음소기반 인식 네트워크에서의 비인식 대상 문장 거부 기능의 비교 연구

김형태,하진영,Kim, Hyung-Tai,Ha, Jin-Young 대한음성학회 2006 말소리 Vol.59 No.-

For speech recognition systems, rejection function as well as decoding function is necessary to improve the reliability. There have been many research efforts on out-of-vocabulary word rejection, however, little attention has been paid on non-target sentence rejection. Recently pronunciation approaches using speech recognition increase the need for non-target sentence rejection to provide more accurate and robust results. In this paper, we proposed filler model method and word/phoneme detection ratio method to implement non-target sentence rejection system. We made performance evaluation of filler model along to word-level, phoneme-level, and sentence-level filler models respectively. We also perform the similar experiment using word-level and phoneme-level word/phoneme detection ratio method. For the performance evaluation, the minimized average of FAR and FRR is used for comparing the effectiveness of each method along with the number of words of given sentences. From the experimental results, we got to know that word-level method outperforms the other methods, and word-level filler mode shows slightly better results than that of word detection ratio method.

문장 거부를 위한 음소기반 인식 네트워크에서의 필러 모델 비율과 단어 검출률의 성능비교

김형태(Hyung-Tai Kim),이병혁(Byung-Hyuk Lee),하진영(Jin-Young Ha) 한국정보과학회 2005 한국정보과학회 학술발표논문집 Vol.32 No.1

음성인식 시스템에서 입력된 음성 데이터에 대해 비인식 대상을 거부하는 기능은 신뢰도 보장 측면에 있어서 상당히 중요하며, 신뢰도를 높이기 위해서는 단순한 인식기능 외에 부적절한 입력 패턴의 거부 기능이 필요하다. 본 논문에서는 이러한 신뢰성 문제를 해결하기 위하여 음소기반 인식 네트워크에서 필러 모델 방법과 단어 검출률 방법을 사용하여 실험하였고, 문장의 단어 수에 따른 두 방법의 문장 거부 성능을 FAR과 FRR의 평균을 최소화 하는 값을 각각 구함으로써 비교·분석 하였다. 그 결과 필러모델 방법이 좀 더 나은 거부 성능을 보였고, 단어 검출률을 이용하는 방법이 인식 네트워크를 전부 거치지 않아도 되므로 실행속도와 메모리 절약에서 효과적이었다.

온라인 필기 한자인식을 위한 체인코드열과 구조코드열의 성능평가

김형태(Hyung-Tai Kim),하진영(Jin-Young Ha) 한국정보과학회 2006 한국정보과학회 학술발표논문집 Vol.33 No.2B

본 논문에서는 보다 실용적인 온라인 한자인식기 개발을 위하여 한자 검정 능력 1급 쓰기 수준을 모두 포함하는 한자 필기 데이터로부터 16방향의 체인코드열과 부분획의 구조를 반영하는 구조코드열을 만들어 성능평가를 하였다. 인식 방법으로는 DP 매칭 방법과 HMM을 사용하여 2,362 종류의 한자에 대해 인식 실험을 하였다. 그 결과 체인코드열을 사용한 DP 매칭 방법에 의한 결과가 96.54%로 가장 높은 인식률을 보였으며, 구조코드열을 사용하여 HMM에 의한 인식실험 결과가 95.65%로 그 뒤를 이었다. 인식 속도면에서는 체인코드 보다 코드열의 길이가 짧은 구조코드열을 사용한 방법이 상대적으로 유리했고, 클래스 당 1개의 모델을 사용한 HMM에 의한 방법이 클래스 당 복수개의 모델을 사용한 DP 매칭 방법에 비해 모델의 개수가 훨씬 적기 때문에 속도 면에서 월등히 유리해 더 효율적인 인식 성능을 보인다는 결론을 내릴 수 있었다.

직주균형이 통근통행에 미치는 영향(1990-2005)

김형태(Kim Hyung-Tai) 대한국토·도시계획학회 2009 國土計劃 Vol.44 No.7

Commuting trips that link workplace and residence which are the basic elements of the urban spatial structure have been known to cause diseconomies of agglomeration, such as heavy traffic and environmental problems, due to spatial and rush hour concentration. This study addresses such issues as improving the efficiency of commuting, the main cause of traffic jams in the Capital Region, mainly from the urban spatial perspective of urban economics. To that end, the study takes advantage of the 1990, 1995, 2000, and 2005 census data of the Capital Region to see whether policy interventions such as jobs-housing balance can work as an effective transportation policy, through regressing commuting distance by land use such as jobs-housing balance after controling for travel mode and socio-economic characteristics of an individual or household. The finding shows that the commuting distance has continued to grow while the commuting time has continued to be reduced between 1990 and 2005, which means that jobs-housing mismatch has worsened while mobility has improved. The finding also shows that policies to focus on qualitative, rather than quantitative, jobs-housing balance would be more useful. The regression results show that policies to provide employment centers with more houses could be more useful tools to reduce commuting distance than those to provide residential areas with more job opportunities.

한국 정상 젊은 남성에서의 제 4-5 중족골간 각 및 제 5 근위-종족골간 각

이경태,이영구,임수재,양기원,박신이,김형태,Lee, Kyung-Tai,Lee, Young-Koo,Yim, Su-Jae,Young, Ki-Won,Park, Shin-Yi,Kim, Hyung-Tae 대한족부족관절학회 2009 대한족부족관절학회지 Vol.13 No.1

Purpose: The purpose of this study is to find out the normal angles of forefoot for diagnosis of bunionette deformity in Korea. This would be helpful as it would provide a basic angular measurement of bunionette deformity that indicates the need for operation. Materials and Methods: Within a period of four months from January 2007 to April 2007. We have established 300 males and 600 feet without trauma history of foot. The source to image distance is 40 inches and erect weight bearing radiographs are obtained in anteroposterior and lateral projections. The significant angular measurements that define a bunionette deformity are the fourth-fifth intermetatarsal angle, the fifth metatarsophalangeal angle Results: The mean age was 21 years(19-22 years) old. The mean $4^{th}-5^{th}$ intermetatarsal angle was $9.4{\pm}4.7^{\circ}$, the mean $5^{th}$ metatarsophalangeal angle was $7.9{\pm}9.2^{\circ}$. Conclusion: We can consider over the $14.1^{\circ}$ on the $4^{th}-5^{th}$ intermetatarsal angle and over $17.1^{\circ}$ on the $5^{th}$ metatarsophalangeal angle as an operational indication.

자가보유 전․월세 가구의 특성과 주거이동성에 관한 연구

마강래(Ma, Kang-Rae),강은택(Kang, Eun-Taek),김형태(Kim, Hyung-tai) 한국지역개발학회 2012 韓國地域開發學會誌 Vol.24 No.1

Over the past few decades, there has been a considerable increase in the number of home-owning households who reside in the rental housing. It has frequently been mentioned in the previous research that the residential mobility of households is associated with the housing tenure. Moving costs increase greatly for the households who own housing, so they are trying to overcome the mobility restrictions by residing in the rental house while owning their home in other places. The main purpose of this study is to investigate the household characteristics that could affect the residential choice behaviors of home-owning renters. Both the 2% sample of 2005 Korean Census of Population and Housing data and 10 wave Korean Labor and Income Panel Survey were used to investigate the socio-economic characteristics of home-owning renters. The empirical evidence revealed that factors such as the number of household member, age, education, marital status, occupation, etc. have a significant impact on the transition to the tenure status of homeowners. It was also found that the transition to rental housing is under the influence of commuting modes, commuting time, dual-income family, and relocation experience.

필러 모델 비율과 단어 누락률을 이용한 영어 문장 거부

김형태,하진영 강원대학교 정보통신연구소 2006 정보통신논문지 Vol.10 No.-

Out-of-vocabulary word or sentence rejection can be used to improve the confidence of speech recognition systems. In this paper, we introduce a standard phoneme based sentence rejection system using filler models and word missing ratio. We found that better performance, in terms of minimizing the average of FAR and FRR, was achieved using filler model method than word missing ratio method form the experimental results. But the latter method used less computation resources than the former method, because it requires no extra models and recognition network is simpler Furthermore there is no extra efforts to build filler models in word missing ratio method.

分離地域에 따른 綠膿菌 藥劑耐性의 比較

金炯泰 全南大醫大 1979 전남의대학술지 Vol.16 No.1

DP-Matching과 HMM을 이용한 온라인 필기 한자 인식

윤병훈,김형태,하진영 강원대학교 정보통신연구소 2007 정보통신논문지 Vol.11 No.-

On-line handwritten Chinese character recognition systems should be fast and yield accurate results in order to be practical systems. DP-matching method for variable length input requires building several allographs (multiple models) per class, which inevitably reduces the recognition speed. HMM is a good alternative, since one HMM per class is usually sufficient to model large variation of data. We proposed a recognition system in which HMM and DP-matching method are combined to improve the recognition speed and recognition accuracy at the same time. The HMM method using structure code features produces recognition candidates, then DP-matching method using chain code features reorders the candidates to get more accurate results. In doing so, we accomplished fast and accurate recognition results for 2,362 Chinese character classes. We also adopted result verification method to discriminate confusable character classes. We got 97.44% recognition accuracy using the proposed method in 113 ms per character in average.

문자 단위 매칭과 유닛 단위 매칭을 이용한 온라인 필기 한자 인식

윤병훈,김형태,박미나,하진영 강원대학교 정보통신연구소 2006 정보통신논문지 Vol.10 No.-

This paper presents on-line Chinese character recognition methods based on DP-Matching. The recognition system consists of character matching and unit matching methods. The character matching method adopts DP-matching between the input data and models as a major matching method, while unit matching method uses relation matching between consecutive units (part of character) in addition to DP-matching between units. A couple of methods combining character matching and unit matching methods were examined also. We found that we got best recognition result applying unit matching methods to the recognition candidates from character matching method.