• Proceedings of the KSPS conference
• /
• 2006.05a
• /
• pp.37-40
• /
• 2006

In this paper, we propose a VQ codebook design of speech recognition feature parameters in order to improve the performance of a distributed speech recognition system. For the context-dependent HMMs, a VQ codebook should be correlated with phonetic distributions in the training data for HMMs. Thus, we focus on a selection method of training data based on phonetic distribution instead of using all the training data for an efficient VQ codebook design. From the speech recognition experiments using the Aurora 4 database, the distributed speech recognition system employing a VQ codebook designed by the proposed method reduced the word error rate (WER) by 10% when compared with that using a VQ codebook trained with the whole training data.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.3 s.303
• /
• pp.81-90
• /
• 2005

When we transmit signals, which are quantized by the vector quantizer (VQ), through noisy channels, the overall performance of the coding system is very dependent on the employed quantization scheme and the channel error effect. In order to design an optimal coding system, the source and channel coding scheme should be jointly optimized as in the channel-optimized VQ. As a suboptimal approach, we may consider the robust VQ (RVQ). In RVQ, we consider developing an index assignment function for mapping the output of quantizers to channel symbols so that the effect of the channel errors is minimized. Recently, a VQ, which can reduce the encoding complexity and is called the sample-adaptive product quantizer (SAPQ), has been proposed. SAPQ has very similar quantizer structure as to the product quantizer (PQ). However, the quantization performance can be better than PQ. Further, the encoding complexity and the memory requirement for the codebooks are lower than the regular full-search VQ case. In this paper, SAPQ is employed in order to design an RVQ to channel errors by reducing the vector dimension. Discussions on the codebook structure of SAPQ and experiments are introduced in an aspect of robustness to noisy channels.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.35S no.8
• /
• pp.96-105
• /
• 1998

The aim of this paper is to propose the method of WDHMM(Weighted DHMM), using the MLP-VQ for the improvement of speaker-independent connect-digit recognition system. MLP neural-network output distribution shows a probability distribution that presents the degree of similarity between each pattern by the non-linear mapping among the input patterns and learning patterns. MLP-VQ is proposed in this paper. It generates codewords by using the output node index which can reach the highest level within MLP neural-network output distribution. Different from the old VQ, the true characteristics of this new MLP-VQ lie in that the degree of similarity between present input patterns and each learned class pattern could be reflected for the recognition model. WDHMM is also proposed. It can use the MLP neural-network output distribution as the way of weighing the symbol generation probability of DHMMs. This newly-suggested method could shorten the time of HMM parameter estimation and recognition. The reason is that it is not necessary to regard symbol generation probability as multi-dimensional normal distribution, as opposed to the old SCHMM. This could also improve the recognition ability by 14.7% higher than DHMM, owing to the increase of small caculation amount. Because it can reflect phone class relations to the recognition model. The result of my research shows that speaker-independent connected-digit recognition, using MLP-VQ and WDHMM, is 84.22%.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.8 no.2
• /
• pp.60-69
• /
• 1998

This paper proposes a speech synthesis method using Fuzzy VQ, and then study how to make choice of fuzziness value which optimizes (controls) the performance of FVQ in order to obtain the synthesized speech which is closer to the original speech. When FVQ is used to synthesize a speech, analysis stage generates membership function values which represents the degree to which an input speech pattern matches each speech patterns in codebook, and synthesis stage reproduces a synthesized speech, using membership function values which is obtained in analysis stage, fuzziness value, and fuzzy-c-means operation. By comparsion of the performance of the FVQ and VQ synthesizer with simmulation, we show that, although the FVQ codebook size is half of a VQ codebook size, the performance of FVQ is almost equal to that of VQ. This results imply that, when Fuzzy VQ is used to obtain the same performance with that of VQ in speech synthesis, we can reduce by half of memory size at a codebook storage. And then we have found that, for the optimized FVQ with maximum SQNR in synthesized speech, the fuzziness value should be small when the variance of analysis frame is relatively large, while fuzziness value should be large, when it is small. As a results of comparsion of the speeches synthesized by VQ and FVQ in their spectrogram of frequency domain, we have found that spectrum bands(formant frequency and pitch frequency) of FVQ synthesized speech are closer to the original speech than those using VQ.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1993.06a
• /
• pp.109-112
• /
• 1993

In this paper, we propose the modified VQ, applied correlation between codewords in order to reduce the error rate due to personal and speakers' temporal variation. Such a modified VQ is used in the stage of preprocessing of HMM and the temporal variation is absorbed by nonlinear Decimation and Interpolation of vowel part that we obtain higher recognition rate than not so case. The objects of experiment are Korea 142 DDD regional names and we show that the proposed method increase the recognition rate.

• The Journal of the Acoustical Society of Korea
• /
• v.15 no.1E
• /
• pp.34-38
• /
• 1996

In this paper, a new fast voronoi divider for vector quantization (VQ) is introduced, which results from Theorem that the nearest vectors in the sense of minimum mean square error(MMSE) have almost the same mean values of their elements. An improved splitting method for a VQ codebook design using the fast voronoi divider is also presented. Experimental results show that the new method reduces the complexity of training a VQ codebook several times with a high signal to noise ratio(SNR) using an appropriate extensive parameter of codebook.

• Proceedings of the Acoustical Society of Korea Conference
• /
• 1991.06a
• /
• pp.50-53
• /
• 1991

벡터 양지화(VQ)는 신호 처리분야에서 정보의 압축을 위해 사용하는 아주 잘 알려진 방법이다. 벡터 양지화는 정보를 대량으로 줄이면서 그 효율을 떨어 뜨리지 않는 방향으로 발전해 왔다. VQ코드북의 크기가 커지면 하나의 코드워드를 찾기위한 시간이 증가하게 된다. 코드북의 빠른 검색을 위하여 다른 방법에 제안 되기도 했으나 최적 검색 방법이라고는 볼 수 없다. 본 고에서는 음성인식에 적용할 목적으로 기존의 방법으로 구성된 코드북의 구성을 변형 하지 않고 검색 속도를 증가 시킬 수 있는 방법을 기수랗고 그 효율에 대해서 설명한다.

• The Journal of the Acoustical Society of Korea
• /
• v.17 no.8
• /
• pp.78-81
• /
• 1998

본 논문에서는 가변블록에 대해 DCT와 VQ를 적용함으로써 복원시 영상의 품질을 원하는 수준으로 유지하면서 비트율을 감소시키는 방법을 제안한다. 각 블록에 대해 임계값 에 따라 블록의 크기를 세분화 시켜 최종적인 블록에 대해 DCT와 VQ를 적용한다. 제안된 방법에서 12개의 영상에 대한 PSNR과 비트율을 비교했을 때 기존의 방법보다 비트율이 크 게 감소함을 보여주고 있다.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.4
• /
• pp.40-49
• /
• 1994

In this paper, we propose methods for improving the performance of word recognition system. The ray stratey of the first method is to apply the inertia to the feature vector sequences of speech signal to stabilize the transitions between VQ cdoes. The second method is generating the new observation probabilities using the transition matrix of VQ codes as weights at the observation probability of the output symbol, so as to take into account the time relation between neighboring frames in DHMM. By applying the inertia to the feature vector sequences, we can reduce the overlapping of probability distribution of the response paths for each word and stabilize state transitions in the HMM. By using the transition matrix of VQ codes as weights in conventional DHMM. we can divide the probability distribution of feature vectors more and more, and restrict the feature distribution to a suitable region so that the performance of recognition system can improve. To evaluate the performance of the proposed methods, we carried out experiments for 50 DDD area names. As a result, the proposed methods improved the recognition rate by $4.2\%$ in the speaker-dependent test and $12.45\%$ in the speaker-independent test, respectively, compared with the conventional DHMM.

• Journal of KIISE:Computing Practices and Letters
• /
• v.7 no.6
• /
• pp.724-732
• /
• 2001

In this paper, a new method of feature extraction of major objects to represent an image using Vector Quantization(VQ) is proposed. The principal features of the image, which are used in a content-based image retrieval system, are color, texture, shape and spatial positions of objects. The representative color and texture features are extracted from the given image using VQ(Vector Quantization) clustering algorithm with a general feature extraction method of color and texture. Since these are used for content-based image retrieval and searched by objects, it is possible to search and retrieve some desirable images regardless of the position, rotation and size of objects. The experimental results show that the representative feature extraction time is much reduced by using VQ, and the highest retrieval rate is given as the weighted values of color and texture are set to 0.5 and 0.5, respectively, and the proposed method provides up to 90% precision and recall rate for 'person'query images.