• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.295-300
• /
• 2023

대화 데이터와 같은 데이터는 사람이 수작업으로 작성해야 하기 때문에 데이터셋 구축에 시간과 비용이 크게 발생한다. 현재 대두되고 있는 Large Language Model은 이러한 대화 생성에서 보다 자연스러운 대화 생성이 가능하다는 이점이 존재한다. 이번 연구에서는 LLM을 통해 사람이 만든 적은 양의 데이터셋을 Fine-tuning 하여 위키백과 문서로부터 데이터셋을 만들어내고, 이를 통해 문서 검색 모델의 성능을 향상시켰다. 그 결과 학습 데이터와 같은 문서집합에서 MRR 3.7%p, 위키백과 전체에서 MRR 4.5%p의 성능 향상을 확인했다.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.1
• /
• pp.70-75
• /
• 2022

The critical bandwidth represents response interactions with respect to a signal tone and their neighboring bands. This study was to analyze the critical bandwidths of a clinically important 500 Hz tone in Korean young male and female subjects (male = 10, female = 10) at a conversational level (60 dB HL). Data were measured with notched band noise and two alternative forced choice methods. Results showed that the critical bandwidth was slightly greater (95 Hz) than the previous Western measures. There were no statistically significant differences in gender, nor were there any significant differences in lateralization of the ear (p > 0.05). These results may have implications in optimizing effective tinnitus masking or the related clinical applications.

• Phonetics and Speech Sciences
• /
• v.15 no.4
• /
• pp.27-34
• /
• 2023

This study aimed to provide knowledge helpful for understanding voice problems related to occupations in the clinical field through an investigation and comparison of subjective vocal symptoms of 12 professional actors and 12 speech-language pathologists Among the 11 symptoms, "Difficulty with high pitch when singing," "Hypertension in the neck when speaking," and "Feel voice fatigue" were the most frequent symptoms in both groups. Additionally, the professional voice users reported a higher frequency of "Difficulty with high pitch when singing" (p=.049), "Hoarse voice" (p=.021), "Difficulty (requiring effort) when speaking" (p=.032), "Pain in the neck when speaking" (p=.009), and "Feel vocal fatigue" (p=.018) than the elite vocal performer group. This may be due to the different voice-related environments and differences in voice demands during occupational activities between the two groups.

• Korean Architects
• /
• no.10 s.462
• /
• pp.70-75
• /
• 2007

• Journal of the Korean Society of Laryngology, Phoniatrics and Logopedics
• /
• v.13 no.2
• /
• pp.124-131
• /
• 2002

Speech characteristics of patients with cleft palates are resonance disorders, articulatory disorders and voice disorders. The purpose of this study is to find the acoustic, physiological and articulatory characteristics of cleft palate speakers. Thirteen control groups and 3 cleft palate patients participated in this experiment. Test words were composed of simple vowels and consonants imbedded in low vowel /a/, /p 'ap'i/ and /sasi/ according to the evaluation experiments. CSL, Video fluoroscopy, Fiberscope and Nasometer were used to analyze VOT, vowel formants, profiles of articulator, VP port images and nasalance. The results are as follows : (1) The nasalance of cleft palate patients in the high vowel /i/, stop sounds and fricative sounds were 60%, 34.8% and 44.1%, respectively. These values were higher than those of the control group. (2) Posterior articulatory movements /k'a/ in patients with cleft palates showed backward movement in comparison with the control group on Video Fluoroscopic images and palatograms. These results suggested that patients with cleft palate have the compensatory oral sounds to close the VP port. (3) The VOT in patients with cleft palates was longer than that of the control group.

• Journal of the Korean Society of Laryngology, Phoniatrics and Logopedics
• /
• v.8 no.1
• /
• pp.44-48
• /
• 1997

Anatomic and physiological changes of the larynx with advancing age result in morphologic changes of the vocal fold and reduced control of the phonatory mechanism in elderly individuals and are reflected in increased unstability of fundamental frequency (Fo). The purpose of this study is to increase current understanding of acoustic and stroboscopic characteristics of normal elderly persons voices. First, phonated /a/ vowel productions by 40 normal adults (20 to 40 years, 20 men and 20 women) and 40 normal elderly persons (60 to 80 years,20 men and 20 women) were analyzed, using CSL (model 4300B) acoustic analysis software, to obtain acoustic measures related to fundamental frequency stability nd vocal resonance characteristics. Second, stroboscopic images of the vocal fold behavior in all subjects were analyzed by experienced specialists. In the men, fundamental frequency variation (vFe) (p＜0.01), jitter. (p＜0.05), and shimmer (p＜0.05) for the older group were significantly higher than the value for the adult group. In the stroboscopic findings, edema of vocal fold had a significant finding in aged men (15%). In the women, vFo (p＜0.05), jitter (p＜0.05), and noise to harmonic ratio (NHR) (p＜0.05) for the older group were significantly higher than the value for e adult group and first formant frequency (F1) (p＜0.01) and second formant frequency (F2) (p＜0.01) for. the older group were significantly lower than the value for the adult group. In the stroboscopic findings, vocal fold atrophy had a significant finding in aged women (25%). Frequency stability, as reflected by vFo, jitter, shimmer, and NHR, decreases with advancing age in men and women and spectral analysis of phonated /a/ vowel productions reveals the lowering of the frequency of F1 and second F2 with advancing age, especially in aged women. Change in the mass of vocal folds, due to atrophy or edema, is considered to be the greatest factor in these acoustic changes.

• Proceedings of the Korean Society for Cognitive Science Conference
• /
• 2002.05a
• /
• pp.211-215
• /
• 2002

본 실험은 종합적 인지과정을 추론할 수 있는 결합 문제(binding problem)를 언어적인지 과정을 통해 알아 본 실험으로, 총 10 명(남:61여:4, 평균나이:24.40 $\pm$ 1.35)의 정상군을 대상으로, 4개의 음소로 이루어진 단음절 명사를 목표 자극(target stimulus)으로 하고, 4개 음소의 임의적인 조합으로서 글자를 이루지 못하는 비목표 자극(non-target stimulus)을, 각각 200 회와 800 회씩 시각적으로 0.5초씩 무작위로 제시하여 128 채널 고밀도 사건관련전위(ERP)를 측정하였다. 이번 실험 결과의 주요 특징은 글자가 아닌 비목표 자극보다 글자인 목표 자극에서 두드러지게 나타난 두정엽 부근의 P500 과 N900 이라고 할 수 있다. 자극 제시 비율의 차이에서 오는 oddball 효과로 인한 기존 P300 의 인지적 의미를 이번 결과의 P500 이 함축한다고 볼 수 있으며, 단음절 단어를 인지할 때, 글자임을 인식하는 순간은 의미적인지 과정이 진행되었다기보다 그 글자의 형태만으로 낯익은 글자인지를 분간하는 것으로 보인다 따라서, 이 경우 기존 언어 실험에 자주 등장하던 의미론적 peak 인 N400 은 보이지 않고, 곧바로 형태적이고, 통사적(syntactic)인 인지 처리 과정인 P500이 나타났다고 해석할 수 있다. 하지만, 이번 실험에서는 N400 대신에 N900 이 나타났다. 이 결과는 이번 ERP 실험과 병행된 프로토콜 분석을 통해, 피험자가 자극 제시 후, 약 900ms 정도에, 이미 제시되고 사라진 글자 자극을 다시 한번 떠올리는 인지 과정이 일어난다는 점과 관련 지어 해석하면, 기존에 의미적(semantic) 인지 과정으로만 해석했던 negative-peak 를 생각(thinking)과 같은 내재적인지 과정(internal cognitive process)으로 확장하여 일반화하는 추론도 생각해 볼 수 있다. 요컨대, 언어인지를 통한 이번 실험을 통해, 뇌파에서 검출되는 negative-peak 은 internal cognitive process로 추측되고, positive-peak 는 external cognitive process 라고 생각된다. 덧붙여, 유의해서 볼 점은 각 peak-topology 에서 Cz 의 진폭이 Fz 보다 크게 나온 점과, 일반적으로 언어 기능을 담당한다는 좌측 측두엽(T7)이 우측(T8)보다 통계적으로 더 유의미한 차이를 보였다는 점등이다.

• Korean Journal of Cognitive Science
• /
• v.27 no.3
• /
• pp.469-499
• /
• 2016

The present study investigated the cognitive mechanism underlying online processing of Korean syntactic (for example, A bed/a clock belongs to/doesn't belong to the furniture "침대는/시계는 가구에 속한다/속하지 않는다") and lexical negation (for example, A tiger/a butterfly has/doesn't have a tail "호랑이는/나비는 꼬리가 있다/없다") using an ERP(Event-related potentials) technique and a truth-value verification task. 23 Korean native speakers were employed for the whole experiment and 15's brain responses (out of 23) were recorded for the ERP analysis. The behavioral results (i.e. verification task scores) show that there is universal pattern of the accuracy and response time for verification process: True-Affirmative (high accuracy and short latency) > False-Affirmative > False-Negated > True-Negated. However, the components (early N400 & P600) reflecting the immediate processing of a negation operator were observed only in lexical negation. Moreover, the ERP patterns reflecting an effect of truth value were not identical: N400 effect was observed in the true condition compared to the false condition in the lexically negated sentences, whereas Positivity effect (like early P600) was observed in the false condition compared to the true condition in the syntactically negated sentences. In conclusion, the form and location of negation operator varied by languages and negation types influences the strategy and pattern of online negation processing, however, the final representation resulting from different computational processing of negation appears to be language universal and is not directly affected by negation types.

• Journal of Families and Better Life
• /
• v.23 no.4 s.76
• /
• pp.17-26
• /
• 2005

The purpose of the this study was to identify the relations among intelligence, parenting environment and self-regulation of preschool children. The subjects were 81 children between the ages 3 and 6 and their mothers from one day-care center in Seoul. T-test, correlation analysis, multiple regression analysis were used for data analysis. First, the result of this study showed that children's self-regulation ability by sex difference was significant to only girls and children's self-regulation ability by no was not significant. Parents who had higher educational background were positively related to children's intelligence and children's self-regulation ability was significant to mothers' educational background. Second, the preschool children who got higher scores in intelligence scale were higher self-regulation ability. Third, the young children's self-regulation ability showed significance only in motor intelligence which was one of the subscales of intelligence scale. Fourth, the effect of children's intelligence and parenting environment on preschoolers' self-regulation were relatively presented as intelligence($\beta$=.34, p<.01), trauma state($\beta$=0.31, p<.05) and development state($\beta$ =.23, p<0.5) which were subscales of parenting environment.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.95-100
• /
• 2018

채팅 시스템을 잘 만들기 위해서는 양질, 대량의 채팅 말뭉치가 굉장히 중요하지만 구축 시 많은 비용이 발생한다는 어려움이 있었다. 따라서 본 논문에서는 영화 자막, 극대본과 같이 대량의 발화 데이터를 이용하여 채팅 말뭉치를 반자동으로 확장하는 방법을 제안한다. 채팅 말뭉치 확장을 위해 미리 구축된 채팅 말뭉치와 유사도 기법을 이용하여 채팅 유사도를 구하고, 채팅 유사도가 실험을 통해 얻은 임계값보다 크다면 올바른 채팅쌍이라고 판단하였다. 그리고 길이가 매우 짧은 채팅성 발화의 채팅 유사도를 효과적으로 계산하기 위해 본 논문에서 제안하는 것은 형태소 단위 임베딩 벡터와 합성곱 신경망 모델을 이용하여 발화 단위 표상을 생성하는 것이다. 실험 결과 기본 발화 단위 표상 생성 방법인 TF를 이용하는 것보다 정확률, 재현율, F1에서 각각 5.16%p, 6.09%p, 5.73%p 상승하여 61.28%, 53.19%, 56.94%의 성능을 가지는 채팅 말뭉치 반자동 구축 모델을 생성할 수 있었다.