• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 1997.11a
• /
• pp.87-90
• /
• 1997

본 논문에서는 영상의 잡음(noise) 제거를 위한 새로운 diffusion모델을 제안한다. MOD모델 (mean curvature diffusion model)은 영상의 잡음 제거 때 유발되는 경계선의 blurring을 지양할 수 있는 장점이 있는 반면에 수렴상태(convergence state)를 갖지 못한 단점을 안고 있다. 본 논문에서는 MCD 모델에 min/max switch를 결합시킴으로써 MCD모델이 갖고 있던 문제점을 개선하였다. 제안하는 diffusion 모델은 scheme의 반복적인 적용에 대해서 실질적으로 그 결과가 더 이상 변동하지 않는 수렴상태(convergence state)를 가진 매우 안정적인 시스템이다.

• The Journal of the Acoustical Society of Korea
• /
• v.40 no.5
• /
• pp.479-487
• /
• 2021

In this paper, we propose an Sound Event Detection (SED) model using self-training based on a noisy student model. The proposed SED model consists of two stages. In the first stage, a mean-teacher model based on an Residual Convolutional Recurrent Neural Network (RCRNN) is constructed to provide target labels regarding weakly labeled or unlabeled data. In the second stage, a self-training-based noisy student model is constructed by applying different noise types. That is, feature noises, such as time-frequency shift, mixup, SpecAugment, and dropout-based model noise are used here. In addition, a semi-supervised loss function is applied to train the noisy student model, which acts as label noise injection. The performance of the proposed SED model is evaluated on the validation set of the Detection and Classification of Acoustic Scenes and Events (DCASE) 2020 Challenge Task 4. The experiments show that the single model and ensemble model of the proposed SED based on the noisy student model improve F1-score by 4.6 % and 3.4 % compared to the top-ranked model in DCASE 2020 challenge Task 4, respectively.

• Proceedings of the KIEE Conference
• /
• 2003.07d
• /
• pp.2696-2698
• /
• 2003

전력선 통신 채널(powerline communication channel)은 원래 전력을 전송하기 위한 선로로서 무선 채널 및 그 밖의 다른 통신 채널들과 같이 부가적 가우시안 잡음 환경 등의 표준 모델로 잘 표현되지 않는 특성이 있다. 그리고 1MHz-30MHz 대역에서의 전력선 채널의 잡음은 배경잡음, 연속 신호(continuous wave) 협대역 및 광대역 잡음, 그리고 임펄스 잡음으로 분류될 수 있으며, 특히 고속 데이터 전송 시스템에서는 임펄스 잡음에 의한 영향이 큰 것으로 알려져 있다. 본 논문에서는 아파트형 주거지에서 실측한 주파수 영역 및 시간영역의 잡음 데이터를 제시하고, 이러한 데이터를 바탕으로 연속 신호 잡음을 포함한 배경잡음과 임펄스 잡음의 모델 및 통계적 특성을 제시한다.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2001.11b
• /
• pp.129-133
• /
• 2001

영상에 포함된 잡음은 화질 및 영상의 압축효율을 저하시킨다. 최근 들어, 영상의 에지 성분을 효율적으로 고려하면서 잡음을 제거하기 위하여 다양한 비정적(nonstationary) 영상 모델에 근거한 잡음제거 알고리즘이 제안되어 왔다. 하지만, 기존의 비정적 영상모델에서는 연산량의 부담을 덜기 위하여 각 화소들 사이에 상관관계(correlation)가 없다는 가정을 하고 있어 영상의 미세한 정보들이 필터링에 의하여 훼손된다. 본 논문에서는 영상의 비정적 상관관계를 고려한 적응적 잡음제거 알고리즘을 제시한다. 영상신호는 비정적 평균을 가진다고 가정되며, 또한 각기 다른 정적(stationary) 상관관계를 가지는 부분 영상으로 분리된다고 가정된다. 제안된 영상 모델에서의 공분산(co-variance) 행렬의 특수한 구조를 이용하여 계산적으로 효율적인 FFT에 기반한 선형 minimum mean square error 필터를 유도한다. 제안된 영상 모델의 정당성 및 알고리즘의 효율성을 제시한다.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.47 no.6
• /
• pp.28-38
• /
• 2010

In order to remove the noise contained in recorded analog video, it is important to recognize the real characteristics and strength of the noise. This paper presents an efficient training-based noise reduction method for recorded analog video after analyzing the noise characteristics of analog video captured in a real broadcasting system. First we show that there is non-negligible noise correlation in recorded analog video and describe the limitations of the traditional noise estimation and reduction methods based on additive white Gaussian noise (AWGN) model. In addition, we show that auto-regressive (AR) model considering noise correlation can be successfully utilized to estimate and synthesize the noise contained in the recorded analog video, and the estimated AR parameters are utilized in the training-based noise reduction scheme to reduce the video noise. Experiment results show that the proposed method can be efficiently applied for noise reduction of recorded analog video with non-negligible noise correlation.

• Journal of IKEEE
• /
• v.3 no.2 s.5
• /
• pp.226-235
• /
• 1999

To investigate SPICE noise model and the behavior of its parameters, 1/f noise of NMOS devices fabricated by BiCMOS process is measured and compared to the various noise models and measured results. For the long channel devices, bias dependence of the drain current noise power spectral density $S_{Id}$ of NMOS is similar to the previous results. Equivalent gate noise power spectral density $S_{Vg}$ shows weak dependence on the gate and drain voltages in long channel NMOS as the previous results. However, it is shown that most of published noise models are difficult to apply to short channel devices. Therefore, in this study, with comparison of our experimental results, we have tried to find the model of 1/f noise, appropriate for our NMOS device fabricated by BiCMOS process.

• Journal of Broadcast Engineering
• /
• v.23 no.4
• /
• pp.511-518
• /
• 2018

This paper deals with improving speech intelligibility by applying binary mask to time-frequency units of speech in noise. The binary mask is set to "0" or "1" according to whether speech is dominant or noise is dominant by comparing signal-to-noise ratio with pre-defined threshold. Bayesian classifier trained with Gaussian mixture model is used to estimate the binary mask of each time-frequency signal. The binary mask based noise suppressor improves speech intelligibility only in noise condition which is included in the training data. In this paper, speaker adaptation techniques for speech recognition are applied to adapt the Gaussian mixture model to a new noise environment. Experiments with noise-corrupted speech are conducted to demonstrate the improvement of speech intelligibility by employing adaption techniques in a new noise environment.

• Proceedings of the Korean Society of Broadcast Engineers Conference
• /
• 2019.11a
• /
• pp.213-216
• /
• 2019

영상 잡음 제거 알고리즘은 잡음으로 오염된 영상으로부터 잡음이 제거된 깨끗한 영상을 추정하여 복원하는 연구이다. 기존의 모델 기반 방법의 영상 잡음 제거 알고리즘은 영상을 복원하는 과정에서 최적화 문제를 풀어야 한다는 단점과 매개변수를 직접 선택을 해주어야 한다는 단점을 가진다. 본 논문에서는 딥러닝을 이용한 학습기반 방법의 영상 잡음 제거 연구를 소개한다. 먼저, 신경망의 구축을 위하여 신경망의 구성 요소는 Instance Normalization 과 컨볼루션 신경망을 이용한 모델을 제안하였고, 여러 연구 분야에서 좋은 성능을 보이는 U-Net 구조를 전체적인 구조로 차용하였다. 신경망의 학습을 위하여 DnCNN 에서 제안한 잡음을 학습하는 잔여 학습 기법을 채택하였고, 기존의 영상 잡음 제거 알고리즘의 단점인 결과 영상이 흐릿해지는 현상을 보완하기 위하여 생성적 적대 신경망 학습 방법을 적용하였다. 본 논문에서 제안한 신경망을 이용한 잡음 제거 영상의 결과가 기존의 연구 방법들 보다 인지적인 측면에서 좋은 결과를 보임을 확인하였다.

• Journal of Broadcast Engineering
• /
• v.20 no.1
• /
• pp.164-170
• /
• 2015

This paper deals with the adaptation of classification model in the binary mask approach to suppress noise in the noisy environment. The binary mask estimation approach is known to improve speech intelligibility of noisy speech. However, the same type of noisy data for the test data should be included in the training data for building the classification model of binary mask estimation. The eigenvoice adaptation is applied to the noise-independent classification model and the adapted model is used as noise-dependent model. The results are reported in Hit rates and False alarm rates. The experimental results confirmed that the accuracy of classification is improved as the number of adaptation sentences increases.

• The Journal of the Acoustical Society of Korea
• /
• v.23 no.6
• /
• pp.473-480
• /
• 2004

In this paper we propose an effective feature compensation scheme based on the speech model in order to achieve robust speech recognition. The proposed feature compensation method is based on parallel combined mixture model (PCMM). The previous PCMM works require a highly sophisticated procedure for estimation of the combined mixture model in order to reflect the time-varying noisy conditions at every utterance. The proposed schemes can cope with the time-varying background noise by employing the interpolation method of the multiple mixture models. We apply the‘data-driven’method to PCMM tot move reliable model combination and introduce a frame-synched version for estimation of environments posteriori. In order to reduce the computational complexity due to multiple models, we propose a technique for mixture sharing. The statistically similar Gaussian components are selected and the smoothed versions are generated for sharing. The performance is examined over Aurora 2.0 and speech corpus recorded while car-driving. The experimental results indicate that the proposed schemes are effective in realizing robust speech recognition and reducing the computational complexities under both simulated environments and real-life conditions.