• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 1998년도 하계종합학술대회논문집
• /
• pp.933-935
• /
• 1998

The HMM-Net is an architecture for a neural network that implements a hidden markov model (HMM). The architecture is developed for the purpose of combining the discriminant power of neural networks with the time-domain modeling capability of HMMs. Criteria used for learning HMM-Net classifiers are maximum likelihood(ML) and minimization of mean squared error(MMSE). In this paper we propose an efficient learning method of HMM_Net classifiers using a ML-MMSE hybrid criterion and report the results of an experimental study comparing the performance of HMM_Net classifiers trained by the gradient descent algorithm with the above criteria. Experimental results for the isolated numeric digits from /0/ to /9/ show that the performance of the proposed method is better than the others in the repects of learning and recognition rates.

• 전자공학회논문지C
• /
• 제34C권9호
• /
• pp.61-67
• /
• 1997

The HMM-Net is an architecture for a neural network that implements a hidden markov model(HMM). The architecture is developed for the purpose of combining the classification power of neural networks with the time-domain modeling capability of HMMs. Criteria which are used for learning HMM_Net classifiers are maximum likelihood(ML), maximum mutual information (MMI), and minimization of mean squared error(MMSE). In this classifiers trained by the gradient descent algorithm with the above criteria. Experimental results for the isolated numbers from /young/to/koo/ show that in the binary inputs the performance of MMSE is better than the others, while in the fuzzy inputs the performance of MMI is better than the others.

• 한국생물정보학회:학술대회논문집
• /
• 한국생물정보시스템생물학회 2000년도 International Symposium on Bioinformatics
• /
• pp.28-31
• /
• 2000

We describe a hidden Markov model, HMMTIR, for general protein sequence based on the I-sites library of sequence-structure motifs. Unlike the linear HMMs used to model individual protein families, HMMSTR has a highly branched topology and captures recurrent local features of protein sequences and structures that transcend protein family boundaries. The model extends the I-sites library by describing the adjacencies of different sequence-structure motifs as observed in the database, and achieves a great reduction in parameters by representing overlapping motifs in a much more compact form. The HMM attributes a considerably higher probability to coding sequence than does an equivalent dipeptide model, predicts secondary structure with an accuracy of 74.6% and backbone torsion angles better than any previously reported method, and predicts the structural context of beta strands and turns with an accuracy that should be useful for tertiary structure prediction. HMMSTR has been incorporated into a public, fully-automated protein structure prediction server.

• The Journal of the Acoustical Society of Korea
• /
• 제17권4E호
• /
• pp.11-16
• /
• 1998

For large vocabulary speech recognition using HMMs, context-dependent subword units have been often employed. However, when context-dependent phone models are used, they result in a system which has too may parameters to train. The problem of too many parameters and too little training data is absolutely crucial in the design of a statistical speech recognizer. Furthermore, when building large vocabulary speech recognition systems, unseen triphone problem is unavoidable. In this paper, we propose the modified phonetic decision tree algorithm for the automatic prediction of unseen triphones which has advantages solving these problems through following two experiments in Japanese contexts. The baseline experimental results show that the modified tree based clustering algorithm is effective for clustering and reducing the number of states without any degradation in performance. The task experimental results show that our proposed algorithm also has the advantage of providing a automatic prediction of unseen triphones.

• 한국정보과학회:학술대회논문집
• /
• 한국정보과학회 2000년도 봄 학술발표논문집 Vol.27 No.1 (B)
• /
• pp.238-240
• /
• 2000

침입탐지시스템은 침입탐지 기법에 따라 크게 오용탐지시스템과 비정상행위탐지시스템으로 나뉜다. 비정상 행위 탐지시스템은 정상사용행위를 모델링한 후 현재 관찰중인 행위가 정상에서 벗어나는지를 검사한다. 시스템 사용시 발생하는 각 이벤트는 동시에 여러 가지 정보를 담고있으므로 여러 각도에서 모델링될 수 있다. 따라서 여러 결과를 종합해서 판정의 안정성을 높을 수 있다. 본 논문에서는 이벤트의 시스템호출에 평가결과와 BSM감사정보 중 시스템호출관련 정보, 파일 접근관련 정보, 이 둘을 모두 고려한 정보를 통합한 평가결과를 투표방식으로 결합하여 판정하는 기법을 제안하였다. 실험결과 두 모델을 별도로 적용하는 경우보다 나아진 판정성능을 보여주었다.

• 음성과학
• /
• 제13권2호
• /
• pp.75-84
• /
• 2006

Various methods have been proposed to overcome the problem of speech recognition in the noisy conditions. Among them, the model compensation methods like the parallel model combination (PMC) and Jacobian adaptation (JA) have been found to perform efficiently. The JA is quite effective when we have hidden Markov models (HMMs) already trained in a similar condition as the target environment. In a previous work, we have proposed an improved method for the JA to make it more robust against the changing environments in recognition. In this paper, we further improved its performance by compensating the delta-mean vectors and covariance matrices of the HMM and investigated its feasibility in the multi-model structure for the noisy speech recognition. From the experimental results, we could find that the proposed improved the robustness of the JA and the multi-model approach could be a viable solution in the noisy speech recognition.

• 한국통신학회논문지
• /
• 제26권6B호
• /
• pp.820-827
• /
• 2001

한글은 둘 또는 세 개의 자모가 사각형 영역 안에 적절히 배치된 구조로 되어 있다. 이와 같은 구성 방법에 따라 글자의 영상을 합성하고 이를 실시간에 Pseudo 2D HMM으로 변환하는 방법을 제안한다. 본 방법에 따라 실시간 합성된 모델과 추가의 필러(filler) 모델, 여백 모델을 문서 영상의 글자 영역에서 핵심어 검출에 적용하였다. 실험 결과 최소한의 설계 변수 조정으로도 오검출, 미검출률이 낮고 언어 모델 없이 숫자 89%, 한글 80%의 검출성능을 보였으며, 따라서 제안된 방법이 인쇄 문자 패턴의 실시간 모델링 및 키워드 검출에 효과가 있음을 보였다. 본 연구 결과는 내용 기반의 광학 문서 색인 등에 활용할 수 있다.

• 한국통신학회논문지
• /
• 제21권6호
• /
• pp.1379-1389
• /
• 1996

한글 필기는 항상 초성, 중성, 종성의 순으로 씌어진다. 본 논문은 이점을 이용하여 자소 탐색 모델을 설계하고 그 탐색 결과에 의거하여 글자를 인식하려는 온라인 필기 인식 방법을 제시하고자 한다. 기본 자소 모델은 은닉 마르코프 모델을 이용하고 자소 탐색 모델은 HMM의 망으로 구성한다. 자소 탐색은 Viterbi 알고리즘에 의한 정합으로 이루어지며 글자 인식은 이들 자소 가설 격자의 탐색으로 이루어진다. 인식 실험 결과는 간단한 인식기 구조에도 불구하고 정자체의 경우 87.47%에 달하는 상당한 인식률을 보였으며, 특히 자연스럽게 쓴 필기에서도 매우 훌륭한 자소 분할 결과를 얻을 수 있었다.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2002년도 하계종합학술대회 논문집(4)
• /
• pp.195-198
• /
• 2002

We propose an efficient mixture Gaussian synthesis method for decision tree based state tying that produces better context-dependent models in a short period of training time. This method makes it possible to handle mixture Gaussian HMMs in decision tree based state tying algorithm, and provides higher recognition performance compared to the conventional HMM training procedure using decision tree based state tying on single Gaussian GMMs. This method also reduces the steps of HMM training procedure. We applied this method to training of PBS, and we expect to achieve a little point improvement in phoneme accuarcy and reduction in training time.

• 전자공학회논문지B
• /
• 제31B권6호
• /
• pp.17-24
• /
• 1994

This paper is a study on Korean word recognition and suggest the method that stabilizes the state-transition in the HMM by applying the `inertia' to the feature vector sequences. In order to reduce the quantized distortion considering probability distribution of input vectors, we used SOFM, an unsupervised learning method, as a vector quantizer, By applying inertia to the feature vector sequences, the overlapping of probability distributions for the response path of each word on the self organizing feature map can be reduced and the state-transition in the Hmm can be Stabilized. In order to evaluate the performance of the method, we carried out experiments for 50 DDD area names. The results showed that applying inertia to the feature vector sequence improved the recognition rate by 7.4% and can make more HMMs available without reducing the recognition rate for the SOFM having the fixed number of neuron.