• Journal of Computing Science and Engineering
• /
• v.4 no.1
• /
• pp.1-22
• /
• 2010

A field of spoken dialog systems is a rapidly growing research area because the performance improvement of speech technologies motivates the possibility of building systems that a human can easily operate in order to access useful information via spoken languages. Among the components in a spoken dialog system, the dialog management plays major roles such as discourse analysis, database access, error handling, and system action prediction. This survey covers design issues and recent approaches to the dialog management techniques for modeling the dialogs. We also explain the user simulation techniques for automatic evaluation of spoken dialog systems.

• MALSORI
• /
• no.50
• /
• pp.111-122
• /
• 2004

Automatic detection of laryngeal diseases by voice is attractive because of its non-intrusive nature. Cepstrum based approach to detect laryngeal cancer shows reliable performance even when the periodicity of voice signals is severely lost, but it has a drawback that it is not robust to channel mismatch due to different microphone characteristics. In this paper, to deal with mismatched training and test microphone conditions, we investigate channel compensation techniques such as Cepstral Mean Subtraction (CMS) and Pole Filtered CMS (PFCMS). According to our experiments, PFCMS yields better performance than CMS. By using PFCMS, we obtained 12% and 40% error reduction over baseline and CMS, respectively.

• ETRI Journal
• /
• v.38 no.3
• /
• pp.487-493
• /
• 2016

Two new methods are proposed for an unsupervised adaptation of a language model (LM) with a single sentence for automatic transcription tasks. At the training phase, training documents are clustered by a method known as Latent Dirichlet allocation (LDA), and then a domain-specific LM is trained for each cluster. At the test phase, an adapted LM is presented as a linear mixture of the now trained domain-specific LMs. Unlike previous adaptation methods, the proposed methods fully utilize a trained LDA model for the estimation of weight values, which are then to be assigned to the now trained domain-specific LMs; therefore, the clustering and weight-estimation algorithms of the trained LDA model are reliable. For the continuous speech recognition benchmark tests, the proposed methods outperform other unsupervised LM adaptation methods based on latent semantic analysis, non-negative matrix factorization, and LDA with n-gram counting.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2022.05a
• /
• pp.279-282
• /
• 2022

컴퓨터가 마이크 등의 소리 센서를 통해 얻은 음향학적 신호를 단어나 문장으로 변환시키는 기술인 음성 인식 기술과 인공지능 기술을 결합한 음성 대화 시스템에 대한 연구 진행 및 제품 출시가 활발하게 이루어지고 있다. 기존의 시스템을 사용하면서 날짜와 시간 외의 정보 추출 정도가 빈약하거나 자동 등록이 되지 않는 문제점을 확인하였다. 음성 인식 기술을 통해 얻은 텍스트에서 보다 많은 정보를 추출하고, 자동 등록 및 알림과 맛집 등 추가 정보 제공 시스템을 구축하는 것을 목표로 하였다.

• International Journal of Computer Science & Network Security
• /
• v.22 no.4
• /
• pp.420-426
• /
• 2022

Breast cancer is among the cancers that may be healed as the disease diagnosed at early times before it is distributed through all the areas of the body. The Automatic Analysis of Diagnostic Tests (AAT) is an automated assistance for physicians that can deliver reliable findings to analyze the critically endangered diseases. Deep learning, a family of machine learning methods, has grown at an astonishing pace in recent years. It is used to search and render diagnoses in fields from banking to medicine to machine learning. We attempt to create a deep learning algorithm that can reliably diagnose the breast cancer in the mammogram. We want the algorithm to identify it as cancer, or this image is not cancer, allowing use of a full testing dataset of either strong clinical annotations in training data or the cancer status only, in which a few images of either cancers or noncancer were annotated. Even with this technique, the photographs would be annotated with the condition; an optional portion of the annotated image will then act as the mark. The final stage of the suggested system doesn't need any based labels to be accessible during model training. Furthermore, the results of the review process suggest that deep learning approaches have surpassed the extent of the level of state-of-of-the-the-the-art in tumor identification, feature extraction, and classification. in these three ways, the paper explains why learning algorithms were applied: train the network from scratch, transplanting certain deep learning concepts and constraints into a network, and (another way) reducing the amount of parameters in the trained nets, are two functions that help expand the scope of the networks. Researchers in economically developing countries have applied deep learning imaging devices to cancer detection; on the other hand, cancer chances have gone through the roof in Africa. Convolutional Neural Network (CNN) is a sort of deep learning that can aid you with a variety of other activities, such as speech recognition, image recognition, and classification. To accomplish this goal in this article, we will use CNN to categorize and identify breast cancer photographs from the available databases from the US Centers for Disease Control and Prevention.

• The Journal of the Acoustical Society of Korea
• /
• v.39 no.2
• /
• pp.131-136
• /
• 2020

Wake Up Word (WUW) is a short utterance used to convert speech recognizer to recognition mode. The WUW defined by the user who actually use the speech recognizer is called user-defined WUW. In this paper, to recognize user-defined WUW, we construct traditional Gaussian Mixture Model-Hidden Markov Model (GMM-HMM), Linear Discriminant Analysis (LDA)-GMM-HMM and LDA-Deep Neural Network (DNN)-HMM based system and compare their performances. Also, to improve recognition accuracy of the WUW system, a threshold method is applied to each model, which significantly reduces the error rate of the WUW recognition and the rejection failure rate of non-WUW simultaneously. For LDA-DNN-HMM system, when the WUW error rate is 9.84 %, the rejection failure rate of non-WUW is 0.0058 %, which is about 4.82 times lower than the LDA-GMM-HMM system. These results demonstrate that LDA-DNN-HMM model developed in this paper proves to be highly effective for constructing user-defined WUW recognition system.

• Phonetics and Speech Sciences
• /
• v.12 no.3
• /
• pp.25-32
• /
• 2020

An analysis of emotional English utterances with 7 emotions (calm, happy, sad, angry, fearful, disgust, surprised) was conducted using the measurement of prosodic distance between 672 emotional and 48 neutral utterances. Applying the technique proposed in the automatic evaluation model of English pronunciation to the present study on emotional utterances, Euclidean distance measurement of 3 prosodic elements such as F0, intensity and duration extracted from emotional and neutral utterances was utilized. This paper, furthermore, extended the analytical methods to include Euclidean distance normalization, z-score and z-score normalization resulting in 4 groups of measurement schemes (sqrF0, sqrINT, sqrDUR; norsqrF0, norsqrINT, norsqrDUR; sqrzF0, sqrzINT, sqrzDUR; norsqrzF0, norsqrzINT, norsqrzDUR). All of the results from perceptual analysis and acoustical analysis of emotional utteances consistently indicated the greater effectiveness of norsqrF0, norsqrINT and norsqrDUR, among 4 groups of measurement schemes, which normalized the Euclidean measurement. The greatest acoustical change of prosodic information influenced by emotion was shown in the values of F0 followed by duration and intensity in descending order according to the effect size based on the estimation of distance between emotional utterances and neutral counterparts. Tukey Post Hoc test revealed 4 homogeneous subsets (calm

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2014.10a
• /
• pp.150-155
• /
• 2014

본 논문에서는, 원거리 음성인식을 위한 서브밴드 필터링 기반의 빠르고 효율적인 에코제거 시스템을 제안한다. 제안하는 에코제거 시스템은 우선 채널간 유사도 (correlation) 가 높을 경우 적응필터가 오작동하는 것을 방지하기 위해 spatial decorrelation 을 적용하게 된다. 그리고 tree 형태를 가지는 IIR filterbank 기반의 subband 구조를 채택함으로써, 적은 차수로도 효과적인 analysis, synthesis 필터링을 수행할 수 있도록 한다. 이 과정에서 불가피하게 발생하는 서브 밴드간 spectral aliasing은 notch filter를 적용해 해결할 수 있다. 또한 적응 필터로는 improved proportionate normalized least-mean-square (IP-NLMS) 알고리즘을 사용해 수렴속도 및 에코제거 성능에서 우수함을 확인하였다. 마지막으로 decision-directed estimation 기반의 residual echo suppressor를 적용해 잔여 에코를 제거하게 된다. 본 논문에서는 각 단계를 구성하게 된 이론적인 배경을 소개하고, 실제 에코가 존재하는 환경에서 ERLE, 원거리 음성 인식률, computational complexity를 통해 제안하는 에코제거 시스템의 효과를 입증하도록 한다.

• Journal of the Korean Society of Laryngology, Phoniatrics and Logopedics
• /
• v.30 no.2
• /
• pp.112-117
• /
• 2019

Background and Objectives The purpose of this study was to develop the differential diagnosis scale containing items from adductor spasmodic dysphonia (ADSD) to muscle tension dysphonia (MTD) and the determine clinical utility of newly developed items. Materials and Method The four parts of pitch, redirected phonation, automatic speech and voiced sound were selected for analyzing the characteristics of ADSD in the literature. One part of tense voiceless sound was developed according to the Korean manner of articulation. The content validity was evaluated based on 5 scales (1-5 point) analysis from 30 experts. One hundred patients (50 ADSD and 50 MTD) were recorded in reading a sentence and sustained phonation. The two speech language pathologist evaluated recorded voices through a blind test using 4 scales (0-3 point) for newly developed items. Results As a result of verifying the content validity of items with experts, it was identified that the differentiated items were valid with 4.2 out of 5. Through the differential diagnosis between two groups according to the items, the correlation between sub-domains and total scores was shown as higher than 0.710. The result of analyzing the reliability on each diagnosis domain was 0.840-0.893, which showed the internal consistency of items was great. Newly developed five parts of ADSD were significantly higher than those of MTD with strong correlation (p<0.01). The reliability among the evaluators was analyzed as high with 0.892. Conclusion In this study, the differential diagnosis scale of ADSD was revealed as having validity and reliability. It is considered that it will be useful for differentiating ADSD and MTD in the clinical field.

• The Journal of the Convergence on Culture Technology
• /
• v.9 no.3
• /
• pp.731-736
• /
• 2023

Recently, the medical field has been applying mandatory Electronic Medical Records (EMRs) and Electronic Health Records (EHRs) systems that computerize and manage medical records, and distributing them throughout the entire medical industry to utilize patients' past medical records for additional medical procedures. However, the conversations between medical professionals and patients that occur during general medical consultations and counseling sessions are not separately recorded or stored, so additional important patient information cannot be efficiently utilized. Therefore, we propose an electronic medical record system that uses speech recognition and natural language processing deep learning to store conversations between medical professionals and patients in text form, automatically extracts and summarizes important medical consultation information, and generates electronic medical records. The system acquires text information through the recognition process of medical professionals and patients' medical consultation content. The acquired text is then divided into multiple sentences, and the importance of multiple keywords included in the generated sentences is calculated. Based on the calculated importance, the system ranks multiple sentences and summarizes them to create the final electronic medical record data. The proposed system's performance is verified to be excellent through quantitative analysis.