• Speech Sciences
• /
• v.5 no.1
• /
• pp.7-21
• /
• 1999

This paper proposes an improved training procedure in speech recognition based on the continuous density of the Hidden Markov Model (CDHMM). Of the three parameters (initial state distribution probability, state transition probability, output probability density function (p.d.f.) of state) governing the CDHMM model, we focus on the third parameter and propose an efficient algorithm that determines the p.d.f. of each state. It is known that the resulting CDHMM model converges to a local maximum point of parameter estimation via the iterative Expectation Maximization procedure. Specifically, we propose two independent algorithms that can be embedded in the segmental K -means training procedure by replacing relevant key steps; the adaptation of the number of mixture Gaussian p.d.f. and the initialization using the CDHMM parameters previously estimated. The proposed adaptation algorithm searches for the optimal number of mixture Gaussian humps to ensure that the p.d.f. is consistently re-estimated, enabling the model to converge toward the global maximum point. By applying an appropriate threshold value, which measures the amount of collective changes of weighted variances, the optimized number of mixture Gaussian branch is determined. The initialization algorithm essentially exploits the CDHMM parameters previously estimated and uses them as the basis for the current initial segmentation subroutine. It captures the trend of previous training history whereas the uniform segmentation decimates it. The recognition performance of the proposed adaptation procedures along with the suggested initialization is verified to be always better than that of existing training procedure using fixed number of mixture Gaussian p.d.f.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1998.06c
• /
• pp.71-74
• /
• 1998

HMM (hidden Markov model)을 이용한 음성인식은 현재 가장 널리 쓰여지고 있는 방법으로, 이 중 CDHMM (continuous observation density HMM)은 상태에서 관측심볼확률을 연속확률밀도를 사용하여 표현한다. 본 논문에서는 가우스 혼합밀도함수를 사용하는 CDHMM의 상태천이확률과, 관측심볼확률을 표현하기 위한 인자인 평균벡터, 공분산 행렬, 가지하중값을 유전자 알고리듬을 사용하여 최적화하는 방법을 제안하였다. 유전자 알고리듬은 매개변수 최적화문제에 대하여 자연의 진화원리를 모방한 알고리듬으로, 염색체 형태로 표현된 개체군 (population) 중에서 환경에 대한 적합도 (fitness)가 높은 개체가 높은 확률로 살아남아 재생 (reproduction)하게 되며, 교배 (crossover)와 돌연변이 (mutation) 연산 후에 다음 세대 개체군을 형성하게 되고, 이러한 과정을 반복하면서 최적의 개체를 구하게 된다. 본 논문에서는 상태천이확률, 평균벡터, 공분산행렬, 가지하중값을 부동소수점수 (floating point number)의 유전자형으로 표현하여 유전자 알고리듬을 수행하였다. 유전자 알고리듬은 복잡한 탐색공간에서 최적의 해를 찾는데 효과적으로 적용되었다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.30 no.9C
• /
• pp.907-912
• /
• 2005

An efficient method for calculation of observation probability in CDHMM(Continous Density Hidden Markov Model) is proposed in this paper. the proposed algorithm, called FCOP(Fast Computation of Observation Probability), approximate obsewation probabilities in CDHMM by eliminating insignificant PDFs(Probability Density Functions) and reduces the computational load. When applied to a speech recognition system, the proposed FCOP algorithm can reduce the instruction cycles by $20\%-30\%$ and can also increase the recognition speed about $30\%$ while minimizing the loss in its recognition rate. When implemented on a practical cellular phone, the FCOP algorithm can increase its recognition speed about $30\%$ while suffering $0.2\%$ loss in recognition rate.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.39 no.2
• /
• pp.116-127
• /
• 2002

A new approach to improve the speaker adaptation algorithm by means of the variable number of observation density functions for CDHMM speech recognizer has been proposed. The proposed method uses the observation density function with more than one mixture in each state to represent speech characteristics in detail. The number of mixtures in each state is determined by the number of frames and the determinant of the variance, respectively. The each MAP Parameter is extracted in every mixture determined by these two methods. In addition, the state segmentation method requiring speaker adaptation can segment the adapting speech more Precisely by using speaker-independent model trained from sufficient database as a priori knowledge. And the state duration distribution is used lot adapting the speech duration information owing to speaker's utterance habit and speed. The recognition rate of the proposed methods are significantly higher than that of the conventional method using one mixture in each state.

• Journal of Korea Society of Industrial Information Systems
• /
• v.7 no.3
• /
• pp.9-14
• /
• 2002

In this paper we propose a method for performance enhancement of speech recognizer under noisy conditions. The parallel combination model which is presented at the PMC method using multiple Gaussian-distributed mixtures have been adapted to the variation of each mixture. The CDHMM(continuous observation density HMM) which has multiple Gaussian distributed mixtures are combined by the proposed PMC method. Also, the EM(expectation maximization) algorithm is used for adapting the model mean parameter in order to reduce the variation of the mixture density. The result of simulation, the proposed PMC adaptation method show better performance than the conventional PMC method.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.35S no.3
• /
• pp.166-175
• /
• 1998

When we make a speaker adapted model using MAPE (maximum a posteriori estimation), the adapted model has one mixture in each state. This is because we cannot estimate a number of a priori distribution from a speaker-independent model in each state. If the model is represented by one mixture in each state, it is not well adadpted to specific speaker because it is difficult to represent various speech informationof the speaker with one mixture. In this paper, we suggest the method using several mixtures to well represent various speech information of the speaker in each state. But, because speaker-specific training dat is not sufficient, this method can't be used in every state. So, we make the number of mixtures in each state variable in proportion to the number of frames and to the determinant ofthe variance matrix in the state. Using the proposed method, we reduced the error rate than methods using one branch in each state.

• The Journal of the Acoustical Society of Korea
• /
• v.17 no.3
• /
• pp.90-95
• /
• 1998

본 논문에서는 CHMM인 CDHMM과 ARHMM을 이용하여 화자적응화 하는 방법을 각각 연구하였다. CDHMM에서는 최대사후화확률 추정법에 의하여 각 상태마다 하나의 가 지를 이용하여 화자에 적응시킨다. 본 논문에서는 음성의 다양한 음향학적 특징을 표현하기 위하여 상태마다 여러 개의 가지를 갖는 방법을 제안하였다. 상태마다의 적절한 가지 수를 결정하기 위하여 각 상태에 속하는 프레임 수와 특징 벡터들의 분산행렬의 행렬식값을 이용 하였다. ARHMM에서는 특징벡터로 선형예측계수를 사용하기 때문에 최대사후화확률 추정 법을 사용할 수 없게 된다. 따라서 화자독립모델을 이용하여 적응화자에 대한 음성을 Viterbi 알고리듬으로 상태별로 분할한 후 k-means 알고리듬을 이용하여 각 상태마다 하나 의 가지를 갖는 모델로 적응시키는 방법을 제안하였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2002.10d
• /
• pp.544-546
• /
• 2002

본 논문은 CDHMM(Continuous Density Hidden Markov Model)의 훈련하는 방법을 동적 다중 그룹 혼합 가중치(Dynamic Mutli-Group mixture weight)을 이용하여 재구성하는 방법을 제안한다. 음성은 Hidden 상태열에 의하여 특성화되고, 각 상태는 가중된 혼합 가우시안 밑도 함수에 의해 표현된다. 음성신호를 더욱더 정확하게 계산하려면 각 상태를 위한 가우시안 함수를 더욱더 많이 사용해야 하며 이것은 많은 계산량이 요구된다. 이러한 문제는 가우시안 분포 확률의 통계적인 평균을 이용하면 계산량을 줄일 수 있다. 그러나 이러한 기존의 방법들은 다양한 화자의 발화속도와 가중치의 적용이 적합하지 못하여 인식률을 저하시키는 단점을 가지고 있다. 이 문제를 다양한 화자의 발화속도에 적합하도록 화자의 화자의 발화속도에 따라 동적으로 5개의 그룹으로 구성하고 동적 다중 그룹 혼합 가중치를 적용하여 CDHMM 파라미터를 재구성함으로써 8.5%의 인식율이 증가되었다.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.69-72
• /
• 2002

본 논문에서 화자공간모델 evolution에 기반한 continuous density hidden Markov model (CDHMM)의 online 적응에 대한 새로운 기법을 제안한다. 학습화자의 a priori knowledge을 나타내는 화자공간모델은 factor analysis (FA) 또는 probabilistic principal component analysis (PPCA)와 같은 은닉변수모델(latent variable model)에 의해 효과적으로 나타내어진다. 은닉 변수모델은 화자공간모델뿐아니라 CDHMM 파라메터의 ajoint prior분포를 표시함으로, maximum a posteriori(MAP)적응기법에 직접 적용되어진다. 화자공간모델의 hyperparameters와 CDHMM파라메터를 동시에 순차적으로 적응하기 위해 quasi-Bayes (QB)추정 기술에 기반한 online 적응기법을 제안한다. 연속숫자음 인식과 관련된 화자적응 실험을 통해 제안된 기법은 적은 적응데이터에서 좋은 성능을 나타내며, 데이터가 증가함에 따라 성능이 지속적으로 증가함을 보여준다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.29 no.5C
• /
• pp.627-631
• /
• 2004

An efficient algorithm for clustering of GPDFs(Gaussian Probability Density Functions) in a speech recognition model is proposed in this paper. The proposed algorithm is based on CNN with the divergence as its distance measure and is applied to a speech recognition. The algorithm is compared with conventional Dk-means(Divergence-based k-means) algorithm in CDHMM(Continuous Density Hidden Markov Model). The results show that it can reduce about 31.3％ of GPDFs over Dk-means algorithm without suffering any recognition performance. When compared with the case that no clustering is employed and full GPDFs are used, the proposed algorithm can save about 61.8％ of GPDFs while preserving the recognition performance.