• 제목/요약/키워드: Korean corpus

검색결과 1,197건 처리시간 0.034초

대형 사전훈련 모델의 파인튜닝을 통한 강건한 한국어 음성인식 모델 구축 (Building robust Korean speech recognition model by fine-tuning large pretrained model)

  • 오창한;김청빈;박기영
    • 말소리와 음성과학
    • /
    • 제15권3호
    • /
    • pp.75-82
    • /
    • 2023
  • 자동 음성 인식(automatic speech recognition, ASR)은 딥러닝 기반 접근 방식으로 혁신되었으며, 그중에서도 자기 지도 학습 방법이 특히 효과적일 수 있음이 입증되고 있다. 본 연구에서는 다국어 ASR 시스템인 OpenAI의 Whisper 모델의 한국어 성능을 향상시키는 것을 목표하여 다국어 음성인식 시스템에서의 비주류 언어의 성능 문제를 개선하고자 한다. Whisper는 대용량 웹 음성 데이터 코퍼스(약 68만 시간)에서 사전 학습되었으며 주요 언어에 대한 강력한 인식 성능을 입증했다. 그러나 훈련 중 주요 언어가 아닌 한국어와 같은 언어를 인식하는 데 어려움을 겪을 수 있다. 우리는 약 1,000시간의 한국어 음성으로 구성된 추가 데이터 세트로 Whisper 모델을 파인튜닝하여 이 문제를 해결한다. 또한 동일한 데이터 세트를 사용하여 전체 훈련된 Transformer 모델을 베이스 라인으로 선정하여 성능을 비교한다. 실험 결과를 통해 Whisper 모델을 파인튜닝하면 문자 오류율(character error rate, CER) 측면에서 한국어 음성 인식 기능이 크게 향상되었음을 확인할 수 있다. 특히 모델 크기가 증가함에 따라 성능이 향상되는 경향을 포착하였다. 그러나 Whisper 모델의 영어 성능은 파인튜닝 후 성능이 저하됨을 확인하여 강력한 다국어 모델을 개발하기 위한 추가 연구의 필요성을 확인할 수 있었다. 추가적으로 우리의 연구는 한국어 음성인식 애플리케이션에 파인튜닝된 Whisper 모델을 활용할 수 있는 가능성을 확인할 수 있다. 향후 연구는 실시간 추론을 위한 다국어 인식과 최적화에 초점을 맞춰 실용적 연구를 이어갈 수 있겠다.

임신우에서 발생된 난포의 기능에 대한 면역조직화학적 관찰 (Immunohistochemical observation on the functions of follicles developed in ovaries of pregnant cows)

  • 곽수동;고필옥;양제훈;원청길;강정부
    • 대한수의학회지
    • /
    • 제43권4호
    • /
    • pp.555-561
    • /
    • 2003
  • Incidence of estrum or abortions in pregnant cows may be affected by large follicles developed together with corpus luteum in pair ovaries of pregnant cows. But the follicles of pregnant phase were not assessed about histological findings. Determination of the healthy and atretic follicles by presence of proliferative cells or apoptotic cells and histological compositions of follicles would be used as important data on measurements of ovarian functions. This study was focussed mainly to investigate macroscopical, histological and immunohistochemical findings of ovarian follicles of pregnant Korean native cows and dairy cows (Holstein). In immunohistochemical methods, assessments of proliferative cells using PCNA antibody and apoptotic cells using TUNEL methods were performed. The follicles were observed on all 24 pregnant cows (17 Korean native cows and 7 Holstein cows). Follicles of greater than 10 mm in daimeter were developed in 37.5% (9/24 heads) of these pregnant cows. largest follicles from in these cows were $16.0{\times}15.0mm$ in diameter in a Korean native cow(l20 days of gestation), $13.4{\times}10.1mm$ in a Korean native cow(50 days of gestation), $12.9{\times}11.5mm$ in a Holstein cow (120 days of gestation). 40.5% among all follicles having diameter of greater than 1.0 mm in pregnant cows were assessed as atretic follicles and in addition, healthy follicles also showed less in number and smaller in size and thinner in wall layer compared with those of cyclic phase ovaries. In immunohistochemical findings, also proliferative positive cells and apoptotic positive cells on the granulosa cell layers in the healthy follicles of pregnant cows appeared less than on those of cyclic follicles. So these follicles were assessed as weakly active follicles. In large follicles, above positive cells were not nearly appeared but granulosa cell debris were more appeared among the granulosa cells. So these large follicles were assessed as inactvie or atretic follicles. The above findings suggest that small follicles of pregnant phase were weakly active or atretic and large follicles were inactive or atretic.

홍삼, 천마, 적하수오 병용투여에 의한 고지혈증 랫드에서의 콜레스테롤 및 발기부전 개선효과 (Beneficial effect of Combination with Korean Red Ginseng, Gastrodia Rhizoma and Polygoni Multiflori on Cholesterol and Erectile Dysfunction in Hyperlipidemia rats)

  • 이윤정;고민철;담서;이재윤;황진석;차정단;최경민;강대길
    • 대한본초학회지
    • /
    • 제30권6호
    • /
    • pp.69-75
    • /
    • 2015
  • Objectives : This study was designed to investigate effects of the combination with Korean Red Ginseng (Panax ginseng C.A. Meyer), Gastrodia Rhizoma (Gastrodia elata Blume) and Polygoni Multiflori Radix (Polygonum multiflorum Thunberg) on metabolic disorders including cholesterol and erectile dysfunction in hyperlipidemia rats.Methods : Animals were divided into six groups; Control with normal diet, high fat/cholesterol-diet (HFCD), fluvastatin, Korean Red Ginseng treated (KRG), and the combination treated (Korean Red Ginseng, Gastrodia Rhizoma and Polygoni Multiflori Radix; 1:1:1 for KGP1 and 2:1:1 for KGP2). The experimental groups initially received HFCD for 10 weeks and then treated orally with fluvastatin, KRG, KGP1 and KGP2 during the final 6 weeks. Erectile function was determined by the measurements of intracavernosal pressure (ICP) and maximal arterial pressure (MAP) after electrical stimulation of the cavernosal nerve.Results : KGP2 decreased the level of total cholesterol and LDL cholesterol in the sera of HFCD rats without no changes of body weights. KRG, KGP1 and KGP2 decreased the level of C-reactive protein (CRP) levels except of fluvastatin, synthetic HMG-CoA reductase inhibitor. KRG, KGP1 and KGP2 significantly increased the ICP, ICP/MAP ratio, area under the curve (AUC) compared with those of normal rat. Morphometric analyses showed that KRG, KGP1 and KGP2 increased the volume of smooth muscle and the regular arrangement of collagen fibers in corpus cavernosum of HFCD rats. The penile expression of eNOS was increased by KRG, KGP1 and KGP2.Conclusions : Based on these results, we suggest that the combination with Korean Red Ginseng, Gastrodia Rhizoma and Polygoni Multiflori may improve hyperlipidemia through regulating the lipid profiles and erectile dysfunction in rats.

CTC를 적용한 CRNN 기반 한국어 음소인식 모델 연구 (CRNN-Based Korean Phoneme Recognition Model with CTC Algorithm)

  • 홍윤석;기경서;권가진
    • 정보처리학회논문지:소프트웨어 및 데이터공학
    • /
    • 제8권3호
    • /
    • pp.115-122
    • /
    • 2019
  • 지금까지의 한국어 음소 인식에는 은닉 마르코프-가우시안 믹스쳐 모델(HMM-GMM)이나 인공신경망-HMM을 결합한 하이브리드 시스템이 주로 사용되어 왔다. 하지만 이 방법은 성능 개선 여지가 적으며, 전문가에 의해 제작된 강제정렬(force-alignment) 코퍼스 없이는 학습이 불가능하다는 단점이 있다. 이 모델의 문제로 인해 타 언어를 대상으로 한 음소 인식 연구에서는 이 단점을 보완하기 위해 순환 신경망(RNN) 계열 구조와 Connectionist Temporal Classification(CTC) 알고리즘을 결합한 신경망 기반 음소 인식 모델이 연구된 바 있다. 그러나 RNN 계열 모델을 학습시키기 위해 많은 음성 말뭉치가 필요하고 구조가 복잡해질 경우 학습이 까다로워, 정제된 말뭉치가 부족하고 기반 연구가 비교적 부족한 한국어의 경우 사용에 제약이 있었다. 이에 본 연구는 강제정렬이 불필요한 CTC 알고리즘을 도입하되, RNN에 비해 더 학습 속도가 빠르고 더 적은 말뭉치로도 학습이 가능한 합성곱 신경망(CNN)을 기반으로 한국어 음소 인식 모델을 구축하여 보고자 시도하였다. 총 2가지의 비교 실험을 통해 본 연구에서는 한국어에 존재하는 49가지의 음소를 판별하는 음소 인식기 모델을 제작하였으며, 실험 결과 최종적으로 선정된 음소 인식 모델은 CNN과 3층의 Bidirectional LSTM을 결합한 구조로, 이 모델의 최종 PER(Phoneme Error Rate)은 3.26으로 나타났다. 이는 한국어 음소 인식 분야에서 보고된 기존 선행 연구들의 PER인 10~12와 비교하면 상당한 성능 향상이라고 할 수 있다.

익지인(益智仁), 두충(杜沖), 백강잠(白殭蠶) 혼합추출물이 남성갱년기 증상 개선에 미치는 영향 (Effects of Fructus Amomi Amari, Eucommiae Cortex, Bombyx Batryticatus Extract on Improving Symptoms of Late-onset Hypogonadism)

  • 박선영;안상현;김호현
    • 동의생리병리학회지
    • /
    • 제33권2호
    • /
    • pp.89-101
    • /
    • 2019
  • In recent times, the number of men with late-onset hypogonadism has increased, and interest on this topic has also increased. This study was conducted to investigate effects of the mixture extract of Fructus amomi Amari, Eucommiae cortex, Bombyx batryticatus on improve late-onset hypogonadism. The experimental subjects consisted of three groups: a control group consisting of 8-week-old male ICR mice that had undergone no treatment, an aging-elicited group (AE group) consisting of 50-week-old ICR male mice that had undergone no treatment, and a Mixed herbal extract treatment group (MT group) consisting of 50-week-old ICR male mice that had undergone the mixture extract of Fructus amomi Amari, Eucommiae cortex, Bombyx batryticatus treatment (0.1 g/kg/day) for 6 months. After the experiment, the mice from all the experimental groups were dissected, and they were analyzed through histochemical and immunohistochemical methods. The mixture extract of Fructus amomi Amari, Eucommiae cortex, Bombyx batryticatus reduces aging-induced cell damage and oxidative stress and increases the secretion of serotonin and B-endorphin in aged mice, and promotes spermatogenesis in seminiferous tubules and reduces apoptosis and oxidative stress, and increases androgen receptor, $17{\beta}-HSD$ and GnRH, increases the ratio of smooth muscle to collagen fibers in the corpus cavernosum, increases eNOS, decreases PDE-5 and oxidative stress in aged mice, so it improves depression, reproductive, sexual problems caused by Late-onset hypogonadism. the mixture extract of Fructus amomi Amari, Eucommiae cortex, Bombyx batryticatus inhibits the induction of osteoporosis by increasing decreased bone matrix distribution due to aging, increasing the activities of OPC and OPN, which are produced in osteoblasts, and decreasing RANKL, MMP-3 activity, increasing OPG activity. It also reduces muscle damage, oxidative stress, inflammation and apoptosis of muscle tissue, and increases Myo-D in the sartorius muscle of aged mice for improving muscle atrophy caused by by Late-onset hypogonadism.

콘포머 기반 FastSpeech2를 이용한 한국어 음식 주문 문장 음성합성기 (A Korean menu-ordering sentence text-to-speech system using conformer-based FastSpeech2)

  • 최예린;장재후;구명완
    • 한국음향학회지
    • /
    • 제41권3호
    • /
    • pp.359-366
    • /
    • 2022
  • 본 논문에서는 콘포머 기반 FastSpeech2를 이용한 한국어 메뉴 음성합성기를 제안한다. 콘포머는 본래 음성 인식 분야에서 제안된 것으로, 합성곱 신경망과 트랜스포머를 결합하여 광역과 지역 정보를 모두 잘 추출할 수 있도록 한 구조다. 이를 위해 순방향 신경망을 반으로 나누어 제일 처음과 마지막에 위치시켜 멀티 헤드 셀프 어텐션 모듈과 합성곱 신경망을 감싸는 마카론 구조를 구성했다. 본 연구에서는 한국어 음성인식에서 좋은 성능이 확인된 콘포머 구조를 한국어 음성합성에 도입하였다. 기존 음성합성 모델과의 비교를 위하여 트랜스포머 기반의 FastSpeech2와 콘포머 기반의 FastSpeech2를 학습하였다. 이때 데이터셋은 음소 분포를 고려한 자체 제작 데이터셋을 이용하였다. 특히 일반대화 뿐만 아니라, 음식 주문 문장 특화 코퍼스를 제작하고 이를 음성합성 훈련에 사용하였다. 이를 통해 외래어 발음에 대한 기존 음성합성 시스템의 문제점을 보완하였다. ParallelWave GAN을 이용하여 합성음을 생성하고 평가한 결과, 콘포머 기반의 FastSpeech2가 월등한 성능인 MOS 4.04을 달성했다. 본 연구를 통해 한국어 음성합성 모델에서, 동일한 구조를 트랜스포머에서 콘포머로 변경하였을 때 성능이 개선됨을 확인하였다.

안태음이 임신랫드와 태자에 미치는 영향 (The Effects of the Administration on Oriental Medicine, Antaeeum, in the Pregnant Rat and Their Fetuses)

  • 김창석;박해모;이선동;이장우;김판기;신헌태
    • 한국환경보건학회지
    • /
    • 제33권4호
    • /
    • pp.306-316
    • /
    • 2007
  • This study have a object to found out the effects of oriental herb medicine, Antaeeum, to dams of rats and their offsprings. The Antaeeum was savaged to female Sprague-Dawley rats at a dose of 5 mg/kg/day for 3 weeks during gestation periods. Dams of rat were sacrificed at 20th day of gestation, and were observed major internal and reproductive organs. Approximately live fetuses in the 20th days of gestation were selected randomly and examined with stereo microscopes. Others offsprings were fixed with 95% ethanol for skeletal examinations. The fixed fetuses were stained with alcian blue and alizarin red S to observe skeletal variations or malformations. Maternal body weight of Antaeeum treated dams have a tendency of increasing compared with control dams. There were no significant difference in internal and reproductive organs of weight or findings. The spleenic organ relative weight of treated dams were decreased compared with the control significaltly (p<0.05). There were no significant changes between two groups in blood chemistry and hematological values. There were no significant changes in number of corpus luteum, implantation, live fetuses and implantation rate, delivery rate, late resorption rate and sex ratio. But in the Antaeeum treated group showed lower early resorption rate than that of the control dams. Fetal body weight and number of fetus a dam at Antaeeum treated group were higher than that of control group. The fetuses of dams treated with Antaeeum didn't induced external malformations. Vertebral and sternal variations were observed in Antaeeum group, but compared with the control, those variations were not significant. The ossification numbers of rib, cervical, thoracic, and lumber were normal. Fetuses treated with Antaeeum to the dams showed no significant difference in the number of caudal vertebra (P>0.01). From these results, it can be concluded that Antaeeum showed no toxicity effects on maternal side especially on body weight, early resorption rate, and number of live fetuses. Also there were no significant changes on maternal organ weights except spleen, hematological data, reproductive organs. Although skeletal variations were examined at vertebra and sternum, this Antaeeum could not induced significant choses in bone malformation.

달생탕이 랫드의 모체와 태자에 미치는 영향에 대한 연구 (THe Effects of the Administration on Oriental Medicine, Dalsaengtang, in the Pregnant Rat and Their Fetuses)

  • 박해모;김창석;이선동;이장우;유재홍;김판기
    • 한국환경보건학회지
    • /
    • 제32권4호
    • /
    • pp.342-352
    • /
    • 2006
  • The experiments was undertaken to evaluate the effects of herbal medicine, Dalsaengtang, in pregnant rats and fetuses. Female Sprague-Dawley rats were orally administered with the Dalsaengtang at dose of 5 mg/kg/day for 20 days. Pregnant rats were sacrificed at 20th day of gestation, and observed internal and reproductive organs. Approximately live fetuses in the 20th day of gestation were randomly selected and fixed in 95% ethanol. To observe skeletal malformations, fetuses were stained with alcian blue and alizarin red S. Maternal body weight of dalsaengtang treated group has a tendency to increase compared to that of control group. The relative liver and kidney weights of dalsaengtang treated group were also increased to that of control group. There were no significant changes between two groups in blood chemistry and hematological values. There were no significant changes in number of corpus luteum, implantation, live fetuses and implantation rate, delivery rate, late resorption rate and sex ratio. But Dalsaengtang administered group showed lower early resorption rate than the control group. From the sex ratio, number of females, bigger than number of males in the control group, and more males than females in Dalsaengtang administered group. Neonatal body weight and number of fetus of Dalsaengtang group were increased to that of control group. The fetuses of dams treated with Dalsaengtang didn't showed external malformation. Vertebral and sternal variations were observed in Dalsaengtang group but, compared to the control, those variations were insignificant. The number of ribs, cervical, thoracic, and lumber were normal. The number of sacral and caudal vertebrae were increased. Fetuses showed significant difference in the number of caudal vertebra (P<0.01). From these results, it can be concluded that Dalsaengtang showed no toxicity effects on maternal body weight, early resorption rate, and number of live fetuses. There were no significant changes in organ weight, hematological data, reproductive organs. Although skeletal variations were showed in vertebrate and sternum, Dalsaengtang did not shown significant changes in bone malformation.

재래돼지에서 수정란의 회수 및 동결보존에 관한 연구 (Studies on Recovery and Cryopreservation of Embryos in Korean Native Swine)

  • 손동수;연성흠;허태영;강석진;서국현;최선호;류일선;이규승;박창식
    • 한국수정란이식학회지
    • /
    • 제19권3호
    • /
    • pp.257-264
    • /
    • 2004
  • 멸종위험이 큰 우리나라 재래돼지를 유전자원으로서 안전하게 보존하고 유전적 다양성을 유지하기 위한 수단으로 수정란을 채취하여 동결보존하기 위해서 미경산 재래돼지에서 과배란 유기를 위한 적정 호르몬의 수준과 수정란의 회수 및 동결보존 방법을 확립하고자 수행한 결과는 다음과 같다. 1. hCG 500IU와 PMSG를 500, 750, 1,000IU 및 hCG 750IU와 PMSG 1,000IU를 각각 투여한 재래돼지의 배란황체와 미배란난포의 수는 각각 12.4, 13.6, 30.0 및 23.3개로 PMSG 1,000IU와 hCG 500IU를 투여한 재래돼지가 다른 용량의 처리돼지보다 난소반응이 양호하였으나 유의적인 차이는 없었으며, 배란황체수에 대한 수정란 회수율은 59.4-79.2% 수준이었다. 2. 과배란처리된 공란돈에서 수정후 4일에 회수된 수정란의 발육단계는 상실기의 수정란이 수정후 5일보다 유의적으로 많이 회수되었으며(P<0.01), 수정후 5일에 회수된 배 반포기의 수정란을 수정후 4일보다 유의적으로 많이 회수되었다(P<0.05). 3. 확장배반포기 수정란을 1.4M glycerol의 항동 해제를 이용하여 관행의 완만동결법으로 동결한 처리에서 생존율은 25.3%였다.

Pregnancy Rate of In Vitro Produced Korean Cattle Embryos according to Transport Time Course

  • Park, Hyo-Young;Kim, Eun-Young;Kim, Young-Hun;Mun, Seong-Ho;Oh, Chang-Eon;Han, Young-Joon;Kim, Nam-Hyung;Lee, Sung-Soo;Ko, Moon-Suck;Riu, Key Zung;Park, Se-Pill
    • Reproductive and Developmental Biology
    • /
    • 제33권4호
    • /
    • pp.257-262
    • /
    • 2009
  • This study was to investigate pregnancy rate of IVM/IVF/IVC Korean cattle (registered in government) embryos according to transport time course. For the production of embryos, oocytes recovered from slaughtered excellent grade cow and highly motile frozen-thawed bull semen (purchased from LIMC, KPN#497) was used. In vitro produced embryos were cultured in CR1aa medium for 8 days and some of them were frozen. The rate of average cleavage (>2-cell) was 83.0% (308/371) and blastocyst rate at day 8 was 34.7% (107/308). Among in vitro produced blastocyst embryos at day 8, most healthy embryos were freshly transferred on production day and some frozen embryos were direct transferred on appropriate day. These embryos were produced in a laboratory, embryo transfer (ET) was planned in 10 areas of the remote island (Jeju) from the laboratory by airplane. Thus, we examined the pregnancy rate in recipient cow according to embryo of transport time course before ET. From embryo transferred 44 recipient cows, overall pregnancy was 40.9% (18/44), these 18 cows were all calved [single, 94% (17/18); twin, 6% (1/18)] and total embryo implantation rate was 26% (19/66). Comparing transport time in the base of 6 hr, pregnancy rate in ET group required less 4 hr (60%, 9/15) was significantly higher than that required more 6 hr (26.3%, 5/19). In direct ET of freezing embryos, the pregnancy rate was 40% (4/10). However, it was difficult to find the meaning of temperature, pH and corpus luteum quality of recipients on comparison of pregnancy rate. When the cell death level of embryos according to storage time in thermos (straw container) before ET was measured by TUNEL staining, apoptotic index was increased with storage time-dependent. These results demonstrated that long distance transfer of IVM/IVF/IVC embryos is possible and the time of embryo transport is very important for the pregnancy rate on field trial.