• 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 Korean Information Science Society Conference
• /
• 2011.06c
• /
• pp.383-385
• /
• 2011

본 논문에서는 동영상을 대상으로 하는 기존의 시각주의 시스템의 성능을 향상시킨 새로운 시스템에 대하여 설명한다. 제안하는 시스템은 기존의 시스템이 가지고 있던 한계점인 서로 반대되는 특징을 가지는 색상에서 하나의 특징만을 고정적으로 선택하던 것을 극복하여, 서로 반대되는 특징 중 현저함이 더 높은 색상 특징을 선택하여 입력 들어오는 영상에 적응적인 현저함 추출을 하였다. 도한 시간 현저함 정보를 추가적으로 고려할 수 있도록 하여 동영상에 대한 처리도 가능하도록 하였고, 성능 평가 시 인간을 대상으로 한 설문 조사 실험을 추가하여 보다 인간의 시각 인식과 유사한 시스템임을 증명하였다.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.18 no.1
• /
• pp.133-138
• /
• 2008

In this paper, we propose a novel anchor shot detection system, named to MASD (Multi-phase Anchor Shot Detection), which is a core step of the preprocessing process for the news video analysis. The proposed system is composed of four modules and operates sequentially: 1) skin color detection module for reducing the candidate face regions; 2) face detection module for finding the key-frames with a facial data; 3) vector representation module for the key-frame images using a non-negative matrix factorization; 4) one class SVM module for determining the anchor shots using a support vector data description. Besides the qualitative analysis, our experiments validate that the proposed system shows not only the comparable accuracy to the recently developed methods, but also more faster detection rate than those of others.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2015.11a
• /
• pp.16-17
• /
• 2015

본 논문에서는 실내 환경 중 TV 시청환경에서 마이크로폰 어레이를 이용하여 다양한 다중 음원 방향을 추정하는 기법을 제안한다. 제안된 기법은 기존의 하나의 음원에 특화되어 있는 GCC-PHAT 기반의 방법을 GCC-PHAT 버퍼와 NMF를 도입하여 다중음원의 방향 추정을 가능하게 만들었다. 제안된 기법의 성능을 평가하기 위해서 실 거주 환경에서 발생하는 소음원과 TV 소리 방향 추정 결과에 대한 실측치와 추정치 간의 오차인 절대 평균오차를 측정하였으며, 실험 결과 제안한 기법이 기존의 방법인 GCC-PHAT보다 우수한 추정 성능을 보임을 확인하였다.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.18 no.9
• /
• pp.2155-2160
• /
• 2014

In this paper, we proposes a new document summarization method using the extracted semantic feature which the semantic feature is extracted by distributed parallel processing based Hadoop. The proposed method can well represent the inherent structure of documents using the semantic feature by the non-negative matrix factorization (NMF). In addition, it can summarize the big data document using Hadoop. The experimental results demonstrate that the proposed method can summarize the big data document which a single computer can not summarize those.

• Annual Conference of KIPS
• /
• 2009.11a
• /
• pp.715-716
• /
• 2009

인터넷의 발전은 대량의 정보를 양산하였고, 이러한 대량의 정보 집합 내에서는 비슷한 정보가 재활용 되거나 반복되는 정보중복문제를 가지고 있다. 중복되는 정보들로부터 사용자에게 원하는 정보를 신속히 검색할 수 있도록 하는 정보 요약에 대한 필요성은 점차 증가하고 있다. 본 논문은 비음수 행렬 인수분해(NMF, non-negative matrix factorization)에 의한 문서의 의미특징을 이용하여 주제기반의 다중문서를 요약하는 새로운 방법을 제안한다. 본 논문에서는 다중문서가 포함하고 있는 문서들 간의 고유구조를 문서요약에 이용하여서 요약의 질을 높일 수 있고, 주제와 문장 간의 유사성과 다양성 고려하여서 쉽게 과잉정보를 제거하여 문장을 요약할 수 있는 장점을 갖는다.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.40 no.4
• /
• pp.619-627
• /
• 2015

This paper presents a speech enhancement method using non-negative matrix factorization (NMF). In the training phase, each basis matrix of source signal is obtained from a proper database, and these basis matrices are utilized for the source separation. In this case, the performance of speech enhancement relies heavily on the basis matrix. The proposed method for which speech basis matrix is made a high reconstruction error for noise signal shows a better performance than the standard NMF which basis matrix is trained independently. For comparison, we propose another method, and evaluate one of previous method. In the experiment result, the performance is evaluated by perceptual evaluation speech quality and signal to distortion ratio, and the proposed method outperformed the other methods.

• The Journal of Information Systems
• /
• v.31 no.1
• /
• pp.271-287
• /
• 2022

Purpose This research aims to study the changes in recruitment needed for the growth and survival of companies in the rapidly changing industry. In particular, we built a real company's worklist accounting for the rapidly advancing data-driven digital transformation, and presented the capabilities and conditions required for work. Design/methodology/approach we selected 37 jobs based on NCS to develop the employment search requirements by analyzing the business characteristics and work capabilities of the industry and company. The business specification indicators were converted into a matrix through the TF-IDF process, and the NMF algorithm is used to extract the features of each document. Also, the cosine distance measurement method is utilized to determine the similarity of the job specification conditions. Findings Companies tended to prefer "IT competency," which is a specification related to computer use and certification, and "experience competency," which is a specification for experience and internship. In addition, 'foreign language competency' was additionally preferred depending on the job. This analysis and development of job requirements would not only help companies to find the talents but also be useful for the jobseekers to easily decide the priority of their specification activities.

• Journal of KIISE:Software and Applications
• /
• v.33 no.5
• /
• pp.481-488
• /
• 2006

The performance of face recognition methods using subspace projection is directly related to the characteristics of their basis images, especially in the cases of local distortion or partial occlusion. In order for a subspace projection method to be robust to local distortion and partial occlusion, the basis images generated by the method should exhibit a part-based local representation. We propose an effective part-based local representation method named locally salient ICA (LS-ICA) method for face recognition that is robust to local distortion and partial occlusion. The LS-ICA method only employs locally salient information from important facial parts in order to maximize the benefit of applying the idea of 'recognition by parts.' It creates part-based local basis images by imposing additional localization constraint in the process of computing ICA architecture I basis images. We have contrasted the LS-ICA method with other part-based representations such as LNMF (Localized Non-negative Matrix Factorization) and LFA (Local Feature Analysis). Experimental results show that the LS-ICA method performs better than PCA, ICA architecture I, ICA architectureII, LFA, and LNMF methods, especially in the cases of partial occlusions and local distortions.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.38C no.6
• /
• pp.503-511
• /
• 2013

This paper presents a speech enhancement method using non-negative matrix factorization. In training phase, we can obtain each basis matrix from speech and specific noise database. After training phase, the noisy signal is separated from the speech and noise estimate using basis matrix in enhancement phase. In order to improve the performance, we model the change of encoding matrix from training phase to enhancement phase using independent Gaussian distribution models, and then use the constraint of the objective function almost same as that of the above Gaussian models. Also, we perform a smoothing operation to the encoding matrix by taking into account previous value. Last, we apply the Log-Spectral Amplitude type algorithm as gain function.