• 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.

• Journal of Korean Institute of Industrial Engineers
• /
• v.24 no.1
• /
• pp.105-113
• /
• 1998

This paper deals with a production line which consists of two production stages that are separated by a finite storage buffer. The inventory level in the storage buffer controls the production rate of the preceding stage. That is, the production rate becomes high (low) when the buffer inventory is low (high). We analyze the system characteristics utilizing the Markov process theory and then find an optimal control policy which maximizes a given system profit function. Also, a sensitivity analysis is made to examine the effects of various system parameters on the system performances.

• Journal of the Korean Society for Industrial and Applied Mathematics
• /
• v.19 no.1
• /
• pp.23-45
• /
• 2015

We consider counterparty risk in CDS rates. To do so, we use a multivariate jump diffusion process for obligors' default intensity, where jumps (i.e. magnitude of contribution of primary events to default intensities) occur simultaneously and their sizes are dependent. For these simultaneous jumps and their sizes, a homogeneous Poisson process. We apply copula-dependent default intensities of multivariate Cox process to derive the joint Laplace transform that provides us with joint survival/default probability and other relevant joint probabilities. For that purpose, the piecewise deterministic Markov process (PDMP) theory developed in [7] and the martingale methodology in [6] are used. We compute survival/default probability using three copulas, which are Farlie-Gumbel-Morgenstern (FGM), Gaussian and Student-t copulas, with exponential marginal distributions. We then apply the results to calculate CDS rates assuming deterministic rate of interest and recovery rate. We also conduct sensitivity analysis for the CDS rates by changing the relevant parameters and provide their figures.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.06e
• /
• pp.373-376
• /
• 1998

통신 수단의 발달로 휴대단말기의 사용이 증가하고 있으며, 이와 함께 휴대단말기에서의 음성인식에 대한 수요도 증가하고 있다. 휴대단말기의 경우 저 전송율을 가지는 음성 부호화기를 사용하게 되며, 이러한 저전송율의 음성 부호화기에서의 음성인식을 수행할 경우 인식 성능이 저하되는 현상을 보이게 된다. 본 논문에서는 이러한 문제를 해결하기 위하여 LSP 파라메터 기반의 거리척도에 관하여 비교 검토하였으며, 적은 훈련 데이터에서 사용 가능한 화자 종속 음성인식 방법으로 Dynamic Time Warping(DTW)과 변형된 Hidden Markov Model(HMM)에 관하여 검토하였다. QCELP 음성 부호화기에서 인식 어휘 당 2번의 훈련 데이터만을 이용한 화자종속 인식방법을 사용한 결과 95% 이상의 인식 성능을 얻을 수 있었다.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.7
• /
• pp.48-52
• /
• 2000

음성 인식에 많이 사용되는 HMM (hidden Markov model)을 개선하기 위하여 분절 특징을 사용한 분절 특징 HMM은 성능이 우수하다고 발표되었다. 그러나, 분절 길이가 증가하고 회귀 차수가 놓아질수록 분절 특징 HMM을 표현하는 매개 변수의 수도 같이 증가된다. 따라서, 본 연구에서는 상태에서 관측 가능한 분절의 분산을 분절 내의 모든 프레임에 대하여 공통적으로 표현하는 고정 분산 방법을 통하여 성능의 저하 없이 매개 변수의 수를 줄이도록 시도하였다. 실험 결과, 두 혼합 밀도인 경우 고정 분산을 이용한 분절 특징 HMM의 성능과 시변 분산을 이용한 성능의 차이가 거의 없어, 제안된 방법의 유효성을 입증하였다.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.18 no.1
• /
• pp.45-54
• /
• 1993

Monitoring processor in a circuit switched network is considered. Monitoring processor monitors communication links offers a grade of service in each link to controller. Such an information is useful for an effective maintenance of system. Two links with asymmetric system parameters and multi-symmetric links are respectively considered. Each links is to be an independent M /M/ 1/ 1/ type. Markov modeling technique is used to represent a model of monitoring processor with FCFS steering protocol. Performance measures considered are ratio of monitored jobs in each link, availability of minitoring processor and throughput of virtual processor in each link. The value of the performance meausres are compared with existing and simulation results.

• Journal of the Korean Operations Research and Management Science Society
• /
• v.18 no.3
• /
• pp.1-12
• /
• 1993

Several alternative manufacturing control strategies are under study in the literature. They are, specifically, push system, pull system, conwip system, and as a special case, infinite buffer system. We focus on modeling, comparison analysis and design of these systems. The event epoch sequences of each system are generated which also enable us to compare their performance. Then the stochastic monotonicity of these enent epoch sequences in several important design parameters are established through the structure of the generalized semi-Markov schemes on which they are based. Finally, we solve the stochastic optimization problem which minimizes these event epochs. Our results supplement the applicability of some previously known results in the literature.

• Communications for Statistical Applications and Methods
• /
• v.8 no.3
• /
• pp.777-786
• /
• 2001

The autoregressive models have been used to describe a wade variety of time series. Then the problem of determining the order in the times series model is very important in data analysis. We consider the Bayesian approach for finding the order of autoregressive(AR) error models using the latent variable which is motivated by Tanner and Wong(1987). The latent variables are combined with the coefficient parameters and the sequential steps are proposed to set up the prior of the latent variables. Markov chain Monte Carlo method(Gibbs sampler and Metropolis-Hasting algorithm) is used in order to overcome the difficulties of Bayesian computations. Three examples including AR(3) error model are presented to illustrate our proposed methodology.

• Journal of Institute of Control, Robotics and Systems
• /
• v.17 no.2
• /
• pp.139-144
• /
• 2011

We consider the stabilization problem for a class of networked control systems with random delays in the discrete-time domain. The controller-to-actuator and sensor-to-controller time-delays are modeled as two Markov chains, and the resulting closed-loop systems are Markovian jump nonlinear systems with two modes. The T-S (Takagi-Sugeno) fuzzy model is employed to represent a nonlinear system with Markovian jump parameters. The aim is to design a fuzzy controller such that the closed-loop Markovian jump fuzzy system is stochastically stable. The necessary and sufficient conditions on the existence of stabilizing fuzzy controllers are established in terms of LMIs (Linear Matrix Inequalities). It is shown that fuzzy controller gains are mode-dependent. Finally, a simulation example is presented to illustrate the effectiveness of the proposed design method.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.4E
• /
• pp.21-27
• /
• 1996

In this paper, we propose a discriminative training algorithm for the stochastic segment model (SSM) in continuous speech recognition. As the SSM is usually trained by maximum likelihood estimation (MLE), a discriminative training algorithm is required to improve the recognition performance. Since the SSM does not assume the conditional independence of observation sequence as is done in hidden Markov models (HMMs), the search space for decoding an unknown input utterance is increased considerably. To reduce the computational complexity and starch space amount in an iterative training algorithm for discriminative SSMs, a hybrid architecture of SSMs and HMMs is programming using HMMs. Given the segment boundaries, the parameters of the SSM are discriminatively trained by the minimum error classification criterion based on a generalized probabilistic descent (GPD) method. With the discriminative training of the SSM, the word error rate is reduced by 17% compared with the MLE-trained SSM in speaker-independent continuous speech recognition.