• The Journal of the Acoustical Society of Korea
• /
• v.16 no.5
• /
• pp.50-59
• /
• 1997

In this paper, we implement an automatic speech segmentation and labeling system which marks phone boundaries automatically for constructing the Korean speech database. We specify and implement the system based on conventional speech segmentation and labeling techniques, and also develop the graphic user interface(GUI) on Hangul $Motif^{TM}$ environment for the users to examine the automatic alignment boundaries and to refine them easily. The developed system is applied to 16kHz sampled speech, and the labeling unit is composed of 46 phoneme-like units(PLUs) and silence. The system uses both of the phonetic and orthographic transcription as input methods of linguistic information. For pattern-matching method, hidden Markov models(HMM) is employed. Each phoneme model is trained using the manually segmented 445 phonetically balanced word (PBW) database. In order to evaluate the performance of the system, we test it using another database consisting of sentence-type speech. According to our experiment, 74.7% of phoneme boundaries are within 20ms of the true boundary and 92.8% are within 40ms.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.40 no.5
• /
• pp.266-279
• /
• 2003

In this paper, we propose machine printed character recognition methods which utilize the character type information and divide the character clusters. The characters are subdivided into a total of seven types, of which six types are for Hangul according to the grapheme combination fashions and one type for English characters, numerals, and symbols. According to the character type, we separate input character image into several recognition units and recognize them by using the direction angle feature. The recognition for each character type is completed by combining recognition units which are recognized by neural networks respectively For combining a total of seven character recognizers, we implemented seven methods such as switching method, integrating method, and their several variants. As experimental results, we obtained 98.2% recognition rate of simple switching method, 90.54% of integrating one, and between 97.35% and 98.65% of five variants.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.3
• /
• pp.600-611
• /
• 1996

Recently, neural network-based speech recognition has been studied to utilize the adaptivity and learnability of neural network models. However, conventional neural network models have difficulty in the co-articulation processing and the boundary detection of similar phonmes of the Korean speech. Also, in case of using one phonotopic map, learning speed may dramatically increase and inaccuracies may be caused because homogeneous learning and recognition method should be applied for heterogenous data. Hence, in this paper, a neural net typewriter has been designed using a hierarchical self-organizing classifier(HSOC), and related algorithms are presented. This HSOC, during its learing stage, distributed phoneme data on hierarchically structured multiple phonotopic maps, using Kohonen's self-organizing feature maps(SOFM). Presented and experimented in this paper were the algorithms for deciding the number of maps, map sizes, the selection of phonemes and their placement per map, an approapriate learning and preprocessing method per map. If maps are divided according to a priorlinguistic knowledge, we would have difficulty in acquiring linguistic knowledge and how to alpply it(e.g., processing extended phonemes). Contrarily, our HSOC has an advantage that multiple phonotopic maps suitable for given input data are self-organizable. The resulting three korean phonotopic maps are optimally labelled and have their own optimal preprocessing schemes, and also confirm to the conventional linguistic knowledge.

• Journal of Korea Society of Industrial Information Systems
• /
• v.21 no.5
• /
• pp.19-30
• /
• 2016

In these days, people get information from internet easily. However, there are too many information. It makes interrupt and inefficient for searching data. Therefore, we need user customized web search system which provides appropriate information. In this paper, we propose a searching system that can collect semi-automatically conditions of users such as weather, location and time and provide essential information to users. Using these context data, the proposed system can understand what information users want in specific situations and can provide more useful information to users than existing systems. The proposed system based on 'Production/Sharing Service of Personal Korean Contents with Voluntary Sharing Economy System' and we add data parsing algorithm in each input, store and search part. In the experiments, we compare and analyze the results of existing system and the proposed system using some general key words.

• Annual Conference on Human and Language Technology
• /
• 1996.10a
• /
• pp.158-162
• /
• 1996

본 논문에서는 문자출력이 가능한 무선호출기에 음성인식 기술을 접목한, 특성화된 한 음성인식 시스템에 대하여 설명하고자 한다. 시스템 동작 과정은, 일단 호출자가 음성인식 서버와 접속하게 되면 서버는 호출자의 자연스런 입력음성을 인식, 그 결과를 문장 형태로 피호출자의 호출기 단말기에 출력시키는 방식으로 되어 있다. 본 시스템에서는 통계적 음성인식 기법을 도입하여, 각 단어를 연속 HMM으로 모델링하였다. 가우시안 혼합 확률밀도함수를 사용하는 각 모델은 전통적인 HMM 학습법들 중의 하나인 Baum-Welch 알고리듬에 의해 학습되고 인식시에는 이들에 비터비 빔 탐색을 적용하여 최선의 결과를 얻도록 한다. MFCC와 파워를 혼용한 26 차원 특징벡터를 각 프레임으로부터 추출하여, 최종적으로, 83 개의 도메인 어휘들 및 무음과 같은 특수어휘들에 대한 모델링을 완성하게 된다. 여기에 구문론적 기능과 의미론적 기능을 함께 수행하는 FSN을 결합시켜 자연발화음성에 대한 연속음성인식 시스템을 구성한다. 본문에서는 이상의 사항들 외에도 음성 데이터베이스, 레이블링 등과 갈이 시스템 성능과 직결되는 시스템의 외적 요소들에 대해 고찰하고, 시스템에 구현되어 있는 다양한 특성들에 대해 밝히며, 실험 결과 및 앞으로의 개선 방향 등에 대해 논의하기로 한다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.197-202
• /
• 2019

토큰화는 입력 텍스트를 더 작은 단위의 텍스트로 분절하는 과정으로 주로 기계 학습 과정의 효율화를 위해 수행되는 전처리 작업이다. 현재까지 자연어 처리 분야 과업에 적용하기 위해 다양한 토큰화 방법이 제안되어 왔으나, 주로 텍스트를 효율적으로 분절하는데 초점을 맞춘 연구만이 이루어져 왔을 뿐, 한국어 데이터를 대상으로 최신 기계 학습 기법을 적용하고자 할 때 적합한 토큰화 방법이 무엇일지 탐구 해보기 위한 연구는 거의 이루어지지 않았다. 본 논문에서는 한국어 데이터를 대상으로 최신 기계 학습 기법인 전이 학습 기반의 자연어 처리 방법론을 적용하는데 있어 가장 적합한 토큰화 방법이 무엇인지 알아보기 위한 탐구 연구를 진행했다. 실험을 위해서는 대표적인 전이 학습 모형이면서 가장 좋은 성능을 보이고 있는 모형인 BERT를 이용했으며, 최종 성능 비교를 위해 토큰화 방법에 따라 성능이 크게 좌우되는 과업 중 하나인 기계 독해 과업을 채택했다. 비교 실험을 위한 토큰화 방법으로는 통상적으로 사용되는 음절, 어절, 형태소 단위뿐만 아니라 최근 각광을 받고 있는 토큰화 방식인 Byte Pair Encoding (BPE)를 채택했으며, 이와 더불어 새로운 토큰화 방법인 형태소 분절 단위 위에 BPE를 적용하는 혼합 토큰화 방법을 제안 한 뒤 성능 비교를 실시했다. 실험 결과, 어휘집 축소 효과 및 언어 모델의 퍼플렉시티 관점에서는 음절 단위 토큰화가 우수한 성능을 보였으나, 토큰 자체의 의미 내포 능력이 중요한 기계 독해 과업의 경우 형태소 단위의 토큰화가 우수한 성능을 보임을 확인할 수 있었다. 또한, BPE 토큰화가 종합적으로 우수한 성능을 보이는 가운데, 본 연구에서 새로이 제안한 형태소 분절과 BPE를 동시에 이용하는 혼합 토큰화 방법이 가장 우수한 성능을 보임을 확인할 수 있었다.

• Journal of Practical Engineering Education
• /
• v.14 no.3
• /
• pp.491-498
• /
• 2022

In this paper, as a liberal arts course for non-major students, a case study of recurrent neural network SW practice, which is essential for designing a basic recurrent neural network subject curriculum, was developed. The developed SW practice case focused on understanding the operation principle of the recurrent neural network, and used a spreadsheet to check the entire visualized operation process. The developed recurrent neural network practice case consisted of creating supervised text completion training data, implementing the input layer, hidden layer, state layer (context node), and output layer in sequence, and testing the performance of the recurrent neural network on text data. The recurrent neural network practice case developed in this paper automatically completes words with various numbers of characters. Using the proposed recurrent neural network practice case, it is possible to create an artificial intelligence SW practice case that automatically completes by expanding the maximum number of characters constituting Korean or English words in various ways. Therefore, it can be said that the utilization of this case of basic practice of recurrent neural network is high.

• Journal of Internet Computing and Services
• /
• v.21 no.4
• /
• pp.87-95
• /
• 2020

The Science and Technology Information Service provided by the Korea Institute of Science and Technology Information (KISTI) is a service designed to allow users to easily and conveniently search and view content that is built similar to the general information service. NDSL is KISTI's core science, technology and information service, providing about 138 million content and having about 93 million page views in a year of 2019. In this paper, various insights were derived through the analysis of how science and technology information such as academic papers, reports and patents provided by NDSL is searched and utilized through web services (https://www.ndsl.kr) and search query words. In addition to general statistics such as the status of content construction, utilization status and utilization methods by type of content, monthly/weekly/time-of-day content usage, content view rate per one-time search by content type, the comparison of the use status of academic papers by year, the relationship between the utilization of domestic academic papers and the KCI index we analyzed the usability of each content type, such as academic papers and patents. We analyzed query words such as the language form of query words, the number of words of query words, and the relationship between query words and timeliness by content type. Based on the results of these analyses, we would like to propose ways to improve the service. We suggest that NDSL improvements include ways to dynamically reflect the results of content utilization behavior in the search results rankings, to extend query and to establish profile information through non-login user identification for targeted services.

• Journal of Internet Computing and Services
• /
• v.22 no.1
• /
• pp.109-135
• /
• 2021

his study classified the development process of artificial intelligence (AI) speakers through analysis of the news text of artificial intelligence (AI) speakers shown in traditional news reports, and identified the characteristics of each product by period. The theoretical background used in the analysis are news frames and topic frames. As analysis methods, topic modeling and semantic network analysis using the LDA method were used. The research method was a content analysis method. From 2014 to 2019, 2710 news related to AI speakers were first collected, and secondly, topic frames were analyzed using Nodexl algorithm. The result of this study is that, first, the trend of topic frames by AI speaker provider type was different according to the characteristics of the four operators (communication service provider, online platform, OS provider, and IT device manufacturer). Specifically, online platform operators (Google, Naver, Amazon, Kakao) appeared as a frame that uses AI speakers as'search or input devices'. On the other hand, telecommunications operators (SKT, KT) showed prominent frames for IPTV, which is the parent company's flagship business, and 'auxiliary device' of the telecommunication business. Furthermore, the frame of "personalization of products and voice service" was remarkable for OS operators (MS, Apple), and the frame for IT device manufacturers (Samsung) was "Internet of Things (IoT) Integrated Intelligence System". The econd, result id that the trend of the topic frame by AI speaker development period (by year) showed a tendency to develop around AI technology in the first phase (2014-2016), and in the second phase (2017-2018), the social relationship between AI technology and users It was related to interaction, and in the third phase (2019), there was a trend of shifting from AI technology-centered to user-centered. As a result of QAP analysis, it was found that news frames by business operator and development period in AI speaker development are socially constituted by determinants of media discourse. The implication of this study was that the evolution of AI speakers was found by the characteristics of the parent company and the process of co-evolution due to interactions between users by business operator and development period. The implications of this study are that the results of this study are important indicators for predicting the future prospects of AI speakers and presenting directions accordingly.