• 음성과학
• /
• 제15권2호
• /
• pp.101-110
• /
• 2008

The purpose of this study was to investigate the speech intelligibility of voice disordered patients when we consider the severity and utterance level as variables. Based on the severity level, 12 patients were divided into three groups, G1, G2, and G3 group, respectively. Words, phrases and sentences produced by the speakers were judged by four listeners with normal hearing, and we compared the intelligibility scores of the three groups. As a result, the speech intelligibility was decreased as the severity level was increased, and the difference was statistically significant. However, the mean difference among words, phrases and sentences was not significant, and the variation of intelligibility according to the utterance level was not under the regular rules.

• ETRI Journal
• /
• 제36권5호
• /
• pp.714-720
• /
• 2014

In this paper, we propose the use of data-driven probabilistic utterance-level decision logic to improve Weighted Finite State Transducer (WFST)-based endpoint detection. In general, endpoint detection is dealt with using two cascaded decision processes. The first process is frame-level speech/non-speech classification based on statistical hypothesis testing, and the second process is a heuristic-knowledge-based utterance-level speech boundary decision. To handle these two processes within a unified framework, we propose a WFST-based approach. However, a WFST-based approach has the same limitations as conventional approaches in that the utterance-level decision is based on heuristic knowledge and the decision parameters are tuned sequentially. Therefore, to obtain decision knowledge from a speech corpus and optimize the parameters at the same time, we propose the use of data-driven probabilistic utterance-level decision logic. The proposed method reduces the average detection failure rate by about 14% for various noisy-speech corpora collected for an endpoint detection evaluation.

• 음성과학
• /
• 제8권1호
• /
• pp.135-142
• /
• 2001

The present study was performed to investigate adequate levels of utterances which can give essential as well as useful information about the patients' voice, by examining the degrees of correlation between the levels of utterances (vowels, words, and phrase paragraph reading) and the entire utterance including all of the levels. For this purpose, a total of 10 individual utterance samples (5 vowels, 3 words, 1 phrase, 1 paragraph reading) were collected from each of the 30 subjects with voice disorder patients, and four experienced voice therapists evaluated them using GRBAS. The results showed that four therapists highly agreed upon on 'G' parameter. The coefficient of the correlation between each level of utterance and entire utterance tended to be above 0.70. Judgements of the vowel /$\varepsilon$/ as well as /o/ highly correlated with the judgement of the entire utterance. Regardless of severity, the judgement of the entire utterance highly correlated with the judgements of the vowel /u/ and the paragraph reading. These results suggest that experienced voice therapists can precisely evaluate patients' voice quality with only one sustained vowel in the clinic field, as is done with the entire utterance evaluation.

• 말소리와 음성과학
• /
• 제1권1호
• /
• pp.55-62
• /
• 2009

This paper proposes an improved method to verify a keyword segment that results from a word spotting system. First a baseline word spotting system is implemented. In order to improve performance of the word spotting systems, we use a two-pass structure which consists of a word spotting system and an utterance verification system. Using the basic likelihood ratio test (LRT) based utterance verification system to verify the keywords, there have been certain problems which lead to performance degradation. So, we propose a method which uses phone-level log-likelihood ratios (PLLR) patterns in computing confidence measures for each keyword. The proposed method generates weights according to the PLLR patterns and assigns different weights to each phone in the process of generating confidence measures for the keywords. This proposed method has shown to be more appropriate to word spotting systems and we can achieve improvement in final word spotting accuracy.

• 말소리와 음성과학
• /
• 제4권1호
• /
• pp.111-118
• /
• 2012

The purpose of this study was to characterize and determine variances of speaking fundamental frequency and vocal intensity depending on gender and three utterance conditions (spontaneous speech, reading, and counting). A total of 65 undergraduate students (32 male students, 33 female students) attending universities in Daegu, South Korea participated in this study. The subjects were all in their 20s. This study used KayPENTAX's Visi-Pitch IV (Model 3950) to measure the variances of speaking fundamental frequency (SFF0) and vocal intensity (VI). As a result, this study came to the following conclusions. First, it was found that both males and females showed no significant difference in SFF0 and vocal intensity among three utterance conditions. Second, this study sought to analyze differences in the variances of SFF0 between males and females. As a result, it was found that females showed significantly higher levels of four measured variances (SFF0 $SD^{**}$, SFF0 $range^{***}$, Min $SFF0^{***}$ and Max $SFF0^{***}$) than males on spontaneous speech. However, it was found that there was no significant difference between males and females in SFF0 range on reading or in SFF0 SD and SFF0 range on counting. It was found that there was no significant difference between males and females in the level of measured variances of vocal intensity depending on utterance conditions. Finally, this study made a comparison and analysis on differences in the variances of SFF0 and vocal intensity among utterance conditions. As a result, it was found that all the measured variances of SFF0 in males were most significantly reduced depending upon spontaneous speech which was followed by reading and counting respectively (SFF0 SD: p<.001, SFF0 range: p<.05, Max SFF0: p<.05). Females however, show no significant difference in the measured variances of SFF0 depending upon three utterance conditions. It was also found that the measured variances of vocal intensity in females were most significantly reduced depending on spontaneous speech that was followed by reading and counting (VI SD: p<.001, VI range: p<.001, Min VI: p<.01 Max VI: p<.05), while males showed no significant difference in the measured variances of vocal intensity depending on three utterance conditions. In sum, these findings suggest that variances of SFF0 in males are affected by three utterance conditions, while variances of vocal intensity in females are affected by three utterance conditions.

• 대한음성학회지:말소리
• /
• 제60호
• /
• pp.165-180
• /
• 2006

In this paper, we present several confidence measures (CM) for speech recognition systems to evaluate the reliability of recognition results. We propose heuristic CMs such as mean log-likelihood score, N-best word log-likelihood ratio, likelihood sequence fluctuation and likelihood ratio testing(LRT)-based CMs using several types of anti-models. Furthermore, we propose new algorithms to add weighting terms on phone-level log-likelihood ratio to merge word-level log-likelihood ratios. These weighting terms are computed from the distance between acoustic models and knowledge-based phoneme classifications. LRT-based CMs show better performance than heuristic CMs excessively, and LRT-based CMs using phonetic information show that the relative reduction in equal error rate ranges between $8{\sim}13%$ compared to the baseline LRT-based CMs. We use the support vector machine to fuse several CMs and improve the performance of utterance verification. From our experiments, we know that selection of CMs with low correlation is more effective than CMs with high correlation.

• 영어어문교육
• /
• 제9권spc호
• /
• pp.149-165
• /
• 2003

This paper is aimed at the investigation of pragmatic contributions to the identification of explicatures. An explicature is the result of fleshing out the semantic representation of an utterance. The basic assumption of the paper is that the process of the developing the semantic representation into an explicature depends heavily on contextual information. Therefore, we are concerned with the way in which hearers use contextual information to flesh rut or develop the semantic representation of an utterance. The identification of explicatures includes both the recovery of the proposition expressed and the recovery of what we called higher-level explicatures. There are three subtasks involved in the recovery of the proposition expressed: reference assignment disambiguation and enrichment On the other hand, there are two subtasks involved in the recovery of higher-level explicatures: attitudes and speech acts.

• 비교문화연구
• /
• 제22권
• /
• pp.303-327
• /
• 2011

This study aims to uncover the learner's language proficiency shown in the writing data of Chinese elementary/intermediate level learners. Language proficiency of the learners acquired by error analysis provides only partial information, and thus this study analyses the interlanguage of Korean learners in terms of 'Mean Length of Utterance, MLU' to discover the overall aspect of learner's language proficiency more symmetrically. The analysis of vocabulary area is to be enforced after generally studying the learner's language development aspect in accordance with MLU-m(orpheme) and MLU-(w)ord found in compositions by Chinese speaking Korean language learners. In terms of MLU, it has been slightly increased as the level of proficiency between elementary level and intermediate level learners; however, the morpheme seemed to be difficult to use, since the difference between Chinese learners and Korean university students has been notably shown. Vocabulary diversity, using aspect for each word class, and using aspect of the predicate are studied for vocabulary area; more various and numerous vocabulary tend to be used as the level of proficiency increases. In terms of predicate use, Chinese learners use less numerous vocabulary types.

• 한국음향학회지
• /
• 제22권3호
• /
• pp.194-201
• /
• 2003

음성인식에서 신뢰도 측정이란 인식된 결과에 대한 신뢰 여부를 결정하는 것이다. 신뢰도는 프레임을 음소 및 단어 수준으로 통합하여 측정된다. 단어 인식의 경우, 신뢰도를 이용하여 인식 결과와 미등록 어휘를 검증한다. 따라서 이러한 후처리를 통해 이를 인식 결과로 승인하지 않음으로써 성능을 높일 수 있다. 본 논문에서는 기존의 신뢰도 측정 방법인 로그 유사도 비를 수정하여 신뢰도를 측정하였다. 제안된 방법은 프레임 수준에서 음소 수준으로 신뢰도를 통합할 때 로그 유사도 비가 음수인 것만을 누적하는 것이다. 단어 인식기의 인식 결과에 대한 검증 성능을 기존의 방법과 비교한 결과, CAR (Correct Acceptance Ratio)이 90%인 지점에서 FAR (False Acceptance Ratio)을 미등록 어휘에 대해서는 약 3.49%, 오인식에 대해서는 15.25% 감소시킬 수 있었다

• 전자공학회논문지B
• /
• 제33B권3호
• /
• pp.159-168
• /
• 1996

In this paper, the natural language dialogue understanding sytem, based on discourse information and plan recognition, is designed and implemented. The system needs to analyze the user's input utterance and acquire the discoruse information to perform plan recognition and facilitate cooperative response. This paper proposes the mehtod of controlling a dialogue, based on the algorithm for extracting the discourse information. When the discourse information for dialogue understanding is extracted, the information-based value in feature structure that is obtained form korean parser is used. And the system makes use of the structure. Thus it can offer the response that the user wants to take, and let the dialogue to study in utterance level and enhance the efficiency of dialogue understanding. In this paper, we apply the system to the hotel reservation domain and show the mehtod of using the discoruse information to control the dialogue.