• The Journal of the Acoustical Society of Korea
• /
• v.41 no.6
• /
• pp.647-654
• /
• 2022

Although the fields in which is utilized children's speech recognition is on the rise, the lack of quality data is an obstacle to improving children's speech recognition performance. This paper proposes a new method for improving children's speech recognition performance by additionally using adult speech data. The proposed method is a transformer based domain adversarial training using dynamically weighted loss to effectively address the data imbalance gap between age that grows as the amount of adult training data increases. Specifically, the degree of class imbalance in the mini-batch during training was quantified, and the loss function was defined and used so that the smaller the data, the greater the weight. Experiments validate the utility of proposed domain adversarial training following asymmetry between adults and children training data. Experiments show that the proposed method has higher children's speech recognition performance than traditional domain adversarial training method under all conditions in which asymmetry between age occurs in the training data.

• Journal of IKEEE
• /
• v.13 no.2
• /
• pp.241-247
• /
• 2009

This paper suggests the expansion methods of semantic dictionary, taking Korean semantic features account. These methods will be used to extract a practical topic word in the automatic summarization. The first is the method which is constructed the synonym dictionary for improving the performance of semantic-marker analysis. The second is the method which is extracted the probabilistic information from the subcategorization dictionary for resolving the syntactic and semantic ambiguity. The third is the method which is predicted the subcategorization patterns of the unregistered predicate, for the resolution of an affix-derived predicate.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 2003.09b
• /
• pp.71-75
• /
• 2003

오류 역전파 학습 알고리즘을 이용하여 영상 인식에 적용 할 경우에는 은닉층의 노드 수를 경험적으로 설정하므로, 학습시간과 지역최소화 및 정체현상이 발생한다. 그리고 ARTI 알고리즘은 입력 패턴과 저장 패턴간의 측정 방법인 유사성 검증 방법과 경계 변수의 설정에 따라 인식률이 좌우된다. 경계 변수의 값이 크면 입력 패턴과 저장 패턴사이에 약간의 차이만 있어도 새로운 카테고리(Category)로 분류하고, 반대로 경계 변수의 값이 적으면 입력 패턴과 저장 패턴 사이에 많은 차이가 있더라도 유사성이 인정되어 입력 패턴들을 대략적으로 분류한다. 따라서 ART1 알고리즘을 영상 인식에 적용하기 위해서는 경계 변수를 경험적으로 설정하므로 인식률에 부정적인 영향을 갖는 문제점이 있다. 따라서 본 논문에서는 개선된 ART1 알고리즘과 지도 학습 방법을 결합하여 신경망의 은닉층 노드를 동적으로 변화시키는 자가 생성지도 학습 알고리즘을 제안한다. 제안된 신경망에서 입력층과 은닉층의 학습 구조에는 ART1 알고리즘을 개선하여 적용하고, 은닉층과 출력층의 학습 구조에는 은닉층에서 승자로 선택된 노드와 출력층 노드와 연결된 가중치만을 조정하고 Delta-Bar-Delta 알고리즘을 적용한다. 제안된 방법의 학습 성능을 분석하기 위하여 학생증 영상에서 추출한 학번 패턴 분류에 적용한 결과, 기존의 신경망 학습 알고리즘보다 학습 성능이 개선됨을 확인하였다.

• Journal of the Institute of Electronics Engineers of Korea SD
• /
• v.38 no.2
• /
• pp.137-144
• /
• 2001

In this paper, a state assignment algorithm was proposed to reduce power consumption in control-flow oriented finite state machines. The Markov chain model is used to reduce the switching activities, which closely relate with dynamic power dissipation in VLSI circuits. Based on the Markov probabilistic description model of finite state machines, the hamming distance between the codes of neighbor states was minimized. To express the switching activities, the cost function, which also accounts for the structure of a machine, is used. The proposed state assignment algorithm is tested with Logic Synthesis Benchmarks, and reduced the cost up to 57.42% compared to the Lakshmikant's algorithm.

• Proceedings of the Korean Society of Precision Engineering Conference
• /
• 1997.10a
• /
• pp.23-27
• /
• 1997

This paper introduces the structural design optimization of a high speed machining center using multi-step optimization combined with G.A.(Genetic Algorithm) and Weighted Method. In this case, the design problem is to find out the best design variables which minimize the static compliance, the dynamic compliance, and the weight of the machine structure simultaneously. Dimensional thicknesses of the thirteen structural members of the machine structure are adopted as design variables. The first step is the cross-section configuration optimization, in which the area moment of inertia of the cross-section for each structural member is maximized while its area is kept constant The second step is a static design optimization, In which the static compliance and the weight of the machine structure are minimized under some dimensional and safety constraints. The third step IS a dynamic design optimization, where the dynamic compliance and the structure weight are minimized under the same constraints. After optunization, static and dynamic compliances were reduced to 62.3% and 95.7% Eorn the initial design, while the weight of the moving bodies are also in the feaslble range.

• Proceedings of the Korean Society of Computer Information Conference
• /
• 2011.01a
• /
• pp.311-313
• /
• 2011

대부문의 이진화 알고리즘은 임계치를 결정하기 위하여 히스토그램을 사용하여 밝기분포를 분석한다. 배경과 물체의 명도차이가 큰 경우에는 분할을 위해 양봉(bimadal) 히스토그램으로 표현하여 최적의 임계치를 찾기 위해 히스토그램 골짜기(valley)를 선택하는 것만으로도 양호한 임계치 결과를 얻을 수 있다. 하지만 배경과 물체의 밝기 차이가 크지 않거나 밝기 분포가 양봉 특성이 보이지 않을 때는 히스토그램 분석만으로 적절한 임계치를 얻기 어렵다. 그리고 한 영상에서는 넓은 영역에 걸쳐 명암도 변화가 일어나고 다양한 유형의 물체가 있을 때 스케치 특징점의 유무를 판별하는 임계치의 결정에는 애매모호함이 존재한다. 따라서, 본 논문에서는 영상에 대한 삼각형 타입의 소속함수를 적용하여 임계치를 동적으로 설정하고 영상을 이진화하는 알고리즘을 제안한다. 제안된 퍼지 이진화 알고리즘은 원 영상을 특정 크기의 윈도우로 나누어서 윈도우의 소속 함수에 대한 소속도를 구하여 영상을 이진화한다. 다양한 영상에 적용한 결과, 기존의 이진화 기법보다 제안된 퍼지 이진화 알고리즘이 효율적인 것을 알 수 있었다.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2001.06a
• /
• pp.141-144
• /
• 2001

In this paper, we proposed a neural network controller with a fuzzy compensator which compensate a output of neural network controller. Even if learn by neural network controller, it can occur a bad results from disturbance or load variations. So in order to adjust above case. we used the fuzzy compensator to get an expected results. And the weight of main neural network can be changed with the result of learning an inverse model neural network of plant, so a expected dynamic characteristics of plant can be got. In order to confirm a performance of the proposed controller, we implemented the controller using the DSP processor and applied in a hydraulic servo system. And then we observed an experimental results.

• Journal of the Institute of Convergence Signal Processing
• /
• v.13 no.4
• /
• pp.233-237
• /
• 2012

In this paper, We propose a control system which compensate a output of a main Neual Network using a RTDNN(Recurrent Time Delayed Neural Network) with a FLC(Fuzzy Logic Controller)After a learn of main neural network, it can occur a Over shoot or Under shoot from a disturbance or a load variations. In order to adjust above case, we used the fuzzy compensator to get an expected results. And the weight of main neural network can be changed with the result of learning a inverse model neural network of plant, so a expected dynamic characteristics of plant can be got. We can confirm good response characteristics of proposed neural network controller by the results of simulation.

• The Journal of the Korea institute of electronic communication sciences
• /
• v.4 no.1
• /
• pp.66-69
• /
• 2009

An adaptive echo canceller is necessary for an application such as a speakerphone, 3G image telephony and VoIP service system. These echo cancellers need to have many taps for filtering echo signals. Many taps cause computation data to increase and convergence speed to be low. To overcome these problems, An adaptive echo canceller with the advanced convergence speed is proposed in this paper. To improve the speed, we divide an echo band into subbands and place a subband filter to be adaptive for each subband. Each subband filter recognizes the echo signal as subband echo signals. So, dynamic range of subband is small, the convergence speed is fast. Moreover, as the number of Tap and weight update are estimated in each subband, the implementation complex of a adaptive filter is low.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.36 no.4B
• /
• pp.305-313
• /
• 2011

This paper presents a P2P (Peer-to-Peer) overlay multicast tree construction algorithm to support stable multimedia service over the Internet. While constructing a multicast tree, it takes into account not only the link delay, but also peer stability. Since peers actually show dynamic and unstable behavior over P2P-based network, it is essential to consider peer stability. Furthermore, the weighting factor between link delay and peer stability is adaptively controlled according to the characteristics of the multicast tree. Basically, Genetic algorithm is employed to obtain a near optimal solution with low computational complexity. Finally, simulation results are provided to show the performance of the proposed algorithm.