• 말소리와 음성과학
• /
• 제6권2호
• /
• pp.29-34
• /
• 2014

In the Vector Taylor Series (VTS)-based noisy speech recognition methods, Hidden Markov Models (HMM) are usually trained with clean speech. However, better performance is expected by training the HMM with noisy speech. In a previous study, we could find that Minimum Mean Square Error (MMSE) estimation of the training noisy speech in the log-spectrum domain produce improved recognition results, but since the proposed algorithm was done in the log-spectrum domain, it could not be used for the HMM adaptation. In this paper, we modify the previous algorithm to derive a novel mathematical relation between test and training noisy speech in the cepstrum domain and the mean and covariance of the Multi-condition TRaining (MTR) trained noisy speech HMM are adapted. In the noisy speech recognition experiments on the Aurora 2 database, the proposed method produced 10.6% of relative improvement in Word Error Rates (WERs) over the MTR method while the previous MMSE estimation of the training noisy speech produced 4.3% of relative improvement, which shows the superiority of the proposed method.

• 대한전기학회:학술대회논문집
• /
• 대한전기학회 2007년도 심포지엄 논문집 정보 및 제어부문
• /
• pp.149-150
• /
• 2007

In this paper, we propose a novel defects inspection algorithm for TFT-LCD panels. We first compensate the distorted image caused by the camera distortion and the uneven illumination environment using the least squares method and the bezier surface. We find a starting point of each pattern. The reference frame, made by subtract method using several clean patterns, is compared to each pattern to find defects. The simulation example shows that our algorithm not only inspects the defects well, but also is robust to the 1-pixel error.

• 음성과학
• /
• 제1권
• /
• pp.277-283
• /
• 1997

In this paper, we present a robust recognition algorithm based on the Wiener filtering method as a research tool to develop the Korean Speech recognition system. We especially used Wiener filtering method in cepstrum-domain, because the method in frequency-domain is computationally expensive and complex. Evaluation of the effectiveness of this method has been conducted in speaker-independent isolated Korean digit recognition tasks using discrete HMM speech recognition systems. In these tasks, we used 12th order weighted cepstral as a feature vector and added computer simulated white gaussian noise of different levels to clean speech signals for recognition experiments under noisy conditions. Experimental results show that the presented algorithm can provide an improvement in recognition of as much as from $5\%\;to\;\20\%$ in comparison to spectral subtraction method.

• 한국CDE학회논문집
• /
• 제7권3호
• /
• pp.184-188
• /
• 2002

In 2D-Profile machining using cutter radius compensation cutter interferences are very common. To prevent the cutter interferences undercuts are inevitable in some regions of the profile. The undercut regions require cleanup machining using smaller radius tools. This paper considers a procedure of the tool path generation for the cleanup profile machining. And two methods are introduced for an efficient tool path generation. One is how to reduce the machining time by uniting adjacent tool paths of undercut regions, and the other is how to find the tool path with the minimal distance by applying TSP algorithm.

• KIEE International Transaction on Electrical Machinery and Energy Conversion Systems
• /
• 제3B권3호
• /
• pp.139-146
• /
• 2003

In this paper, the steady-state operating performance analysis for the three-phase squirrel cage rotor self-excited induction generator (SEIG) driven by a variable-speed prime mover (VSPM) in addition to a constant-speed prime mover (CSPM) is presented on the basis of an effective algorithm based on its frequency-domain equivalent circuit. The operating characteristics of the three-phase SEIG coupled by a VSPM and/or a CSPM are evaluated on line processing under the condition of the electrical passive load parameters variations with simple and efficient computation processing procedure in unregulated voltage control loop scheme. A three-phase SEIG prototype setup with a VSPM as well as a CSPM is implemented for the small-scale clean renewable and alternative energy utilizations. The experimental operating characteristic results are illustrated and give good agreements with the simulation ones.

• 한국전자파학회논문지
• /
• 제24권9호
• /
• pp.928-935
• /
• 2013

본 논문에서는 레이더 표적 인식 성능을 향상시키기 위한 방법으로 특성 벡터 융합 기법을 제안하였다. 제안하는 방법은 두 개의 수신기로 입력되는 신호로부터 추출된 특성 벡터를 서로 융합해서 사용함으로써 표적에 대해 더 많은 정보를 획득할 수 있는 장점을 가지고 있다. 제안하는 방법의 성능을 검증하기 위해 먼저, 세 가지의 서로 다른 전투기의 실스케일 캐드 모델들에 대해 모노스태틱 및 바이스태틱 RCS(Radar Cross Section)를 계산하였다. 계산된 RCS로부터 표적의 특성 벡터인 산란점 정보를 추출하기 위해 시간 영역의 1차원 FFT(Fast Fourier Transform) 기반의 CLEAN 알고리즘을 이용하였다. 추출된 특성 벡터는 신경망 구분기의 입력으로 사용되어 표적 구분 실험을 수행한 결과, 제안하는 방법이 모노스태틱 및 바이스태틱 특성 벡터를 따로 사용했을 때보다 표적 인식 성능을 향상시킬 수 있음을 확인하였다.

• 한국멀티미디어학회논문지
• /
• 제23권11호
• /
• pp.1349-1360
• /
• 2020

In recent years, Convolutional Neural Networks (CNNs) have been successfully implemented in different tasks of computer vision. Since CNN models are the representatives of supervised learning algorithms, they demand large amount of data in order to train the classifiers. Thus, obtaining data with correct labels is imperative to attain the state-of-the-art performance of the CNN models. However, labelling datasets is quite tedious and expensive process, therefore real-life datasets often exhibit incorrect labels. Although the issue of poorly labelled datasets has been studied before, we have noticed that the methods are very complex and hard to reproduce. Therefore, in this research work, we propose Deep CleanNet - a considerably simple system that achieves competitive results when compared to the existing methods. We use K-means clustering algorithm for selecting data with correct labels and train the new dataset using a deep CNN model. The technique achieves competitive results in both training and validation stages. We conducted experiments using MNIST database of handwritten digits with 50% corrupted labels and achieved up to 10 and 20% increase in training and validation sets accuracy scores, respectively.

• 한국통신학회논문지
• /
• 제39A권2호
• /
• pp.109-116
• /
• 2014

본 논문에서는 기상 레이다의 효율적 주파수 사용을 위해 직교 파형과 간섭 제거 기술을 이용한 주파수 재사용 기술을 제안하였다. 주파수 공유로 발생한 레이다간 간섭 영향을 줄이기 위해 직교 코드 시퀀스 간 상호 상관 최대값이 가장 작은 조합으로 선택하였으며, 잔여 간섭 성분 제거를 위해 CLEAN 알고리즘을 사용하였다. 전산 모의 실험 결과 재사용 기술 적용 후에도 성능 요구 조건을 만족하는 것으로 나타남으로써 현행 S 밴드 기상 레이다의 점유 주파수 8개를 1개로 줄일 수 있어 신규 주파수 확보 가능성 제시하였다.

• 한국전자통신학회논문지
• /
• 제14권5호
• /
• pp.811-816
• /
• 2019

본 논문에서는 잡음이 포함된 4채널의 음원신호를 주파수 대역의 독립벡터분석 알고리즘에 의하여 깨끗한 음성신호와 혼합잡음신호를 분리하는 기법을 먼저 제안한다. 제안한 독립벡터분석 알고리즘에 의하여 분리된 음원신호를 주파수대역 지연합 빔형성기로부터 출력되는 신호와 독립벡터분석으로부터 분리된 출력신호 간의 상호 상관성을 이용하여 향상된 출력음성신호를 구한다. 본 실험에서는 백색잡음이 포함된 0dB, -5dB의 SNR의 입력 혼합잡음음성에 대하여, 본 논문에서 제안하고 있는 알고리즘이 주파수대역 지연합 빔형성기 알고리즘만을 사용하였을 때 보다 최대 10.90dB의 SNR 및 10.02dB의 Segmental SNR이 개선되었음을 확인하였다. 따라서 본 논문의 알고리즘 기법이 주파수대역 지연합 빔형성기와 비교하여 음성품질이 향상된 것을 실험 및 고찰을 통하여 확인할 수 있었다.

• 전력전자학회논문지
• /
• 제21권6호
• /
• pp.514-524
• /
• 2016

This work suggests a study on 5 kW plasma power supply design and reactor capacitance estimation algorithm for a wide range of linear output control to operate a plasma reactor. The suggested study is designed to use a two-stage circuit and control the full-bridge circuit of the two-stage circuit using the buck converter output voltage of the single-stage circuit. The switching frequency of the full-bridge circuit is designed to operate through high-frequency switching and obtain maximum output using LC parallel resonance. Soft switching technique(ZVS) is used to reduce the loss caused by high-frequency switching, and duty control of the buck converter is applied to control a wide range of linear output. The internal capacitance of the reactor cannot easily be extracted, and thus, the reactor cannot be operated in an optimized resonant state. To address this issue, this work designs the internal capacitance of the reactor such that estimations can be performed with the developed reactor capacitance estimation algorithm applied to the internal capacitance of the reactor. A 5 kW plasma power supply is designed for a wide range of linear output control, and the validity of the study on the reactor capacitance estimation algorithm is verified.