• Journal of Broadcast Engineering
• /
• v.25 no.2
• /
• pp.157-165
• /
• 2020

In this paper. we propose an effective compensation scheme to the quantization error arisen from quantized learning in a machine learning on an embedded system. In the machine learning based on a gradient descent or nonlinear signal processing, the quantization error generates early vanishing of a gradient and occurs the degradation of learning performance. To compensate such quantization error, we derive an orthogonal compensation vector with respect to a maximum component of the gradient vector. Moreover, instead of the conventional constant learning rate, we propose the adaptive learning rate algorithm without any inner loop to select the step size, based on a nonlinear optimization technique. The simulation results show that the optimization solver based on the proposed quantized method represents sufficient learning performance.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.19 no.7
• /
• pp.1203-1212
• /
• 1994

This paper proposes a variable block size vector quantization based on a biorthogonal wavelet transform for image compression. An image is first decomposed with the biorthogonal wavelet transform into multiresolution image and the wavelet coefficients of the middle frequency bands are segmented using the quadtree sturcture to extract the perceptually important regions in the middle frequency bands. A sedges of middle frequency bands exist the corresponding position of high frequency bands, the complicated quadtree structure of middle frequency bands is equally applied to the high frequency bands. Therefore the overhaed information of the quadtree codes needed to segment the high frequency bands can be reduced. The segmented subblocks are encoded with the codebook designed at the each scales and directions. The simulation results showed that the proposed methods could reproduce higher quality image with bit rate reduced about 20(%) than of the preceding VQ method and sufficiently reduce the bolck effect and the edge degradation.

• Journal of Digital Convergence
• /
• v.16 no.11
• /
• pp.335-340
• /
• 2018

Commercialized speech recognition systems that have an accuracy recognition rates are used a learning model from a type of speaker dependent isolated data. However, it has a problem that shows a decrease in the speech recognition performance according to the quantity of data in noise environments. In this paper, we proposed the vector quantization based speech recognition performance improvement using maximum log likelihood in Gaussian distribution. The proposed method is the best learning model configuration method for increasing the accuracy of speech recognition for similar speech using the vector quantization and Maximum Log Likelihood with speech characteristic extraction method. It is used a method of extracting a speech feature based on the hidden markov model. It can improve the accuracy of inaccurate speech model for speech models been produced at the existing system with the use of the proposed system may constitute a robust model for speech recognition. The proposed method shows the improved recognition accuracy in a speech recognition system.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.49 no.1
• /
• pp.11-18
• /
• 2012

In this paper, an entropy constrained vector quantizer for high bit-rates is proposed. The sample-adaptive product quantizer (SAPQ), which is based on the product codebooks, is employed, and a design algorithm for the entropy constrained sample adaptive product quantizer (ECSAPQ) is proposed. The performance of the proposed ECSAPQ is better than the case of the entropy constrained vector quantizer by 0.5dB. It is also shown that the ECSAPQ distortion curve, which is based on the scalar quantizer, is lower than the high-rate theoretical curve of the entropy constrained scalar quantizer, where the theoretical curve have 1.53dB difference from Shannon's lower bound.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.259-262
• /
• 2002

벡터양자화에서 주어진 학습벡터를 가장 잘 대표할 수 있는 코드벡터의 집합인 코드북을 구하는 것은 가장 중요한 문제이다. 이러한 코드북을 구하는 알고리즘 중에서 가장 대표적인 방법은 K-means 알고리즘으로 그 성능이 초기 코드북에 크게 의존한다는 문제점을 가지고 있어 여러 가지 초기 코드북을 설계하는 알고리즘이 제안되어 왔다. 본 논문에서는 splitting 방법을 이용한 수정된 초기 코드북 생성 알고리즘을 제안하고자 한다. 제안된 방법에서는 기존외 splitting 방법을 적용하여 초기 코드북을 생성하되, 미소분리 과정 시 학습벡터의 수렴 빈도가 가장 낮은 코드벡터를 제거하고 수렴 빈도가 가장 높은 코드벡터를 미소분리 하여 수렴 빈도가 가장 낮은 코드벡터와 대체해가며 초기 코드북을 설계 한다. 제안된 방법의 적용온 기존 방법에서 MSE(mean square error)의 감소율이 가장 작은 미소분리 과정에서 시작하여 원하는 코드북 크기를 얻을 때까지 반복한다. 제안된 방법으로 생성된 초기 코드북을 사용하여 K-means 알고리즘을 수행한 결과 기존의 splitting 방법으로 생성된 초기 코드북을 사용한 경우보다 코드북의 성능이 향상되었다.

• Journal of Broadcast Engineering
• /
• v.2 no.2
• /
• pp.136-145
• /
• 1997

As a 3D mesh model consists of a lot of vertices and polygons and each vertex position is represented by three 32 bit floating-point numbers in a 3D coordinate, the amount of data needed for representing the model is very excessive. Thus, in order to store and/or transmit the 3D model efficiently, a 3D model compression is necessarily required. In this paper, a 3D model compression method using PRVQ (predictive residual vector quantization) is proposed. Its underlying idea is based on the characteristics such as high correlation between the neighboring vertex positions and the vectorial property inherent to a vertex position. Experimental results show that the proposed method obtains higher compression ratio than that of the existing methods and has the advantage of being capable of transmitting the vertex position data progressively.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.5
• /
• pp.262-270
• /
• 2005

In this paper, a new temporal decomposition method is proposed. where not oniy distortion but also entropy are involved in segmentation. The interpolation functions and the target feature vectors are determined by a dynamic Programing technique. where both distortion and entropy are simultaneously minimized. The interpolation functions are built by using a training speech corpus. An iterative method. where segmentation and estimation are iteratively performed. finds the locally optimum Points in the sense of minimizing both distortion and entropy. Simulation results -3how that in terms of both distortion and entropy. the Proposed temporal decomposition method Produced superior results to the conventional split vector-quantization method which is widely employed in the current speech coding methods. According to the results from the subjective listening test, the Proposed method reveals superior Performance in terms of qualify. comparing to the Previous vector quantization method.

• Journal of Sensor Science and Technology
• /
• v.6 no.5
• /
• pp.376-384
• /
• 1997

In this paper, we propose a coding method of remotely sensed satellite image with edge region compensation. This method classifies each pixel vector considering spectral reflection characteristics of satellite image data. For each class, we perform classified intraband VQ and classified interband prediction to remove intraband and interband redundancies, respectively. In edge region case, edge region is compensated using class information of neighboring blocks and gray value of quantized reference bands. Then we perform classified interband prediction using compensated class information to remove interband redundancy, effectively. Experiments on LANDSAT-TM satellite images show that coding efficiency of the proposed method is better than that of the conventional methods.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.38 no.4
• /
• pp.362-370
• /
• 2001

In this paper, we proposed multispectral image compression method using CIP (classified inter-channel prediction) and SVQ (selective vector quantization) in wavelet domain. First, multispectral image is wavelet transformed and classified into one of three classes considering reflection characteristics of the subband with the lowest resolution. Then, for a reference channel which has the highest correlation and the same resolution with other channels, the variable VQ is performed in the classified intra-channel to remove spatial redundancy. For other channels, the CIP is performed to remove spectral redundancy. Finally, the prediction error is reduced by performing SVQ. Experiments are carried out on a multispectral image. The results show that the proposed method reduce the bit rate at higher reconstructed image quality and improve the compression efficiency compared to conventional methods. Index Terms-Multispectral image compression, wavelet transform, classfied inter-channel prediction, selective vetor quantization, subband with lowest resolution.

• The Transactions of the Korea Information Processing Society
• /
• v.4 no.1
• /
• pp.149-160
• /
• 1997

This paper is concerned with the problem of speaker-adaptive speech synthes is method using a mapped codebook designed by fuzzy mapping on FLVQ (Fuzzy Learning Vector Quantization). The FLVQ is used to design both input and reference speaker's codebook. This algorithm is incorporated fuzzy membership function into the LVQ(learning vector quantization) networks. Unlike the LVQ algorithm, this algorithm minimizes the network output errors which are the differences of clas s membership target and actual membership values, and results to minimize the distances between training patterns and competing neurons. Speaker Adaptation in speech synthesis is performed as follow;input speaker's codebook is mapped a reference speaker's codebook in fuzzy concepts. The Fuzzy VQ mapping replaces a codevector preserving its fuzzy membership function. The codevector correspondence histogram is obtained by accumulating the vector correspondence along the DTW optimal path. We use the Fuzzy VQ mapping to design a mapped codebook. The mapped codebook is defined as a linear combination of reference speaker's vectors using each fuzzy histogram as a weighting function with membership values. In adaptive-speech synthesis stage, input speech is fuzzy vector-quantized by the mapped codcbook, and then FCM arithmetic is used to synthesize speech adapted to input speaker. The speaker adaption experiments are carried out using speech of males in their thirties as input speaker's speech, and a female in her twenties as reference speaker's speech. Speeches used in experiments are sentences /anyoung hasim nika/ and /good morning/. As a results of experiments, we obtained a synthesized speech adapted to input speaker.