• Journal of Internet Computing and Services
• /
• v.2 no.1
• /
• pp.31-38
• /
• 2001

Speech database built of phonemes is significant in the studies of speech recognition, speech synthesis and analysis, Phoneme, consist of voiced sounds and unvoiced ones, Though there are many feature differences in voiced and unvoiced sounds, the traditional algorithms for detecting the boundary between phonemes do not reflect on them and determine the boundary between phonemes by comparing parameters of current frame with those of previous frame in time domain, In this paper, we propose the assort algorithm, which is based on a block and reflecting upon the feature differences between voiced and unvoiced sounds for phoneme segmentation, The assort algorithm uses the distance measure based upon MFCC(Mel-Frequency Cepstrum Coefficient) as a comparing spectrum measure, and uses the energy, zero crossing rate, spectral energy ratio, the formant frequency to separate voiced sounds from unvoiced sounds, N, the result of out experiment, the proposed system showed about 79 percents precision subject to the 3 or 4 syllables isolated words, and improved about 8 percents in the precision over the existing phonemes segmentation system.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.9 no.11
• /
• pp.1221-1226
• /
• 2014

In an effort to improve the efficiency of the speech recognition, we investigate the effect of the number of phoneme clusters. For this purpose, codebooks of varied number of phoneme clusters are prepared by modified k-means clustering algorithm. The subsequent processing is fuzzy vector quantization (FVQ) and hidden Markov model (HMM) for speech recognition test. The result shows that there are two distinct regimes. For large number of phoneme clusters, the recognition performance is roughly independent of it. For small number of phoneme clusters, however, the recognition error rate increases nonlinearly as it is decreased. From numerical calculation, it is found that this nonlinear regime might be modeled by a power law function. The result also shows that about 166 phoneme clusters would be the optimal number for recognition of 300 isolated words. This amounts to roughly 3 variations per phoneme.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.87-90
• /
• 2000

문-음성 합성기의 자연감을 높이기 위해 주로 자연음에 존재하는 운률 법칙을 정확히 구현해 주어야 한다. 일반적으로 언어학적 정보를 이용하거나 자연음으로부터 추출한 운률 정보를 추출한 운률 법칙을 합성에 이용하고 있다. 이와 같이 구한 운률 법칙이 자연음에 존재하는 모든 운률 법칙을 포함할 수 있으면, 자연스러운 합성음을 들을 수 있겠으나, 실질적으로는 모든 법칙을 구현한다는 것은 어려운 실정이고, 자연음으로부터 추출한 운률 법칙이 잘못 구현되는 경우 합성음의 자연성이 떨어지는 것을 피할 수 없을 것이다. 이런 점을 고려하여 우리는 자연음에 내재하는 운율 법칙을 훈련을 통해 학습할 수 있는 인공 신경망을 제안하였다 운률의 세 가지 요소는 피치, 지속시간, 크기 변화가 있는데, 인공 신경망은 문장이 입력되면, 각 해당 음소의 지속시간에 따른 피치 변화와 크기 변화를 학습할 수 있도록 설계하였다. 신경망을 훈련시키기 위해 고립 단어군과 음소균형 문장군을 화자로 하여금 발성하게 하여, 녹음하고, 분석하여 운률 데이터베이스를 구축하였다. 자연음의 각 음소에 대해 지속시간과 피치변화 그리고 크기 변화를 구하여 곡선 적응 방법을 이용하여 각 변화 곡선에 대한 계수를 구해 데이터베이스를 구축한다. 이렇게 구축한 데이터베이스를 이용해 인공 신경망을 훈련시켜 평가한 결과 훈련용 데이터를 계속 확장하면 좀 더 자연스러운 운률을 발생시킬 수 있음을 관찰하였다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.31 no.9C
• /
• pp.853-858
• /
• 2006

TMHMM(Tied Mixture Hidden Markov Model) is an important approach to reduce the number of free parameters in speech recognition. However, this model suffers from a degradation in recognition accuracy due to its GPDF (Gaussian Probability Density Function) clustering error. This paper proposes a clustering algorithm, called HCNN(Homogeneous Centroid Neural network), to cluster acoustic feature vectors in TMHMM. Moreover, the HCNN uses the heterogeneous distance measure to allocate more code vectors in the heterogeneous areas where probability densities of different states overlap each other. When applied to Korean digit isolated word recognition, the HCNN reduces the error rate by 9.39% over CNN clustering, and 14.63% over the traditional K-means clustering.

• The Journal of the Acoustical Society of Korea
• /
• v.18 no.5
• /
• pp.78-82
• /
• 1999

This paper is a study on weighted cepstrum used broadly for robust speech recognition. Especially, we propose the weighted function of asymmetric glottal pulse shape. which is used for weighted cepstrum extracted by PLP(Perceptual Linear Predictive) based on auditory model. Also, we analyze this glottal weighted cepstrum from the glottal pulse of glottal model in connection with the cepstrum. And we obtain speech features analyzed by both the glottal model and the auditory model. The isolated-word recognition rate is adopted for the test of proposed method in the car moise and street environment. And the performance of glottal weighted cepstrum is compared with both that of weighted cepstrum extracted by LP(Linear Prediction) and that of weighted cepstrum extracted by PLP. The result of computer simulation shows that recognition rate of the proposed glottal weighted cepstrum is better than those of other weighted cepstrums.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.3
• /
• pp.543-552
• /
• 1994

Dynamic Time Warp(DTW) needs for iterative calculations and the design of PE cell suitable for the operations is very important. Accordingly, this paper aims at real time recognition design enables large dictionary hardware realization using DTW algorithm. The DTW PE cell separated into three large blocks. "MIN" is the one block for counting accumulated minimum distance. "ADD" block calculates these minimum distances, and "ABS" seeks for the absolute values to the total sum of local distances. Circuit design and verification about the three block have been accomplished, and performed layout '||'&'||' DRC(design rule check) using 1.2 m CMOS N-Well rule base.CMOS N-Well rule base.

• The Journal of the Korea Contents Association
• /
• v.5 no.6
• /
• pp.57-64
• /
• 2005

LBL(Layer-By-Layer) algorithms have been proposed to accelerate the training speed of MLPs(Multilayer Perceptrons). In this LBL algorithms, each layer needs a error function for optimization. Especially, error function for hidden layer has a great effect to achieve good performance. In this sense, this paper proposes a new hidden layer error function for improving the performance of LBL algorithm for MLPs. The hidden layer error function is derived from the mean squared error of output layer. Effectiveness of the proposed error function was demonstrated for a handwritten digit recognition and an isolated-word recognition tasks and very fast learning convergence was obtained.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.32B no.11
• /
• pp.1527-1535
• /
• 1995

This paper introduces an isolated word recognition system realized on a neurocomputer called E-MIND II, which is a 2-D torus wavefront array processor consisting of 256 DNP IIs. The DNP II is an all digital VLSI unit processor for the EMIND II featuring the emulation capability of more than thousands of neurons, the 40 MHz clock speed, and the on-chip learning. Built by these PEs in 2-D toroidal mesh architecture, the E- MIND II can be accelerated over 2 Gcps computation speed. In this light, the advantages of the E-MIND II in its capability of computing speed, scalability, computer interface, and learning are especially suitable for real time application such as speech recognition. We show how to map a TDNN structure on this array and how to code the learning and recognition algorithms for a user independent isolated word recognition. Through hardware simulation, we show that recognition rate of this system is about 97% for 30 command words for a robot control.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.53-56
• /
• 2000

무제한 어휘 음성합성 시스템의 문-음성 합성기는 합성음의 자연감을 높이기 위해 여러 가지 방법을 사용하게되는데 그중 하나가 자연음에 내재하는 운을 법칙을 정확히 구현하는 것이다. 합성에 필요한 운율법칙은 언어학적 정보를 이용해 구현하거나, 자연음을 분석해 구한 운을 정보로부터 운율 법칙을 추출하여 합성에 이용하고 있다. 이와 같이 구한 운을 법칙이 자연음에 존재하는 운율 법칙을 전부 반영하지 못했거나, 잘못 구현되는 경우에는 합성음의 자연성이 떨어지게 된다. 이런 점을 고려하여 우리는 자연음의 운율 정보를 이용해 인공 신경망을 훈련시켜, 문장단위 운율을 발생시킬 수 있는 방식을 제안하였다. 운율의 세 가지 요소는 피치, 지속시간, 크기 변화가 있는데, 인공 신경망은 문장이 입력되면, 각 해당 음소의 지속시간에 따른 피치 변화와 크기 변화를 학습할 수 있도록 설계하였다. 신경망을 훈련시키기 위해 고립 단어 군과 음소균형 문장 군을 화자로 하여금 발성하게 하여, 녹음하고, 분석하여 구한 운을 정보를 데이터베이스로 구축하였다. 문장 내의 각 음소에 대해 지속시간과 피치 변화 그리고 크기 변화를 구하고, 곡선적응 방법을 이용하여 각 변화 곡선에 대한 다항식 계수와 초기치를 구해 운을 데이터베이스를 구축한다. 이 운을 데이터베이스의 일부를 인공 신경망을 훈련시키는데 이용하고, 나머지를 이용해 인공 신경망의 성능을 평가한 결과 운을 데이터베이스를 계속 확장하면 좀더 자연스러운 운율을 발생시킬 수 있음을 관찰하였다.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.8
• /
• pp.498-504
• /
• 2005

We introduce a Korean speech recognition platform (ECHOS) developed for education and research Purposes. ECHOS lowers the entry barrier to speech recognition research and can be used as a reference engine by providing elementary speech recognition modules. It has an easy simple object-oriented architecture, implemented in the C++ language with the standard template library. The input of the ECHOS is digital speech data sampled at 8 or 16 kHz. Its output is the 1-best recognition result. N-best recognition results, and a word graph. The recognition engine is composed of MFCC/PLP feature extraction, HMM-based acoustic modeling, n-gram language modeling, finite state network (FSN)- and lexical tree-based search algorithms. It can handle various tasks from isolated word recognition to large vocabulary continuous speech recognition. We compare the performance of ECHOS and hidden Markov model toolkit (HTK) for validation. In an FSN-based task. ECHOS shows similar word accuracy while the recognition time is doubled because of object-oriented implementation. For a 8000-word continuous speech recognition task, using the lexical tree search algorithm different from the algorithm used in HTK, it increases the word error rate by $40\%$ relatively but reduces the recognition time to half.