자동차 내장형 음성합성시스템을 위한 데이터베이스 구현

권오일(Ohil Kwon) 한국자동차공학회 2010 한국자동차공학회 부문종합 학술대회 Vol.2010 No.5

Speech Database is one of the most important part of Text-to-speech(TTS) system. Especially, the embedded TTS system needs smaller size of database than that of the sever TTS system. So, the compression and statistical reduction of database is a very important factor in the embedded TTS system. But this compression and statistical reduction of database always brings out a loss of quality of the synthesized speech. In this paper, we propose a method of constructing database for high quality embedded TTS system and verify the quality of synthesized speech with MOS(Mean Opinion Score) test.

주요 산림복원사업지 내 귀화식물의 특성과 공종 간 영향 관계

전용삼,박준형,권오일,이혜정,임채영,Jeon, Yongsam,Park, Joon Hyung,Kwon, Ohil,Lee, Hye Jeong,Lim, Chaeyoung 한국환경생태학회 2022 한국환경생태학회지 Vol.36 No.5

이 연구는 산림복원사업 이후 유입되는 귀화식물과 생태계교란 식물의 실태 및 사업의 특성에 따라 나타나는 양상을 파악하는데 목적을 두고 있다. 현장조사는 산림청에서 수행한 산림복원사업지 29개소를 대상으로 2020년과 2021년 봄철(5~6월), 여름철(8~9월)로 나누어 연 2회 실시하였고, 사업 범위 밖의 지역은 조사에서 제외하여 실제 복원사업 후 도입 및 유입되었다고 볼 수 있는 식물을 파악할 수 있도록 하였다. 또한 대상지 내 분포하는 귀화식물상과 복원사업 준공내역 확인을 통해 현장에 실제 적용된 공종과의 연관성을 분석하였다. 전체 대상지에서 출현한 귀화식물은 29과 80속 108종 1아종으로 109분류군이며, 생태계교란 식물은 총 3과 7속 8종이 확인되었다. 귀화식물의 분류군 수와 귀화율은 사업 후 소요시간이 증가함에 따라 점차 감소하는 추세를 보였고, 귀화식물 분류군 수와 귀화율은 권역 간 유의적인 차이가 없는 것으로 나타났으나(p>0.05), 훼손 유형별 귀화식물 분류군 수의 경우 백두대간 단절 구간과 채석지, 시설부지 등의 유형에서 출현한 귀화식물 수가 상대적으로 많았다(p<0.05). 복원지 내 나지 및 비탈면의 녹화를 위한 식생도입 방법인 종자 파종, 나무 식재, 초본 식재, 떼붙임 공종의 유무에 따른 귀화율 비교 분석 결과, 종자파종을 할 경우 귀화율은 평균 15.545%로 파종 공종이 없는 경우의 평균 9.167%에 비해 높은 것으로 나타났고(p<0.05), 그 외 다른 식생도입 방법은 공종 적용 유무에 따라 귀화율의 유의적인 차이가 나타나지 않았다(p>0.05). 이는 식물체의 형태를 확인하고 어느 정도 제어할 수 있는 나무 식재와 초본 식재, 떼붙임 공종과는 달리 종자 상태에서 귀화식물 여부의 파악이 까다로운 종자 파종 공종이 산림복원사업지 내 귀화식물 도입에 큰 영향을 미치는 것을 의미한다. 따라서 복원지 내 종자 파종 시 종자 전문가에 의한 검수 과정의 도입이 필요할 것으로 사료된다. 이 연구의 결과는 향후 산림복원사업 시 식생 도입 및 사후관리의 방향설정에 도움이 되는 자료로 활용될 수 있을 것으로 판단된다. This study was designed to identify the actual state of naturalized plants and invasive alien species that cause disturbances to the ecosystem, plants which are introduced after forest restoration, and explore the implications resulting from the project. Onsite examination included 29 sites which have been subjected to forest restoration by the Korea Forest Service. Once these were chosen, activity took place twice a year in the spring (May-June) and in the summer (August-September) in 2020 and 2021. Areas not relevant to the project sites were excluded from this activity so that we could identify the plants that could be understood to have been introduced or brought into the site after the actual forest restoration. And the correlation was analyzed, between the naturalized flora within the project sites and the working types applied to the site through confirmation of completion of the restoration project. The naturalized plants appearing on the entire site cover a total of 109 taxa, which includes 29 families, 80 genera, 108 species and 1 subspecies, while invasive plants included 3 families, 7 genera and 8 species. The number of classifications and the naturalization rate gradually decreased over time, after the project. While there was no significant difference between the number of classification groups and the naturalization rate for naturalized plants between project sites, given the number of taxa of naturalized plants, organized by type of damage, there were relatively more naturalized plants that appeared in the severed section of the Baekdudaegan Mountain Range, as well as at quarry and facility sites. Seeding apparently results in naturalization rates as high as 15.545%, on average, based on comparisons of naturalization rates by sowing, seeding, planting, herb planting, and sod pitching channels, all of these being methods of vegetation for planting/greening of bareland and slopes within the project areas. With no seeding, it was 9.167%, higher than the average. As for other vegetation, there was no significant difference depending on application of the working type. This means that unlike the plants subjected to planting, the working type of seed planting which makes it difficult to identify whether a certain plant is a naturalized plant greatly affects the introduction of naturalized plants to the restoration sites, even when using herb planting and sod pitching to control plants and results. Therefore the study suggests that there be inspection by experts of seeds when sowing within restoration sites. The results of this study suggest good practices that will help to direct effective vegetation restoration and follow-up management.

한국 음성 합성기에서의 운율 향상을 위한 AP, IP 예측

홍문기(Munki Hong),강선미(Sunmee Kang),신지영(Jiyoung Sin),권오일(Ohil Kwon),한장규(JangKyu Han) 한국통신학회 2002 한국통신학회 학술대회논문집 Vol.2002 No.7

실시간 차선 이탈 감지 알고리즘

權五逸,梁玄錫 弘益大學校 科學技術硏究所 2004 科學技術硏究論文集 Vol.15 No.-

As the number of vehicles increases, solving a traffic congestion problem and decreasing casualties by automobile accidents become the most important research areas for vehicular researchers. Traffic congestion problem is solved, theoretically, by constructing lots of roads, which is not practically possible due to the limited budget. Instead, researchers are trying to develop a smart traffic signal system that optimizes the traffic flow and to develop a smart highway system that increases the capacity of highways by allowing a shorter distance between adjacent automobiles. To decrease casualties, it is required to develop a collision detection/avoidance system that pre-detects possible collisions and warns the driver or automatically takes the optimal action to avoid the collision if it is too late to notice to the driver. In this paper, we focus on development a collision detection system, which detects abnormal lane-change such as zig-zag driving or driving while stepping on a lane. First of all, the left and right lanes are recognized from the picture image taken by a camera attached in front of a car. Then, the distance from the center of the car to the left lane is calculated. From this measure, we can detect whether the car is crossing a lane or not. The crossing pattern is analyzed to determine whether it is a normal lane-crossing or a zig-zag driving. Simulations are done for the cases when a car changes a lane normally, is in zig-zag driving, is driving close to the left lane, and is driving while stepping on a lane. we have used 100 pictures taken with a rate of 3 pictures per 1 second while the car is moving with 60km/hr.

차선검출 및 주행선 변경감지 알고리즘의 개발

權五逸,梁玄錫 弘益大學校 科學技術硏究所 2003 科學技術硏究論文集 Vol.14 No.-

As a number of vehicles increases, safety becomes the most important aspect in vehicular research. Many researchers try to find solutions that avoids collisions occurred by driving under influence or sleeping. A good portion of collision can be avoided if zig-zag driving or continuous lane changing is detected and warned to the driver. In this paper, we propose an algorithm that finds the lanes from the picture taken by a camera attached in front of a vehicle. Detecting lanes while moving is the first step in developing a system for lane-crossing detection. We modified currently avaliable algorithm and was able to detect dotted lanes, too. With this result, a method was proposed to detect lane-crossing. The method proposed uses a imaginable bar at the center of the picture. If a vehicle is driving along the center line of the lane, the bar does not touch the left and the right lanes. If the bar touches either lanes, it is possible to notify whether the vehicle is crossing the lanes. In this paper, introductory results are presented and further research is under way.