• 한국전자통신학회논문지
• /
• 제5권4호
• /
• pp.363-371
• /
• 2010

특징벡터의 분류를 개선시켜 화자독립 음성인식의 오류율을 줄이려는 노력의 일환으로서, 우리는 MFCC의 추출에 있어서 푸리에 스펙트럼을 기울이는 방법이 미치는 효과를 연구한다. 음성신호에 FIR 필터링을 적용하는 효과의 조사도 병행된다. 제안된 방법은 두 가지 독립적인 방법에 의해 평가된다. 즉, 피셔의 차별함수에 의한 방법과 은닉 마코브 모델 및 퍼지 벡터양자화를 사용한 음성인식 오류율 조사 방법이다. 실험 결과, 적절한 파라미터의 선택에 의해 기존의 방법에 비해 10% 정도 낮은 인식 오류율이 얻어짐을 확인하였다.

• 한국정보과학회논문지:소프트웨어및응용
• /
• 제31권6호
• /
• pp.799-807
• /
• 2004

본 논문에서는 사람의 얼굴표정과 음성 속에 담긴 6개의 기본감정(기쁨, 슬픔, 화남, 놀람, 혐오, 공포)에 대한 특징을 추출하고 인식하고자 한다. 이를 위해 얼굴표정을 이용한 감정인식에서는 이산 웨이블렛 기반 다해상도 분석을 이용하여 선형판별분석기법으로 특징을 추출하고 최소 거리 분류 방법을 이용하여 감정을 인식한다. 음성에서의 감정인식은 웨이블렛 필터뱅크를 이용하여 독립적인 감정을 확인한 후 다중의사 결정 기법에 외해 감정인식을 한다. 최종적으로 얼굴 표정에서의 감정인식과 음성에서의 감정인식을 융합하는 단계로 퍼지 소속함수를 이용하며, 각 감정에 대하여 소속도로 표현된 매칭 감은 얼굴에서의 감정과 음성에서의 감정별로 더하고 그중 가장 큰 값을 인식 대상의 감정으로 선정한다.

• ETRI Journal
• /
• 제35권6호
• /
• pp.1105-1114
• /
• 2013

Owing to the large number of video programs available, a method for accessing preferred videos efficiently through personalized video summaries and clips is needed. The automatic recognition of user states when viewing a video is essential for extracting meaningful video segments. Although there have been many studies on emotion recognition using various user responses, electroencephalogram (EEG)-based research on preference recognition of videos is at its very early stages. This paper proposes classification models based on linear and nonlinear classifiers using EEG features of band power (BP) values and asymmetry scores for four preference classes. As a result, the quadratic-discriminant-analysis-based model using BP features achieves a classification accuracy of 97.39% (${\pm}0.73%$), and the models based on the other nonlinear classifiers using the BP features achieve an accuracy of over 96%, which is superior to that of previous work only for binary preference classification. The result proves that the proposed approach is sufficient for employment in personalized video segmentation with high accuracy and classification power.

• The Journal of the Acoustical Society of Korea
• /
• 제25권1E호
• /
• pp.20-24
• /
• 2006

In this paper, we focused on the real time implementation of a speech recognition system with medium size of vocabulary considering its application to a mobile phone. First, we developed the PC based variable vocabulary word recognizer having the size of program memory and total acoustic models as small as possible. To reduce the memory size of acoustic models, linear discriminant analysis and phonetic tied mixture were applied in the feature selection process and training HMMs, respectively. In addition, state based Gaussian selection method with the real time cepstral normalization was used for reduction of computational load and robust recognition. Then, we verified the real-time operation of the implemented recognition system on the TMS320C6201 EVM board. The implemented recognition system uses memory size of about 610 kbytes including both program memory and data memory. The recognition rate was 95.86% for ETRI 445DB, and 96.4%, 97.92%, 87.04% for three kinds of name databases collected through the mobile phones.

• 대한인간공학회지
• /
• 제17권1호
• /
• pp.67-77
• /
• 1998

The objective of this study is to extract information from electroencephalogram(EEG) signals with which we can discriminate mental states. Seven university students were participated in this study. Ten stimuli based on IAPS (International Affective Picture Systems) Were presented at random according to the experimental schedule. 8-channel ($O_1$, $O_2$, $F_3$, $F_4$, $F_7$, $F_8$, $FP_1$, and $FP_2$)EEG signals were recorded at a sampling rate of 204.8 Hz for visual stimuli and analyzed. After random ten sequential stimuli presentation, the subject subjectively assessed the stimulus by scaling from -5 to 5. If the stimulus was the best and the worst, it was scored 5 and -5, respectively. Only maximum and minimum scored-EEG signals within each subject were selected on the basis of subjectively assessment for analysis. EEG signals were transformed into feature objects based on scalar autoregressive model coefficients. They were classified with Discriminant Analysis for each channel. The features produced results with the best classification accuracy of 85.7 % in $O_1$ and $O_2$ for visual stimuli. This study could be extended to establish an algorithm which quantify and classify emotions evoked by visual stimulus using autoregressive models.

• 제어로봇시스템학회논문지
• /
• 제19권11호
• /
• pp.1022-1028
• /
• 2013

It is well knowns that based on the CSP (Common Spatial Pattern) algorithm, the linear projection of an EEG (Electroencephalography) signal can be made to spaces that optimize the discriminant between two patterns. Sharing disadvantages from linear time invariant systems, CSP suffers from the non-stationary nature of EEGs causing the performance of the classification in a BCI (Brain-Computer Interface) system to drop significantly when comparing the training data and test data. The author has suggested a simple idea based on the parallel model of CSP filters to improve the performance of BCI systems. The model was tested with a simple CSP algorithm (without any elaborate regularizing methods) and a perceptron learning algorithm as a classifier to determine the improvement of the system. The simulation showed that the parallel model could improve classification performance by over 10% compared to conventional CSP methods.

• 대한전자공학회:학술대회논문집
• /
• 대한전자공학회 2004년도 ICEIC The International Conference on Electronics Informations and Communications
• /
• pp.391-394
• /
• 2004

This paper focuses on the DSP implementation of an HMM-based speech recognizer that can handle several hundred words of vocabulary size as well as speaker independency. First, we develop an HMM-based speech recognition system on the PC that operates on the frame basis with parallel processing of feature extraction and Viterbi decoding to make the processing delay as small as possible. Many techniques such as linear discriminant analysis, state-based Gaussian selection, and phonetic tied mixture model are employed for reduction of computational burden and memory size. The system is then properly optimized and compiled on the TMS320C6711 DSP for real-time operation. The implemented system uses 486kbytes of memory for data and acoustic models, and 24.5kbytes for program code. Maximum required time of 29.2ms for processing a frame of 32ms of speech validates real-time operation of the implemented system.

• 한국정보처리학회:학술대회논문집
• /
• 한국정보처리학회 2016년도 춘계학술발표대회
• /
• pp.612-615
• /
• 2016

본 논문에서는 EPIC(Electric Potential Integrated Circuit) 센서를 통해 추출된 동작신호에 대해 이산 웨이블릿 변환（Discrete Wavelet Transform : DWT)과 선형 판별분석（Linear Discriminant Analysis : LDA), Support Vector Machine(SVM)을 사용하는 동작 분류 시스템을 제안한다. EPIC 센서 신호에 대해 이산 웨이블릿 변환을 사용하여 웨이블릿 계수인 근사계수(approximation coefficients)와 상세계수(detail coefficients)를 구한 후, 각각의 웨이블릿 계수에 대해 특징 파라미터를 추출한다. 이 때, 특징 파라미터는 14개의 통계적 특징 추출 파라미터 중에 유전자 알고리즘(Genetic Algorithm : GA)을 통하여 선택한 우수한 특징 파라미터이다. 웨이블릿 계수들에서 추출한 특징 파라미터는 선형 판별분석을 적용하여 차원을 축소하고 SVM의 훈련 및 분류에 사용한다. 실험결과, 4가지 동작에 대한 EPIC 센서 신호분류에서 제안된 방법의 분류율이 99.75%로 원신호에 대한 HMM 분류율 97% 보다 높은 정확률을 보여주었다.

• 대한인간공학회지
• /
• 제18권1호
• /
• pp.13-24
• /
• 1999

The objective of the study is to discriminate EEG(electroencephalogram) due to emotional changes. Emotion was evoked by the series of auditory stimuli which were selected from the natural sounds in the sound effect collection of compact disc. Seventeen university students participated and experienced positive or negative emotions by six auditory stimuli with intermission between stimuli. Temporal EEG ($T_3$, $T_4$, $T_5$, and $T_6$) was recorded at the same time and a subjective test was performed on the eleven point scales after the experiment. The maximum and minimum scores of the EEG among six stimuli EEG were analyzed for discrimination of emotion. The EEG signals were transformed into feature objects based on scalar intervention model coefficients. Auditory stimulus was considered as intervention variable. They were classified by Discriminant Analysis for each channel. The features showed results with the best classification accuracy of 91.2 % in $T_4$ for auditory stimuli. This study could be extended to establish an algorithm which quantifies and classifies emotions evoked by auditory stimulus using time-series models.

• 한국지능정보시스템학회:학술대회논문집
• /
• 한국지능정보시스템학회 2005년도 공동추계학술대회
• /
• pp.516-525
• /
• 2005

Case-based reasoning (CBR) has been widely used in various areas due to its convenience and strength in complex problem solving. Generally, in order to obtain successful results from CBR, effective retrieval of useful prior cases for the given problem is essential. However, designing a good matching and retrieval mechanism for CBR systems is still a controversial research issue. Most prior studies have tried to optimize the weights of the features or selection process of appropriate instances. But, these approaches have been performed independently until now. Simultaneous optimization of these components may lead to better performance than in naive models. In particular, there have been few attempts to simultaneously optimize the weight of the features and selection of the instances for CBR. Here we suggest a simultaneous optimization model of these components using a genetic algorithm (GA). We apply it to a customer classification model which utilizes demographic characteristics of customers as inputs to predict their buying behavior for a specific product. Experimental results show that simultaneously optimized CBR may improve the classification accuracy and outperform various optimized models of CBR as well as other classification models including logistic regression, multiple discriminant analysis, artificial neural networks and support vector machines.