• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1
• /
• pp.24-31
• /
• 1994

In this paper, we report on the development of a speaker independent continuous speech recognition system using continuous hidden Markov models. The continuous hidden Markov model consists of mean and covariance matrices and directly models speech signal parameters, therefore does not have quantization error. Filter bank coefficients with their 1st and 2nd-order derivatives are used as feature vectors to represent the dynamic features of speech signal. We use the segmental K-means algorithm as a training algorithm and triphone as a recognition unit to alleviate performance degradation due to coarticulation problems critical in continuous speech recognition. Also, we use the one-pass search algorithm that Is advantageous in speeding-up the recognition time. Experimental results show that the system attains the recognition accuracy of $83\%$ without grammar and $94\%$ with finite state networks in speaker-indepdent speech recognition.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38B no.4
• /
• pp.305-313
• /
• 2013

As an FMT (Filtered Multi-Tone) transmission method of Wideband VHF communication system specified by the ETS (European Telecommunications Standards) EN 300 392-2, this paper introduces three existing realization methods, i.e., the direct filtering method using different band SRRC (Square-Root Raised Cosine) filters for each subcarrier, the PPN-DFT method using the IDFT-PPN (Poly-Phase Network) and PPN-DFT at the transmitter and receiver, respectively, and the Extended DFT method. Then, it proposes the extended IDFT-SDFT (Sliding Discrete Fourier Transform) that computes the DFT values only for interested subcarriers every sample time, and shows that it has an advantage of blind symbol timing (using no training symbol) individually for each user signal (independently of other users' signals) in the multi-user environment where the subcarriers are assigned in contiguous or interleaved blocks to each user and each user signal possibly experiences different channels.

• Journal of the Institute of Convergence Signal Processing
• /
• v.2 no.2
• /
• pp.1-6
• /
• 2001

Eigenface method in face recognition is useful due to its insensitivity to large variations in facial expression and facial details. However its low recognition rate necessitates additional researches. In this paper, we present an efficient method for improving the recognition rate in face recognition using eigenface feature. For this, we performs a comparative study of three different classifiers which are i) a single prototype (SP) classifier, ii) a nearest neighbor (NN) classifier, and iii) a standard feedforward neural network (FNN) classifier. By evaluating and analyzing the performance of these three classifiers, we shows that the distribution of eigenface features of face image is not compact and that selections of classifier and sample training data are important for obtaining higher recognition rate. Our experiments with the ORL face database show that 1-NN classifier outperforms the SP and FNN classifiers. We have achieved a recognition rate of 91.0% by selecting sample trainging data properly and using 1-NN classifier.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.25 no.1A
• /
• pp.156-166
• /
• 2000

In discrete multi-tone (DMT)-based very high bit-rate digital subscriber line (VDSL) systems, the ingressed RFI (Radio Frequency Interference) accompanied by transmitted signal at the receiver is known to cause the spectralleakage by the finite-point FFT, resulting in significant performance degradation.0 this paper, we propose a RFIcancellation technique which can compensate the ingressed RFI efficiently, especially for a high data-rate VDSLsystem. The proposed technique compensates the performance degradation of e VDSL system due to RFI byusing a time-domain RFI canceller whose coefficients are obtained from the estimated center frequency of RFI inthe frequency domain under the assumption that the ingressed RFI is a narrow-band signal compared to VDSLsampling frequency. The proposed technique requires no training symbol and convergence period, and worksproperly even when spectral shape of the ingressed RFI is unknown or arbitrary. Feasibility of the proposedtechnique is demonstrated via a computer simulation by comparing its performance with the performance of theprevious RFI cancellation technique.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.5
• /
• pp.447-453
• /
• 2020

In this paper, we propose a system to extract effective speaker representations from a speech signal using a deep learning method. Based on the fact that speech signal contains identity unrelated information such as text content, emotion, background noise, and so on, we perform a training such that the extracted features only represent speaker-related information but do not represent speaker-unrelated information. Specifically, we propose an auto-encoder based disentanglement method that outputs both speaker-related and speaker-unrelated embeddings using effective loss functions. To further improve the reconstruction performance in the decoding process, we also introduce a discriminator popularly used in Generative Adversarial Network (GAN) structure. Since improving the decoding capability is helpful for preserving speaker information and disentanglement, it results in the improvement of speaker verification performance. Experimental results demonstrate the effectiveness of our proposed method by improving Equal Error Rate (EER) on benchmark dataset, Voxceleb1.

• Hwankyungkyoyuk
• /
• v.18 no.1 s.26
• /
• pp.55-69
• /
• 2005

This study aims to investigate and practically examine the effect of environmental qualifications based on the theoretical background on the area. First, the economic effect of the acquisition of the qualifications was to be studied from the viewpoints of individuals and from an actual analysis on it, its effect was to be proved. Second, its non-economic effect was to be proved from the same way as the first analysis. On the basis of theoretical background, a study model was formulated in a way that the effect of qualifications in individuals side was divided into an economic effect (wage, employment, promotion, job-switching) and non-economic effect (self-development, self-efficiency, satisfaction on the job, ability to cope with the advancement of technology, job performance, signal effect, the settlement of uneasiness at unemployment). Then, survey was carried out with questions designed in accordance with this model. The hypotheses were proved as the following. First, for hypothesis 1(Environmental qualifications will bring up positive impacts on an individuals economic effect), environmental qualifications was shown to have positive impacts on wage, job-switching in personal economic effect. But, no personal economic effect appeared for employment and promotion. Second, for hypothesis 2(Environmental qualifications will bring up positive impacts on an individuals non-economic effect), environmental qualifications appeared to have positive impacts on self-efficiency, ability to cope with the advancement of technology, job performance and signal effect. Besides, no impact was shown in satisfaction on the job and the settlement of uneasiness at unemployment and self-development.

• Journal of the Institute of Convergence Signal Processing
• /
• v.5 no.4
• /
• pp.333-337
• /
• 2004

This paper describe a design of 5GHz OFDM baseband chip for IEEE 802.11a wireless LAN. The proposed device is consists of transmitter and receiver within a single FPGA chip. We applied single tap equalizer that use Normalized LMS algorithm to remove ISI that happen at high speed data transmission. And also, we used carrier wave frequency offset algorithm that use training symbol to remove ICI. The simulation results show the correct transmission without errors the between transmitter and receiver And we can remarkably reduce the number of register through the synthesized circuits by using DSP block and EMB(Embedded Memory Block). The target device for implementation of the synthesized circuits is Altera Stratix EPIS25FC672 FPGA and design platform is VHDL.

• The Journal of the Acoustical Society of Korea
• /
• v.31 no.1
• /
• pp.1-10
• /
• 2012

The linear and decision-feedback equalization can mitigate time-varying intersymbol interference (ISI) caused by time-varying multipath propagation for underwater acoustic channels. The perfect elimination of interference components, however, is difficult using the linear equalization and the decision feedback equalizer has an error propagation problem. To overcome these shortcomings, this paper proposes an equalizer mode selection method using training sequences. The proposed method selects an equalization mode corresponding to the signal-to-noise ratio (SNR). If the SNR is low, the proposed system operates the linear equalizer for preventing the error propagation and if the SNR is high, the decision feedback equalizer for eliminating the residual ISI. Therefore, the proposed method can improve the error performance compared to the conventional equalizers. The computer simulation shows the proposed method improves the bit error performance using practical underwater channels responses acquired from the sea experiment.

• Journal of the Korea Convergence Society
• /
• v.11 no.3
• /
• pp.19-26
• /
• 2020

In this paper, we propose a method to analyze the loss region of the dictionary-based super resolution result learned for image quality improvement and to map the learning data according to the analyzed loss region. In the conventional learned dictionary-based method, a result different from the feature configuration of the input image may be generated according to the learning image, and an unintended artifact may occur. The proposed method estimate loss information of low resolution images by analyzing the reconstructed contents to reduce inconsistent feature composition and unintended artifacts in the example-based super resolution process. By mapping the training data according to the final interpolation feature map, which improves the noise and pixel imbalance of the estimated loss information using a Gaussian-based kernel, it generates super resolution with improved noise, artifacts, and staircase compared to the existing super resolution. For the evaluation, the results of the existing super resolution generation algorithms and the proposed method are compared with the high-definition image, which is 4% better in the PSNR (Peak Signal to Noise Ratio) and 3% in the SSIM (Structural SIMilarity Index).

• The Transactions of the Korean Institute of Electrical Engineers D
• /
• v.55 no.11
• /
• pp.502-510
• /
• 2006

In this paper, we describe implementation of a computer access device for the severly motor-disability. Many people with severe motor disabilities need an augmentative communication technology. Those who are totally paralyzed, or 'locked-in' cannot use conventional augmentative technologies, all of which require some measure of muscle control. The forehead is often the last site to suffer degradation in cases of severe disability and degenerative disease. For example, In ALS(Amyotrophic Lateral Sclerosis) and MD(Muscular dystrophy) the ocular motorneurons and ocular muscles are usually spared permitting at least gross eye movements, but not precise eye pointing. We use brain and body forehead bio-potentials in a novel way to generate multiple signals for computer control inputs. A bio-amplifier within this device separates the forehead signal into three frequency channels. The lowest channel is responsive to bio-potentials resulting from an eye motion, and second channel is the band pass derived between 0.5 and 45Hz, falling within the accepted Electroencephalographic(EEG) range. A digital processing station subdivides this region into eleven components frequency bands using FFT algorithm. The third channel is defined as an Electromyographic(EMG) signal. It responds to contractions of facial muscles and is well suited to discrete on/off switch closures, keyboard commands. These signals are transmitted to a PC that analyzes in a time series and a frequency region and discriminates user's intentions. That software graphically displays user's bio-potential signals in the real time, therefore user can see their own bio-potentials and control their physiological signals little by little after some training sessions. As a result, we confirmed the performance and availability of the developed system with experimental user's bio-potentials.