-800
• /
• 2010

Embedding a large vocabulary speech recognition system in mobile devices requires a reduced acoustic model obtained by eliminating redundant model parameters. In conventional optimization methods based on the minimum description length (MDL) criterion, a binary Gaussian tree is built at each state of a hidden Markov model by iteratively finding and merging similar mixture components. An optimal subset of the tree nodes is then selected to generate a downsized acoustic model. To obtain a better binary Gaussian tree by improving the process of finding the most similar Gaussian components, this paper proposes a new distance measure that exploits the difference in likelihood values for cases before and after two components are combined. The mixture weight of Gaussian components is also introduced in the component merging step. Experimental results show that the proposed method outperforms MDL-based optimization using either a Kullback-Leibler (KL) divergence or weighted KL divergence measure. The proposed method could also reduce the acoustic model size by 50% with less than a 1.5% increase in error rate compared to a baseline system.

• Nuclear Engineering and Technology
• /
• v.56 no.2
• /
• pp.715-727
• /
• 2024

During fast neutron imaging, besides the dark current noise and readout noise of the CCD camera, the main noise in fast neutron imaging comes from high-energy gamma rays generated by neutron nuclear reactions in and around the experimental setup. These high-energy gamma rays result in the presence of high-density gamma white spots (GWS) in the fast neutron image. Due to the microscopic quantum characteristics of the neutron beam itself and environmental scattering effects, fast neutron images typically exhibit a mixture of Gaussian noise. Existing denoising methods in neutron images are difficult to handle when dealing with a mixture of GWS and Gaussian noise. Herein we put forward a deep learning approach based on the Swin Transformer UNet (SUNet) model to remove high-density GWS-Gaussian mixture noise from fast neutron images. The improved denoising model utilizes a customized loss function for training, which combines perceptual loss and mean squared error loss to avoid grid-like artifacts caused by using a single perceptual loss. To address the high cost of acquiring real fast neutron images, this study introduces Monte Carlo method to simulate noise data with GWS characteristics by computing the interaction between gamma rays and sensors based on the principle of GWS generation. Ultimately, the experimental scenarios involving simulated neutron noise images and real fast neutron images demonstrate that the proposed method not only improves the quality and signal-to-noise ratio of fast neutron images but also preserves the details of the original images during denoising.

• Phonetics and Speech Sciences
• /
• v.7 no.3
• /
• pp.131-138
• /
• 2015

In this paper, we propose L1-norm regularization for state vector adaptation of subspace Gaussian mixture model (SGMM). When you design a speaker adaptation system with GMM-HMM acoustic model, MAP is the most typical technique to be considered. However, in MAP adaptation procedure, large number of parameters should be updated simultaneously. We can adopt sparse adaptation such as L1-norm regularization or sparse MAP to cope with that, but the performance of sparse adaptation is not good as MAP adaptation. However, SGMM does not suffer a lot from sparse adaptation as GMM-HMM because each Gaussian mean vector in SGMM is defined as a weighted sum of basis vectors, which is much robust to the fluctuation of parameters. Since there are only a few adaptation techniques appropriate for SGMM, our proposed method could be powerful especially when the number of adaptation data is limited. Experimental results show that error reduction rate of the proposed method is better than the result of MAP adaptation of SGMM, even with small adaptation data.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.17 no.6
• /
• pp.725-730
• /
• 2007

Gray-level histogram-based threshold selection methods such as Otsu's method, Huang and Wang's method, and etc. have been widely used for the threshold selection in image processing. They are simple and effective, but take too much time to determine the optimal multilevel threshold values as the number of thresholds are increased. In this paper, we measure correlation between gray-levels by using the Gaussian function and define a Gaussian-type finite mixture distribution which is combination of the Gaussian distribution function with the gray-level histogram, and propose a fast and effective threshold selection method using it. We show the effectiveness of the proposed through experimental results applied it to three images and the efficiency though comparison of the computational complexity of the proposed with that of Otsu's method.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.13 no.5
• /
• pp.512-519
• /
• 2003

In this paper, we propose a neuro-fuzzy modeling to improve the performance using the hierarchical clustering and Gaussian Mixture Model(GMM). The hierarchical clustering algorithm has a property of producing unique parameters for the given data because it does not use the object function to perform the clustering. After optimizing the obtained parameters using the GMM, we apply them as initial parameters for Adaptive Network-based Fuzzy Inference System. Here, the number of fuzzy rules becomes to the cluster numbers. From this, we can improve the performance index and reduce the number of rules simultaneously. The proposed method is verified by applying to a neuro-fuzzy modeling for Box-Jenkins s gas furnace data and Sugeno's nonlinear system, which yields better results than previous oiles.

• Journal of the Korean Society for Aeronautical & Space Sciences
• /
• v.47 no.4
• /
• pp.274-282
• /
• 2019

Vision-based navigation of unmaned aerial vehicle is a significant technology that can reinforce the vulnerability of the widely used GPS/INS integrated navigation system. However, the existing image matching algorithms are not suitable for matching the aerial image with the database. For the reason, this paper proposes particle filters using Gaussian mixture models to deal with matching between aerial image and database for vision-based navigation. The particle filters estimate the position of the aircraft by comparing the correspondences of aerial image and database under the assumption of Gaussian mixture model. Finally, Monte Carlo simulation is presented to demonstrate performance of the proposed method.

• Journal of the Korea Institute of Military Science and Technology
• /
• v.22 no.1
• /
• pp.27-34
• /
• 2019

This paper presents a chamber pressure model development of K2 rifle by applying Gaussian mixture model. In order to materialize a real recoil force of a virtual reality shooting rifle in military combat training, the chamber pressure which is one of major components of the recoil force needs to be investigated and modeled. Over 200,000 data of the chamber pressure were collected by implementing live fire experiments with both K100 and M193 of 5.56 mm bullets. Gaussian mixture method was also applied to create a mathematical model that satisfies nonlinear, asymmetry, and deviations of the chamber pressure which is caused by irregular characteristics of propellant combustion. In addition, Polynomial and Fourier Regression were used for comparison of results, and the sum of squared errors, the coefficient of determination and root-mean-square errors were analyzed for performance measurement.

• The KIPS Transactions:PartB
• /
• v.12B no.4 s.100
• /
• pp.437-442
• /
• 2005

For the purpose of speaker normalization in speaker independent speech recognition systems, experiments are conducted on a method based on Gaussian mixture model(GMM). The method, which is an improvement of the previous study based on vector quantizer, consists of modeling the probability distribution of canonical feature vectors by a GMM with an appropriate number of clusters, and of estimating the warp factor of a test speaker by making use of the obtained probabilistic model. The purpose of this study is twofold: improving the existing ML based methods, and comparing the performance of what is called 'soft decision' method with that of the previous study based on vector quantizer. The effectiveness of the proposed method is investigated by recognition experiments on the TIMIT corpus. The experimental results showed that a little improvement could be obtained tv adjusting the number of clusters in GMM appropriately.

• Journal of Korea Multimedia Society
• /
• v.7 no.11
• /
• pp.1531-1539
• /
• 2004

In this paper, we propose an efficient method for locating human lips. Based on Point Distribution Model and Principle Component Analysis, a lip shape model is built. Lip boundary model is represented based on the concatenated gray level distribution model. We calculate the distribution of shape parameters using Gaussian mixture. The problem to locate lip is simplified as the minimization problem of matching object function. The Down Hill Simplex Algorithm is used for the minimization with Gaussian Mixture for setting initial condition and refining estimate of lip shape parameter, which can refrain iteration from converging to local minima. The experiments have been performed for many images, and show very encouraging result.

• International Journal of Fuzzy Logic and Intelligent Systems
• /
• v.13 no.1
• /
• pp.59-66
• /
• 2013

In this paper, we propose a probabilistic framework for model-based clustering of direction of arrival (DOA) data to obtain stable sound source localization (SSL) estimates. Model-based clustering has been shown capable of handling highly overlapped and noisy datasets, such as those involved in DOA detection. Although the Gaussian mixture model is commonly used for model-based clustering, we propose use of the von Mises mixture model as more befitting circular DOA data than a Gaussian distribution. The EM framework for the von Mises mixture model in a unit hyper sphere is degenerated for the 2D case and used as such in the proposed method. We also use a histogram of the dataset to initialize the number of clusters and the initial values of parameters, thereby saving calculation time and improving the efficiency. Experiments using simulated and real-world datasets demonstrate the performance of the proposed method.