• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.5
• /
• pp.765-776
• /
• 1995

This paper describes a speaker adaptation method to improve the recognition performance of MLP(multiLayer Perceptron) based HMM(Hidden Markov Model) speech recognizer. In this method, we use lst-order linear transformation network to fit data of a new speaker to the MLP. Transformation parameters are adjusted by back-propagating classification error to the transformation network while leaving the MLP classifier fixed. The recognition system is based on semicontinuous HMM's which use the MLP as a fuzzy vector quantizer. The experimental results show that rapid speaker adaptation resulting in high recognition performance can be accomplished by this method. Namely, for supervised adaptation, the error rate is signifecantly reduced from 9.2% for the baseline system to 5.6% after speaker adaptation. And for unsupervised adaptation, the error rate is reduced to 5.1%, without any information from new speakers.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.2
• /
• pp.28-34
• /
• 1994

In this paper, a new time-domain parameter extraction method for speech recognition is proposed. The suggested emthod is based on the fact that the local peak-to-peak interval, i.e., the interval between maxima and minima of speech waveform is closely related to the frequency component of the speech signal. The parameterization is achieved by a sort of filter bank technique in the time domain. To test the proposed parameter extraction emthod, an isolated word recognizer based on Vector Quantization and Hidden Markov Model was constructed. As a test material, 22 words spoken by ten males were used and the recognition rate of 92.9% was obtained. This result leads to the conclusion that the new parameter extraction method can be used for speech recognition system. Since the proposed method is processed in the time domain, the real-time parameter extraction can be implemented in the class of personal computer equipped onlu with an A/D converter without any DSP board.

• Journal of the Institute of Electronics Engineers of Korea TE
• /
• v.37 no.3
• /
• pp.16-20
• /
• 2000

Most of speech recognition systems are using Hidden Markov Model based on statistical modelling frequently. In Korean isolated telephone digit speech recognition, high recognition rate is gained by using HMM if many training data are given. But in Korean continuous telephone digit speech recognition, HMM has some limitations for similar telephone digits. In this paper we suggest a way to overcome some limitations of HMM by using discriminant learning based on minimal classification error criterion in Korean continuous telephone digit speech recognition. The experimental results show our method has high recognition rate for similar telephone digits.

• Proceedings of the IEEK Conference
• /
• 1999.06a
• /
• pp.675-678
• /
• 1999

본 논문에서는 음소별 코드북 개수의 선택과 벡터 양자화에 따른 음소 인식률과 고립단어 인식률에 대하여 다룬다. 음성모델은 이산 확률 밀도를 갖는 DHMM(Discrete Hidden Markov Model)을 사용하였으며, 코드북 생성과 벡터 양자화 알고리즘으로는 K-means 알고리즘과 LBG(Linde, Buzo, Gray) 알고리즘을 사용하였다 음소별 코드북 개수와 벡터 양자화를 최적화함으로써 음소 인식률을 향상시킬 수 있으며, 그 결과 안정된 고립단어 인식률을 얻을 수 있다.

• Annual Conference on Human and Language Technology
• /
• 1994.11a
• /
• pp.389-394
• /
• 1994

말뭉치에 품사를 부여하는 일은 언어연구의 중요한 기초가 된다. 형태소 해석의 모호한 결과로부터 한 가지 품사를 선정하는 작업을 태깅이라고 한다. 한국어에서 은닉 마르코프 모델 (Hidden Markov Model)을 이용한 태깅은 형태소 관계만 흑은 어절관계만을 이용한 방법이 있어 왔다. 본 논문에서는 어절관계와 형태소관계를 동시에 은닉 마르코프 모델에 반영하여 태깅의 정확도를 높인 모델을 제시한다. 제안된 방법은 품사의 변별력은 뛰어나지만 은닉 마르코프 모델의 노드의 수가 커짐으로써 형태소만을 고려한 방법보다 더 많은 학습데이타를 필요로 한다. 실험적으로 본 논문의 방법이 기존의 방법보다 높은 정확성을 가지고 있음이 검증되었다.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.125-128
• /
• 2001

본 연구에서는 음성신호의 왜곡에 대해 음성 부재 확률을 고려한 MMSE(Minimum Mean Square Error) STSA(Short-Time Spectral Amplitude Estimator)를 전처리기로 도입하여 HMM(Hidden Markov Model)에 기반 한 음성인식시스템의 인식성능을 평가하였다. 음성인식 시스템의 실시간 구현을 고려하여, MMSE STSA 기법을 음성개선을 위한 전처리기로 사용할 때 MMSE STSA의 이득계산 과정에서 많은 계산량이 요구되는 modified Bessel 함수를 근사 화하여 사용하였다.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.3E
• /
• pp.31-36
• /
• 1996

In this paper, a two-stage recognition approach based on error pattern hypotheses is proposed to reduce errors of a connected digit recognizer. In the approach, a conventional recognizer is first used to produce N-best candidate strings, and then error patterns are hypothesized by examining the candidate strings. For substitution error pattern hypotheses, error-pattern-dependent classifiers having more discriminative power than the first-stage classifier are used ; and for insertion and deletion errors, word duration and energy contour information are exploited are exploited to discriminated confusing pairs. Simulation results showed that the proposed approach achieves 15% decrease in word error rate for speaker-independent Korean connected digit recognition when a hidden Markov model-based recognizer is used for the first-stage classifier.

• Journal of Korea Multimedia Society
• /
• v.21 no.3
• /
• pp.391-399
• /
• 2018

In this paper, we construct and compare various guitar chord classification systems using perceptron neural network and convolutional neural network without pre-processing other than Fourier transform to identify the optimal chord classification system. Conventional guitar chord classification schemes use, for better feature extraction, computationally demanding pre-processing techniques such as stochastic analysis employing a hidden markov model or an acoustic data filtering and hence are burdensome for real-time chord classifications. For this reason, we construct various perceptron neural networks and convolutional neural networks that use only Fourier tranform for data pre-processing and compare them with dataset obtained by playing an electric guitar. According to our comparison, convolutional neural networks provide optimal performance considering both chord classification acurracy and fast processing time. In particular, convolutional neural networks exhibit robust performance even when only small fraction of low frequency components of the data are used.

• Proceedings of the Korean Information Science Society Conference
• /
• 2012.06b
• /
• pp.477-479
• /
• 2012

개의 발성은 성도의 물리적인 특징에 따라 고유의 특정 포먼트를 만들어 내며 개의 품종에 따라 다른 물리적 특징을 가지므로 개의 발성을 HMM(Hidden Markov Model)으로 모델링하여 개의 품종을 분류하는 연구를 하였다. 주파수 특징은 MFCC(Mel Frequency Cepstral Coefficients) 12차, 에너지 컴포넌트 1차, 델타 13차, 억셀러레이션(Acceleration) 13차, 총 39차 벡터를 사용하였다. 개의 품종 분류에 적합한 HMM 구조의 설계를 위하여 기본 좌우 모델, 좌우 모델, 좌우 모델2, 전후진 모델, 총 4가지를 제안하고 실험하여 성능을 비교분석하였다. 이 중 전후진 모델이 가장 바람직한 모델로 검증 되었다. 본 모델은 다음과 같은 장점을 갖는다. (1) 기본 좌우 모델과 마찬가지로 1~2회 발성을 갖는 데이터가 입력되어도 처음에서 마지막 상태까지의 이동단계가 최소 3번까지 가능하므로 적은 횟수의 발성 데이터도 처리가 가능하다. (2) 다수 반복된 발성 데이터의 신호도 처리가 가능하다. 즉, 본 모델은 상태의 이동이 후진도 가능하므로 5회이상 반복된 발성 데이터의 신호의 처리도 가능하다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.34 no.3C
• /
• pp.257-267
• /
• 2009

This paper presents a training method for neural networks and the employment of MSE (mean scare error) values as the basis of a decision regarding the identity claim of a speaker in a recurrent neural networks based speaker verification system. Recurrent neural networks (RNNs) are employed to capture temporally dynamic characteristics of speech signal. In the process of supervised learning for RNNs, target outputs are automatically generated and the generated target outputs are made to represent the temporal variation of input speech sounds. To increase the capability of discriminating between the true speaker and an impostor, a discriminative training method for RNNs is presented. This paper shows the use and the effectiveness of the MSE value, which is obtained from the Euclidean distance between the target outputs and the outputs of networks for test speech sounds of a speaker, as the basis of speaker verification. In terms of equal error rates, results of experiments, which have been performed using the Korean speech database, show that the proposed speaker verification system exhibits better performance than a conventional hidden Markov model based speaker verification system.