• Proceedings of the KSPS conference
• /
• 1997.07a
• /
• pp.81-91
• /
• 1997

This paper deals with the problems of English listening comprehension: the rate of understanding difference in positions and sentence structures, parts of speech easily missed to understand, English sounds only in English(not in Korean), confusion of sounds, unaccented prefixes and suffixes, polysemy, homonym, juncture, understanding as one word by two different words, and sound blending in a normal speed of connected speech. Bearing those in mind I taught Suwon University freshmen video English with the mixed idea of Peterson's bottom-up and top-down methods putting in a meaningful context with thought group rather than word to word understanding. As a consequence, their errors come: prepositions, conjunctions, unstressed prefixes and suffixes, -ing from the present progressives and so forth. Assignments to have students transcribe the TV commercials and the names of reporters or Korean related news from English broadcastings are of use and help.

• Phonetics and Speech Sciences
• /
• v.3 no.3
• /
• pp.173-179
• /
• 2011

The present study was performed to investigate the intelligibility of voice disorder patients by providing the various background noise levels. Four sets of 12-sentence-stimuli produced by 11 voice disorder patients were prepared, and 5 minute-news from radio broadcasting studio were used as a background noise. 30 listeners assigned intelligibility score of each sentence with visual analog scale. Each set of sentences was provided with 20dB, 10dB, 0dB noise (same intensity with stimuli), and, finally, with no noise. As results, as background noise level increased, intelligibility scores were lowered with statistical significance. Even though in the same severity, more loud background noise showed much lower scores than less loud noise. When 10dB noise was provided, intelligibility scores showed the biggest difference among the degree of severity.

• Proceedings of the KSPS conference
• /
• 2002.11a
• /
• pp.123-126
• /
• 2002

In this paper, we describe our LVCSR system for Korean broadcast news transcription. The main focus is to find the most proper morpheme-based lexical model for Korean broadcast news recognition to deal with the inflectional flexibilities in Korean. There are trade-offs between lexicon size and lexical coverage, and between the length of lexical unit and WER. In our system, we analyzed the training corpus to obtain a small 24k-morpheme-based lexicon with 98.8％ coverage. Then, the lexicon is optimized by combining morphemes using statistics of training corpus under monosyllable constraint or maximum length constraint. In experiments, our system reduced the number of monosyllable morphemes from 52％ to 29％ of the lexicon and obtained 13.24％ WER for anchor and 24.97％ for reporter.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2022.01a
• /
• pp.283-284
• /
• 2022

본 논문은 TTS를 사용해 사용자들에게 E-Book 및 뉴스를 보고 들을 수 있는 기능을 제공한다. 사용자 및 개발자가 직접 녹음한 TTS를 사용해 원하는 목소리, 배속과 같은 기능을 제공한다. 기존 TTS를 사용한 E-Book 사이트들은 광고가 많아 가독성의 문제와 유료 서비스인 반면에 본 논문에서 제안한 웹은 다양한 연령층의 사용자들이 사용하기 쉽게 메뉴의 간편화를 통해 다양한 E-Book 및 뉴스 기능을 제공함으로써 보다 직관적이고 쉽게 전자문서를 읽을 수 있도록 하였다.

• MALSORI
• /
• no.43
• /
• pp.1-9
• /
• 2002

This study aims to model Korean prosodic phrasing using CART(classification and regression tree) method. Our data are limited to Korean read speech. We used 400 sentences made up of editorials, essays, novels and news scripts. Professional radio actress read 400sentences for about two hours. We used K-ToBI transcription system. For technical reason, original break indices 1,2 are merged into AP. Differ from original K-ToBI, we have three break index Zero, AP and IP. Linguistic information selected for this study is as follows: the number of syllables in ‘Eojeol’, the location of ‘Eojeol’ in sentence and part-of-speech(POS) of adjacent ‘Eojeol’s. We trained CART tree using above information as variables. Average accuracy of predicting NonIP(Zero and AP) and IP was 90.4% in training data and 88.5% in test data. Average prediction accuracy of Zero and AP was 79.7% in training data and 78.7% in test data.

• 전자공학회논문지 IE
• /
• v.45 no.3
• /
• pp.20-26
• /
• 2008

Two independent component analysis(ICA) algorithms were applied for the recognition of speech signals corrupted by a car engine noise. Speech recognition was performed by hidden markov model(HMM) for the estimated signals and recognition rates were compared with those of orginal speech signals which are not corrupted. Two different ICA methods were applied for the estimation of speech signals, one of which is FastICA algorithm that maximizes negentropy, the other is information-maximization approach that maximizes the mutual information between inputs and outputs to give maximum independence among outputs. Word recognition rate for the Korean news sentences spoken by a male anchor is 87.85%, while there is 1.65% drop of performance on the average for the estimated speech signals by FastICA and 2.02% by information-maximization for the various signal to noise ratio(SNR). There is little difference between the methods.

• The Journal of the Acoustical Society of Korea
• /
• v.19 no.2
• /
• pp.101-105
• /
• 2000

A telegram has been used to transmit the emergency news or celebration message. So, it has been very important media in our life. Although the telegram processing is more and more convenient, on the other hand, the telegram service contains only text message. The voice telegram is that delivering user's voice with text message. So, the voice telegram can be delivered sender's emotions and feelings. However, since voice information contains lots of data, large memory size and high cost processor are needed to deliver itself. In this paper, we proposed a new speech waveform coding method that has low complexity and low cost implementation for the voice telegram system. First, we fixed one basic speech waveform per pitch period and measured the waveform similarity between basic and neighbor speech waveform. Second, if the similarity satisfied threshold values, we compress the neighbor speech waveform with pitch and magnitude value per pitch period and if not, we save speech waveform. When the compression is about 45%, we obtained about 4 point in MOS.

• Phonetics and Speech Sciences
• /
• v.11 no.4
• /
• pp.89-96
• /
• 2019

In this paper, we propose a deep learning architecture that can effectively detect speech segmentation in broadcast contents. We also propose a multi-scale time-dilated layer for learning the temporal changes of feature vectors. We implement several comparison models to verify the performance of proposed model and calculated the frame-by-frame F-score, precision, and recall. Both the proposed model and the comparison model are trained with the same training data, and we train the model using 32 hours of Korean broadcast data which is composed of various genres (drama, news, documentary, and so on). Our proposed model shows the best performance with F-score 91.7% in Korean broadcast data. The British and Spanish broadcast data also show the highest performance with F-score 87.9% and 92.6%. As a result, our proposed model can contribute to the improvement of performance of speech detection by learning the temporal changes of the feature vectors.

• Phonetics and Speech Sciences
• /
• v.10 no.3
• /
• pp.31-39
• /
• 2018

This paper aims to present a taxonomy of non-standard words (NSW) for developing a French text normalization system and to propose a method for evaluating this system based on a corpus. The proposed taxonomy of French NSWs consists of 13 categories, including 2 types of letter-based categories and 9 types of number-based categories. In order to evaluate the text normalization system, a representative test set including NSWs from various text domains, such as news, literature, non-fiction, social-networking services (SNSs), and transcriptions, is constructed, and an evaluation equation is proposed reflecting the distribution of the NSW categories of the target domain to which the system is applied. The error rate of the test set is 1.64%, while the error rate of the whole corpus is 2.08%, reflecting the NSW distribution in the corpus. The results show that the literature and SNS domains are assessed as having higher error rates compared to the test set.

• Proceedings of the KSPS conference
• /
• 1996.10a
• /
• pp.168-168
• /
• 1996

In this paper I will examine some concrete examples of the obstacles faced by non-native speakers of Japanese when learning the language. I will go on to suggest ways in which these obstacles may be overcome. Nowadays there are numerous Japanese language books available for non-native speakers. However, most of these introductory Japanese language books focus on topics such as pronunciation, accent and intonation. Notable, these introductory textbooks provide insufficient emphasis on prosodic features of the Japanese language. The Japanese language has been considered by many teachers as relatively easy compared to other languages, due to its simple phonetic structure. This may be a partial explanation of the reason why the teaching of prosodic features has generally been given insufficient emphasis. To teach Japanese efficiently at a university level I have combined an emphasis on the teaching of prosodic features together with my experience of television announcing. This has entailed using television news programmes and contemporary reading materials in my class. Using taped material I intend to describe a case-study of teaching of Japanese articulation.