• Proceedings of the Korea Multimedia Society Conference
• /
• 2001.11a
• /
• pp.85-90
• /
• 2001

본 논문에서는 한국어 숫자음 인식을 위한 시간 지연 신경망(Time delay neural network-TDNN)과 은닉 마르코프 모델(Midden Markov Model-HMM)의 결합 방법에 대해서 연구하였고 그 성능을 측정하였으며, 기존의 시스템과 비교 평가하였다. 이 알고리즘은 TDNN과 HMM의 구조적인 결합에 기반하고 있는데 TDNN의 두번째 은닉층의 출력이 HMM의 입력으로 들어가도록 구성되었다. 그러면 HMM은 TDNN의 출력으로 각 단어에 대해서 훈련과정을 거치게 된다. 이렇게 구성된 인식알고리즘은 TDNN의 뛰어난 단기간(Short-time)분류 기능과 HMM의 시간 정렬(time-warping) 능력을 동시에 갖게 된다. 위의 과정을 컴퓨터 시뮬레이션을 이용하여 구현하였으며, 한사람의 음성을 녹음하여 실험한 결과 기존의 TDNN만으로 만들어진 인식기보다는 3%, HMM만으로 구성된 인식기 보다는 5.7% 나은 성능을 얻을 수 있었다.

• The Korean Journal of Applied Statistics
• /
• v.33 no.6
• /
• pp.791-804
• /
• 2020

Residential electricity consumption can be predicted more accurately by utilizing the realtime household electricity consumption reference that can be collected by the AMI as the ICT developed under the smart grid circumstance. This paper studied the model that predicts residential power load using the ARIMA, TBATS, NNAR model based on the data of hour unit amount of household electricity consumption, and unlike forecasting the consumption of the whole households at once, it computed the anticipated amount of the electricity consumption by aggregating the predictive value of each established model of cluster that was collected by the households which show the similiar load profile. Especially, as the typical time series data, the electricity consumption data chose the clustering analysis method that is appropriate to the time series data. Therefore, Dynamic Time Warping and Periodogram based method is used in this paper. By the result, forecasting the residential elecrtricity consumption by clustering the similiar household showed better performance than forecasting at once and in summertime, NNAR model performed best, and in wintertime, it was TBATS model. Lastly, clustering method showed most improvements in forecasting capability when the DTW method that was manifested the difference between the patterns of each cluster was used.

• Korean Journal of Ecology and Environment
• /
• v.54 no.4
• /
• pp.390-397
• /
• 2021

The time series data of rotifer community focusing on the species number and total density were collected from 29 reservoirs located at Jeonnam Province from 2008 to 2016 quarterly. The reservoirs had similar weather condition during the study period, but their sizes and water qualities were different. To analyze the temporal dynamics of rotifer community, the medians, ranges, outliers and coefficient of variation (CV) value of rotifer species number and abundance were compared. For the temporal trend analysis, time series of each reservoir data were compared and clustered using the dynamic time warping function of the R package "dtwclust". Small-sized reservoirs showed higher variability in rotifer abundance with more frequent outliers than large-sized reservoirs. On the other hand, apparent pattern was not observed for the rotifer species number. For the temporal pattern of rotifer density, COD, phytoplankton abundance fluctuation, and cladoceran abundance fluctuation have been suggested as potential factor affecting the rotifer abundance dynamics.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 1993.04a
• /
• pp.275-281
• /
• 1993

본 발표에서는 신경회로망을 이용하여 연속음성중에서 키워드를 인식하는 방법을 설명한다. 연속음성에서 파형소편 및 음절을 식별하는 휴리스틱 알고리즘을 개발하였고, 연속음성을 음절단위로 파형소편 스펙트럼분석(선형예측법)으로 특성치를 추출하였다. 음절의 특성치는 코호넨 신경회로망을 통하여 학습을 시켰으며, 연속음성중 키워드인식은 먼저 음절을 인식하여 단어를 찾고, 인식된 단어가 키워드와 일치하는가를 확인한다. 본 연구의 의의는 파형소편 및 음절식별 알고리즘을 통하여, 크기불변성(Scaling invariance), 시간불변성(Time warping 및 Time-shift invariance), 중복성제거의 문제점을 해결하였고, 신경회로망의 학습을 통하여 화자독립적인 연속음성인식시스템 구축의 기반을 확립한데 있다. 본 음성인식모델은 학교구내 전화번호 안내시스템으로 활용단계에 있으며 전화번호뿐만아니라 주소안내시스템으로도 활용될 예정이다. 또한 자동차 운전보조시스템 및 주행안내시스템의 음성명령에 응용될 수 있는데, 예로 음성명령은 "핸들 좌로 20도", "시청까지 주행", "시청 지도안내"등이 될 수 있다. 현재 자동차 운전보조시스템은 컴퓨터 화면상 모의동작시스템으로 운영되고 있다. 본 음성인식모델은 화자종속시 90%이상, 화자독립시 70%의 인식결과를 보였다.시 90%이상, 화자독립시 70%의 인식결과를 보였다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2003.04c
• /
• pp.241-243
• /
• 2003

일반적으로 사용되는 반도체 공정에 대한 진단 기법은 한 공정을 진행하기 전에 테스트 공정을 수행하여 공정의 진행 여부를 결정하고, 한 공정의 진행을 완료한 후에 다시 테스트 공정을 수행하여 공정의 결과를 진단하는 방법이다. 본 논문에서 제안하는 실시간 자동 진단 시스템은 기존 방법의 문제점인 자원의 낭비를 막고, 실시간으로 진단함으로써 시간의 낭비를 막는 진단 시스템을 제안한다. 실시간 자동 진단 시스템은 크게 시스템 초기화 단계, 학습 단계 그리고 예측 단계로 나누어진다. 초기화 단계는 진단할 공정에 대한 사전 입력값을 받아 시스템을 초기화하는 과정으로 공정장비 파라미터별 중요도 자동 설정 과정과 초기화 클러스터링으로 이루어진다. 학습 단계는 실시간으로 저장된 공정장치별 데이터와 계측기로부터 획득된 데이터를 이용하여 최적의 유사 클래스를 결정하는 단계와 결정된 유사 클래스를 이용하여 가중치를 학습하는 단계로 나누어진다. 예측 단계는 공정 진행 중 획득된 실시간 데이터를 학습 단계에서 결정된 파라미터별 가중치를 사용하여 공정에 대한 진단을 한다. 본 시스템에서 사용하는 클러스터링 알고리즘은 DTW(Dynamic Time Warping)를 이용하여 파라미터 데이터에 대한 특징을 추출하고 LBG(Linde, Buzo and Gray) 알고리즘을 사용하여 데이터를 군집화 한다.

• Speech Sciences
• /
• v.14 no.4
• /
• pp.157-167
• /
• 2007

In this paper, we implemented a single-chip speech recognizer using the TMS320C2000 DSPs. For this implementation, we had developed very small-sized speaker-dependent recognition engine based on dynamic time warping, which is especially suited for embedded systems where the system resources are severely limited. We carried out some optimizations including speed optimization by programming time-critical functions in assembly language, and code size optimization and effective memory allocation. For the TMS320F2801 DSP which has 12Kbyte SRAM and 32Kbyte flash ROM, the recognizer developed can recognize 10 commands. For the TMS320F2808 DSP which has 36Kbyte SRAM and 128Kbyte flash ROM, it has additional capability of outputting the speech sound corresponding to the recognition result. The speech sounds for response, which are captured when the user trains commands, are encoded using ADPCM and saved on flash ROM. The single-chip recognizer needs few parts except for a DSP itself and an OP amp for amplifying microphone output and anti-aliasing. Therefore, this recognizer may play a similar role to dedicated speech recognition chips.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.14 no.2
• /
• pp.201-206
• /
• 2004

This study attempted to develop a speech recognition module applied to a wheelchair for the physically handicapped. In the proposed speech recognition module, TMS320C32 was used as a main processor and Mel-Cepstrum 12 Order was applied to the pro-processor step to increase the recognition rate in a noisy environment. DTW (Dynamic Time Warping) was used and proven to be excellent output for the speaker-dependent recognition part. In order to utilize this algorithm more effectively, the reference data was compressed to 1/12 using vector quantization so as to decrease memory. In this paper, the necessary diverse technology (End-point detection, DMA processing, etc.) was managed so as to utilize the speech recognition system in real time

• Proceedings of the Korean Society For Composite Materials Conference
• /
• 2004.10a
• /
• pp.282-285
• /
• 2004

The main purpose of this study is the vibration suppression of rotating composite blade containing distributed piezoelectric sensors and actuators. The blade is modeled by thin-walled, single cell composite beam including the warping function, centrifugal force, Coriolis acceleration and piezoelectric effect. Further, the numerical study is performed m ing finite element method. The vibration of composite rotor is suppressed by piezocomposite actuators and PVDF sensors that are embedded between composite layers. A velocity feedback control algorithm coupling the direct and converse piezoelectric effect is used to actively control the' dynamic response of an integrated structure through a closed control loop. Responses of the rotating blade are investigated. Newmark time integration method is used to calculate the time response of the model. In the numerical simulation, the effect of parameters such as rotating speed, fiber orientation of the blade and size of actuators are studied in detail.

• ETRI Journal
• /
• v.29 no.5
• /
• pp.655-666
• /
• 2007

There are nodes and edges in a topological map. Node data has been used as a main source of information for the localization of mobile robots. In contrast, edge data is regarded as a minor source of information, and it has been used in an intuitive and heuristic way. However, edge data also can be used as a good source of information and provide a way to use edge data efficiently. For that purpose, we define a data format which describes the shape of an edge. This format is called local generalized Voronoi graph's angle (LGA). However, the LGA is constituted of too many samples; therefore, real time localization cannot be performed. To reduce the number of samples, we propose a compression method which utilizes wavelet transformation. This method abstracts the LGA by key factors using far fewer samples than the LGA. Experiments show that the LGA accurately describes the shape of the edges and that the key factors preserve most information of the LGA while reducing the number of samples.

• Journal of Electrical Engineering and Technology
• /
• v.9 no.3
• /
• pp.1016-1022
• /
• 2014

Kinect is a motion sensing camera released by Microsoft in November 2010 for the Xbox360 that is used to produce depth and color images. Because Kinect uses an infrared pattern, it generates holes and noises around an object's boundaries in the obtained images. The flickering phenomenon and unmatched edges also occur. In this paper, we propose a real time virtual-view video synthesis algorithm which results in a high quality virtual view by solving these problems stated above. The experimental results show that the proposed algorithm performs much better than the conventional algorithms.