• Phonetics and Speech Sciences
• /
• v.12 no.2
• /
• pp.29-37
• /
• 2020

Recently, the neural network-based deep learning algorithm has dramatically improved performance compared to the classical Gaussian mixture model based hidden Markov model (GMM-HMM) automatic speech recognition (ASR) system. In addition, researches on end-to-end (E2E) speech recognition systems integrating language modeling and decoding processes have been actively conducted to better utilize the advantages of deep learning techniques. In general, E2E ASR systems consist of multiple layers of encoder-decoder structure with attention. Therefore, E2E ASR systems require data with a large amount of speech-text paired data in order to achieve good performance. Obtaining speech-text paired data requires a lot of human labor and time, and is a high barrier to building E2E ASR system. Therefore, there are previous studies that improve the performance of E2E ASR system using relatively small amount of speech-text paired data, but most studies have been conducted by using only speech-only data or text-only data. In this study, we proposed a semi-supervised training method that enables E2E ASR system to perform well in corpus in different domains by using both speech or text only data. The proposed method works effectively by adapting to different domains, showing good performance in the target domain and not degrading much in the source domain.

• Economic and Environmental Geology
• /
• v.33 no.1
• /
• pp.41-50
• /
• 2000

Geo-radar survey which has the advantage of high-resolution and relatively fast survey has been widely used for engineering and environmental problems. Three-dimensional effects have to be considered in the interpretation of geo-radar for high-resolution. However, there exists a trouble on the analysis of the three dimensional effects. To solve this problem an efficient three dimension numerical modeling algorithm is needed. Numerical radar modeling in three dimensional case requires large memory and long calculating time. In this paper, a finite difference method time domain solution to Maxwell's equations for simulating electromagnetic wave propagation in three dimensional media was developed to make economic algorithm which requires smaller memory and shorter calculating time. And in using boundary condition Liao absorption boundary. The numerical result of cross-hole radar survey for tunnel is compared with real data. The two results are well matched. To prove application to three dimensional analysis, the results with variation of tunnel's incident angle to survey cross-section and the result when the tunnel is parallel to the cross-section were examined. This algorithm is useful in various geo-radar survey and can give basic data to develop dat processing and inversion program.

• Proceedings of the Korean Institute of Building Construction Conference
• /
• 2021.11a
• /
• pp.59-60
• /
• 2021

The unit-water content has a major problem in concrete structures which leads to micro cracks on the concrete during drying time. Thus, the compressive strength and durability of the concrete structures are significantly reduced. Several techniques have been developed to measure the unit-water content in concrete structures such as heating drying, unit volume mass, and capacitance measurements. However, these techniques have problems in during measurement such as longer time, expensive and difficult in analysis of data. Frequency Domain Reflectivity (FDR) is one of the sensors which used to measure the water content. This method has several advantages including easy to measure, inexpensive, and capable of measuring moisture in real time. In this study, an attempt has been made to evaluate the unit-water content in concrete using the FDR sensor and interpret the data with deep learning method.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.41 no.5
• /
• pp.117-124
• /
• 2004

Various multimedia systems have been developed and their application areas widely proliferate. Thus, the interoperability is getting important among various networks and devices. The video transcoding is a technology to solve this interoperability problem among various coding standards. Transcoding can be defined as the conversion of one compressed coded data to another. In this paper, MPEG-2 to MPEG-4 transcoder in the spatial domain is compared with that in the DCT domain. The transcoder is very useful when a video sequence that is originally encoded for digital TV, DVD or satellite broadcasting is served in mobile environment. In order to compare two transcoders, all modules except motion compensation and down sampling are implemented identically. In addition, both transcoders do not search for motion vector. Instead, the decoded information is reused to the encoder. The experimental results show that the transcoder in the spatial domain is usually better than that in the DCT domain with respect to PSNR (Peak Signal-to-Noise Ratio), bitrate and execution time.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.13 no.8
• /
• pp.1497-1504
• /
• 2009

This Paper Propose a method to reduce the calculation time in JPEG2000. That is, if a coefficient is estimate a upper-level subband and its descendents skip the scan process. There is a trade-off relationship between the calculation time and the image quality or the amount of output data, the calculation time and the amount of output data decreases, but the image degradation increases. The experimental results showed that in calculation time was 35% in average, which means that ls ge ses. The ein calculation time and output data can be obtls ed with a cost of an acceptlble image quality degradation.

• Journal of Broadcast Engineering
• /
• v.18 no.2
• /
• pp.311-318
• /
• 2013

This paper deals with a noise reduction algorithm which uses the binary masking in the time-frequency domain. To improve speech intelligibility in noise, noise-masked speech is decomposed into time-frequency units and mask "0" is assigned to masker-dominant region removing time-frequency units where noise is dominant compared to speech. In the previous research, Gaussian mixture models were used to classify the speech-dominant region and noise-dominant region which correspond to mask "1" and mask "0", respectively. In each frequency band, data were collected and trained to build the Gaussian mixture models and detection procedure is performed to the test data where each time-frequency unit belongs to speech-dominant region or noise-dominant region. In this paper, we consider the correlation of masks in the frequency domain and propose a post-processing method which exploits the Viterbi algorithm.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.1
• /
• pp.30-37
• /
• 2022

This paper compares and evaluates model performance from two perspectives according to the learning target and network structure for training Deep Neural Network (DNN)-based speech enhancement models in the frequency domain. In this case, spectrum mapping and Time-Frequency (T-F) masking techniques were used as learning targets, and a real network and a complex network were used for the network structure. The performance of the speech enhancement model was evaluated through two objective evaluation metrics: Perceptual Evaluation of Speech Quality (PESQ) and Short-Time Objective Intelligibility (STOI) depending on the scale of the dataset. Test results show the appropriate size of the training data differs depending on the type of networks and the type of dataset. In addition, they show that, in some cases, using a real network may be a more realistic solution if the number of total parameters is considered because the real network shows relatively higher performance than the complex network depending on the size of the data and the learning target.

• Journal of Power System Engineering
• /
• v.4 no.3
• /
• pp.72-80
• /
• 2000

In order to design a stable robust controller, nominal model, and the upper bound about the uncertainty which is the error of the model are needed. The problem to estimate the nominal model of controlled system and the upper bound of uncertainty at the same time is called robust identification. When the nominal model of controlled system and the upper bound of uncertainty in relation to robust identification are given, the evaluation of the validity of the model and the upper bound makes it possible to distinguish whether there is a model which explains observation data including disturbance among the model set. This paper suggests a method to identity the uncertainty which removes disturbance and expounds observation data by giving a probable postulation and plural data set to disturbance. It also examines the suggested method through a numerical computation simulation and validates its effectiveness.

• Communications for Statistical Applications and Methods
• /
• v.10 no.3
• /
• pp.955-969
• /
• 2003

The classical Fourier analysis, which is the typical frequency domain approach, is used to detect periodic trends that are of the sinusoidal shape in time series data. In this article, using a sequence of periodic step functions, describes an adaptive Fourier series where the patterns may take general periodic shapes that include sinusoidal as a special case. The results, which extend both Fourier analysis and Walsh-Fourier analysis, are applies to investigate the shape of the periodic component. Through the real data, compare the goodness-of-fit of the model using two methods, the adaptive Fourier method which is proposed method in this paper and classical Fourier method.

• Journal of Electrical Engineering and Technology
• /
• v.8 no.5
• /
• pp.1002-1011
• /
• 2013

This paper proposes the new adaptive protection algorithm for inverse-time overcurrent relays (OCRs) to ensure their proper operating time and protective coordination. The application of the proposed algorithm requires digital protection relays with microcontroller and memory. The operating parameters of digital OCRs are adjusted based on the available data whenever system conditions (system with distributed generation (DG)) vary. Moreover, it can reduce the calculation time required to determine the operating parameters for achieving its purpose. To verify its effectiveness, several case studies are performed in time-domain simulation. The results show that the proposed adaptive protection algorithm can keep the proper operating time and provide the protective coordination time interval with fast response.