• 대한음성학회지:말소리
• /
• 제60호
• /
• pp.165-180
• /
• 2006

In this paper, we present several confidence measures (CM) for speech recognition systems to evaluate the reliability of recognition results. We propose heuristic CMs such as mean log-likelihood score, N-best word log-likelihood ratio, likelihood sequence fluctuation and likelihood ratio testing(LRT)-based CMs using several types of anti-models. Furthermore, we propose new algorithms to add weighting terms on phone-level log-likelihood ratio to merge word-level log-likelihood ratios. These weighting terms are computed from the distance between acoustic models and knowledge-based phoneme classifications. LRT-based CMs show better performance than heuristic CMs excessively, and LRT-based CMs using phonetic information show that the relative reduction in equal error rate ranges between $8{\sim}13%$ compared to the baseline LRT-based CMs. We use the support vector machine to fuse several CMs and improve the performance of utterance verification. From our experiments, we know that selection of CMs with low correlation is more effective than CMs with high correlation.

• 디지털융복합연구
• /
• 제16권11호
• /
• pp.335-340
• /
• 2018

정확한 인식률을 보이고 있는 상업적인 음성인식 시스템은 화자종속 고립데이터로부터 학습 모델을 사용한다. 그러나 잡음 환경에서 데이터양에 따라 음성인식의 성능이 저하되는 문제점이 있다. 본 논문에서는 가우시안 분포에서 Maximum Log Likelihood를 이용한 벡터 양자화 기반 음성 인식 성능 향상을 제안한다. 제안하는 방법은 음성에 대한 특징을 가지고 벡터 양자화와 Maximum Log Likelihood 음성 특징 추출 방법을 이용하여 유사 음성에 대한 음성 인식의 정확성을 높이는 최적 학습 모델 구성 방법이다. 이를 위해 HMM을 기반으로 음성 특징을 추출하는 방법을 사용한다. 제안하는 방법을 사용하여 기존 시스템에서 생성되어 사용되는 음성 모델에 대한 부정확한 음성 모델에 대한 정확성을 향상시킬 수 있으므로 음성 인식에 강인한 모델을 구성할 수 있다. 제안하는 방법은 음성 인식 시스템에서 향상된 인식의 정확도를 보인다.

• 한국통신학회논문지
• /
• 제31권2C호
• /
• pp.117-122
• /
• 2006

본 논문에서는 로그 최우비(log likelihood ratio, LLR)를 이용하여 Gray 부호화된 PAM신호를 위한 비트 연판정 계산식을 제안한다. 이 계산식은 Gray 매핑 특성을 이용하여 Max-Log-MAP 알고리듬에서 필요한 max0/min0 함수를 사용하지 않고 산술 연산만을 사용하기 때문에 구현이 간단하다. 제안된 식의 결과는 기존 Max-Log-MAP 알고리듬의 결과와 일치한다. 또한, 식에 사용되는 인자들은 송수신 시스템이 서로 공유하는 정보와 수신된 심벌 값만을 이용하여 계산한다. 따라서 본 논문에서 제안된 알고리듬은 일반적으로 많이 사용하는 이진 반복 복호기 등과 함께 실제 응용 설계에 적용이 가능하며, 특히 실제 설계에 적용되는 기존의 여러 가지 알고리듬에 비하여 구조가 유연하고 효율적이며 정확한 비트별 LLR을 제공하는 효율적인 방법 중의 하나이다.

• 제어로봇시스템학회:학술대회논문집
• /
• 제어로봇시스템학회 1992년도 한국자동제어학술회의논문집(국제학술편); KOEX, Seoul; 19-21 Oct. 1992
• /
• pp.272-275
• /
• 1992

A novel fault diagnosis method based on likelihood decomposition is proposed for linear stochastic systems described by autoregressive (AR) model. Assuming that at some time instant .tau. the fault of one of the following two types is occurs: innovation fault (actuator fault); and observation fault (sensor fault), the log-likelihood function is decomposed into two components based on the observations before and after .tau., respectively, Then, the type of the fault is determined by comparing the log-likelihoods corresponding two types of faults. Numerical examples demonstrate the usefulness of the proposed diagnosis method.

• Journal of the Korean Data and Information Science Society
• /
• 제17권3호
• /
• pp.963-972
• /
• 2006

This paper focuses on the likelihood based confidence intervals for two inverse gaussian distributions when the parameter of interest is common scale parameter. Confidence intervals based on signed loglikelihood ratio statistic and modified signed loglikelihood ratio statistics will be compared in small sample through an illustrative simulation study.

• Journal of the Korean Data and Information Science Society
• /
• 제17권3호
• /
• pp.919-925
• /
• 2006

Maximum likelihood estimate(MLE) is obtained from the partial log-likelihood function for the cell probabilities of two way incomplete contingency tables proposed by Chen and Fienberg(1974). The partial log-likelihood function is modified by adding lagrangian multiplier that constraints can be incorporated with. Variances of MLE estimators of population proportions are derived from the matrix of second derivatives of the loglikelihood with respect to cell probabilities. Simulation results, when data are missing at random, reveal that Complete-case(CC) analysis produces biased estimates of joint probabilities under MAR and less efficient than either MLE or MI. MLE and MI provides consistent results under either the MAR situation. MLE provides more efficient estimates of population proportions than either multiple imputation(MI) based on data augmentation or complete case analysis. The standard errors of MLE from the proposed method using lagrangian multiplier are valid and have less variation than the standard errors from MI and CC.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제7권4호
• /
• pp.767-783
• /
• 2013

This paper aims at designing a selective demodulation scheme based on Log-likelihood Ratio threshold (SDLT) instead of the conventional adaptive demodulation (ADM) scheme, by using rateless codes. The major difference is that the Log-likelihood ratio (LLR) threshold is identified as a key factor to control the demodulation rate, while the ADM uses decision region set (DRS) to adjust the bit rate. In the 16-QAM SDLT scheme, we deduce the decision regions over an additive white Gaussian channel, corresponding to the variation of LLR threshold and channel states. We also derived the equations to calculate demodulation rate and bit error rate (BER), which could be proven by simulation results. We present an adaptation strategy for SDLT, and compare it with ADM and adaptive modulation (AM). The simulation results show that our scheme not only significantly outperforms the ADM in terms of BER, but also achieves a performance as good as the AM scheme. Moreover, the proposed scheme can support much more rate patterns over a wide range of channel states.

• Communications for Statistical Applications and Methods
• /
• 제11권3호
• /
• pp.519-527
• /
• 2004

Relationship between the partial likelihood ratio statistics for logisitic models and the partial goodness-of-fit statistics for corresponding log-linear models is discussed. This paper shows how definitions of suppression in logistic model can be adapted for log-linear model and how they are related to confounding in terms of collapsibility for categorical data. Several $2{times}2{times}2$ contingency tables are illustrated.

• 한국통신학회논문지
• /
• 제41권12호
• /
• pp.1736-1738
• /
• 2016

고차변조와 고효율 채널부호를 사용하는 통신시스템에서는 지수함수의 합 로그 연산을 수반하는 다량의 대수우도비 산출이 필수적이며 이의 근사화 기법에 따라 링크 성능이 좌우된다. 본 논문에서는 신규 근사화 기법을 복조기와 복호기에 결합하여 적용하고 복잡도를 분석하며 모의실험을 통해 도출한 결합 성능을 분석한다.

• Communications for Statistical Applications and Methods
• /
• 제25권6호
• /
• pp.633-645
• /
• 2018

Gaussian error distributions are a common choice in traditional regression models for the maximum likelihood (ML) method. However, this distributional assumption is often suspicious especially when the error distribution is skewed or has heavy tails. In both cases, the ML method under normality could break down or lose efficiency. In this paper, we consider the log-concave and Gaussian scale mixture distributions for error distributions. For the log-concave errors, we propose to use a smoothed maximum likelihood estimator for stable and faster computation. Based on this, we perform comparative simulation studies to see the performance of coefficient estimates under normal, Gaussian scale mixture, and log-concave errors. In addition, we also consider real data analysis using Stack loss plant data and Korean labor and income panel data.