• International journal of advanced smart convergence
• /
• v.9 no.2
• /
• pp.20-27
• /
• 2020

Sound event detection in real-world environments suffers from the interference of non-stationary and time-varying noise. This paper presents an adaptive noise reduction method for sound event detection based on non-negative matrix factorization (NMF). In this paper, we proposed a deep learning model that integrates Convolution Neural Network (CNN) with Non-Negative Matrix Factorization (NMF). To improve the separation quality of the NMF, it includes noise update technique that learns and adapts the characteristics of the current noise in real time. The noise update technique analyzes the sparsity and activity of the noise bias at the present time and decides the update training based on the noise candidate group obtained every frame in the previous noise reduction stage. Noise bias ranks selected as candidates for update training are updated in real time with discrimination NMF training. This NMF was applied to CNN and Hidden Markov Model(HMM) to achieve improvement for performance of sound event detection. Since CNN has a more obvious performance improvement effect, it can be widely used in sound source based CNN algorithm.

• Journal of Korea Multimedia Society
• /
• v.12 no.1
• /
• pp.26-30
• /
• 2009

We propose an adaptive Recursive Least Rank(RLR) L-filter which uses an L-estimator in order statistics and is based on rank estimate in robust statistics. The proposed RLR L-filter is a non-linear adaptive filter using non-linear adaptive algorithm and adapts itself to optimal filter in the sense of least dispersion measure of errors with non-homogeneous step size. Therefore the filter may be suitable for applications when the transmission channel is nonlinear channels such as Gaussian noise or impulsive noise, or when the signal is non-stationary such as image signal.

• Journal of Communications and Networks
• /
• v.9 no.1
• /
• pp.18-27
• /
• 2007

The conventional normalized least mean square (NLMS) algorithm is the most widely used for adaptive identification within a non-stationary input context. The convergence of the NLMS algorithm is independent of environmental changes. However, its steady state performance is impaired during input sequences with low dynamics. In this paper, we propose a new NLMS algorithm which is, in the steady state, insensitive to the time variations of the input dynamics. The square soot (SR)-NLMS algorithm is based on a normalization of the LMS adaptive filter input by the Euclidean norm of the tap-input. The tap-input power of the SR-NLMS adaptive filter is then equal to one even during sequences with low dynamics. Therefore, the amplification of the observation noise power by the tap-input power is cancelled in the misadjustment time evolution. The harmful effect of the low dynamics input sequences, on the steady state performance of the LMS adaptive filter are then reduced. In addition, the square root normalized input is more stationary than the base input. Therefore, the robustness of LMS adaptive filter with respect to the input non stationarity is enhanced. A performance analysis of the first- and the second-order statistic behavior of the proposed SR-NLMS adaptive filter is carried out. In particular, an analytical expression of the step size ensuring stability and mean convergence is derived. In addition, the results of an experimental study demonstrating the good performance of the SR-NLMS algorithm are given. A comparison of these results with those obtained from a standard NLMS algorithm, is performed. It is shown that, within a non-stationary input context, the SR-NLMS algorithm exhibits better performance than the NLMS algorithm.

• Proceedings of the Korea Society for Industrial Systems Conference
• /
• 2006.05a
• /
• pp.111-114
• /
• 2006

비백색잡음(non-white noise)인 잔향(reverberation)이 신호탐지(signal detection)의 주 방해신호인 천해 능동소나(active sonar) 환경에서의 표적탐지는 선백색화기(pre-whitening filter)를 사용하여 수신신호를 백색화한 후 백색잡음에서 최적 탐지기(optimum detector)인 정합필터를 사용한다. 그러나 이 방법은 잔향이 비정상(non-stationary) 특성을 가지기 때문에 구현이 매우 힘들다. 기존의 연구에 따르면 이러한 잔향은 지역적 정상상태(local stationary)라고 가정할 수 있다. 본 논문에서는 먼저 잔향신호의 지역적 정상상태의 범위를 추정(estimation)하고, 이 추정을 바탕으로 천해와 같은 비백색 잔향신호 환경에서 선백색화 블럭 정규화 정합필터(pre-whitening block normalized matched filter)의 성능을 개선할 수 있는 선백색화 기법을 제안하였다. 제안된 잔향신호의 백색화 기법은 표적신호 전 후의 잔향신호를 사용하여 처리블록(processing block)을 백색화하기 때문에 기존의 백색화 기법보다 우수한 성능을 보였다. 제안된 백색화 기법을 이용한 탐지기의 성능을 평가하기 위해 우리나라 인근해역에서 실측된 데이터를 이용하여 컴퓨터 모의실험을 수행하였다. 모의실험 결과 제안된 기법을 사용한 탐지기는 기존의 백색화 기법을 사용한 탐지기보다 우수한 탐지성능을 보였다.

• Smart Structures and Systems
• /
• v.14 no.1
• /
• pp.17-37
• /
• 2014

In this study, four widely accepted and used variants of Evolution Strategies (ES) are adapted and applied to the output-error state-space identification problem. The selection of ES is justified by prior strong indication of superior performance to similar problems, over alternatives like Genetic Algorithms (GA) or Evolutionary Programming (EP). The ES variants that are being tested are (i) the (1+1)-ES, (ii) the $({\mu}/{\rho}+{\lambda})-{\sigma}$-SA-ES, (iii) the $({\mu}_I,{\lambda})-{\sigma}$-SA-ES, and (iv) the (${\mu}_w,{\lambda}$)-CMA-ES. The study is based on a six-degree-of-freedom (DOF) structural model of a shear building that is characterized by light damping (up to 5%). The envisaged analysis is taking place through Monte Carlo experiments under two different excitation types (stationary / non-stationary) and the applied ES are assessed in terms of (i) accurate modal parameters extraction, (ii) statistical consistency, (iii) performance under noise-corrupted data, and (iv) performance under non-stationary data. The results of this suggest that ES are indeed competitive alternatives in the non-linear state-space estimation problem and deserve further attention.

• Proceedings of the Korean Society for Noise and Vibration Engineering Conference
• /
• 2005.11b
• /
• pp.144-147
• /
• 2005

The tire is considered as one of the Important noise sources having an influence on vehicle's performance. The Pattern noise of a tire is the transmission sound of airborne noise. On smooth asphalt road, Pattern noise is amplified with the velocity. In recent, the study on the reduction of Pattern noise is energetically processed. Pattern noise is strongly related with pitch sequence. To reduce the pattern noise, tire's designer has to randomize the sequence of pitch. The FFT is a traditional method to evaluate the level of the randomization of the pitch sequence, but gives no information on time-varying, instantaneous frequency. In the study, we found that Time-Frequency transform is a useful method to non-stationary signal such as tire noise.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.16 no.6 s.111
• /
• pp.627-633
• /
• 2006

The tire is considered as one of the important noise sources having an influence on vehicle's performance. The Pattern noise of a tire is the transmission sound of airborne noise. On smooth asphalt road, Pattern noise is amplified with the velocity. In recent, the study on the reduction of Pattern noise is energetically processed. Pattern noise is strongly related with pitch sequence. To reduce the pattern noise, tire's designer has to randomize the sequence of pitch. The FFT is a traditional method to evaluate the level of the randomization of the pitch sequence, but gives no information on time-varying, instantaneous frequency. In the study, we found that Time-Frequency transform is a useful method to non-stationary signal such as tire noise.

• Journal of Korea Multimedia Society
• /
• v.10 no.8
• /
• pp.1004-1016
• /
• 2007

In this paper we proposed a Weight-Update-Control Adaptive Noise Canceller which improves speech when environmental noise is stationary and it is hard to acquire a reference signal. Adaptive Noise Canceller(ANC) needs a reference signal, but it is not easy to measure pure noise without voice for reference in factory. Because there are mixed various mechanical noise and workers' voice. Therefore ANC is not suitable to reduce background noise. So we proposed the method that uses an arbitrary constant as an input signal and inputs microphone signal to the reference signal. The noise is eliminated using updated weights in non-speech range. In speech range the weight is fixed and the modified voice is acquired then voice is restored through transversal filter. The proposed method is based on facts that the factory noise is stationary and the noise is not changed in short conversation range. As a result of simulation using MATLAB, we confirmed that the proposed method is effective for reducing factory noise and has high signal to noise ratio(SNR).

• Proceedings of the Korean Information Science Society Conference
• /
• 1999.10b
• /
• pp.542-544
• /
• 1999

본 논문에서는 다수의 화자 모델을 구성함으로써 잡음에 강인한 화자확인 방법을 제안한다. Non-stationary한 잡음을 가진 입력음성의 SNR을 측정하는 것은 어렵기 때문에, 각 화자에 대해 잡음이 없을 때의 화자모델에 여러 SNR에 대한 잡음 모델을 결합시킴으로써 여러 개의 잡음 첨가된 화자 모델을 구성한다. 그리고, 화자확인에서는 이렇게 구한 각 모델에 대한 입력 음성의 likelihood를 구해 그 중 가장 큰 likelihood만을 선택한다. 이 값을 이용하여 화자확인을 수행한다. 실험 결과, 제안한 방법은 입력음성의 SNR을 모르는 잡음환경에서 일반적으로 하나의 모델을 사용하는 것보다 훨씬 좋은 성능을 보였다.

• Transactions of the Korean Society for Noise and Vibration Engineering
• /
• v.22 no.12
• /
• pp.1172-1183
• /
• 2012

Aure's tonality was considered as the sound metrics for the expression of the tonality of gear whine sound in a previous research. It was failed to use the Aure's tonality as a sound metric for the tonal impression. Thus Aures's tonality, was developed for tonal impression in previous research. However, this metric did not express well the tonality of gear whine sound since the whine sound is a non-stationary signal with frequency modulation and amplitude modulation. In this study, the new method for the tonality evaluation for a non-stationary signal is presented. It is developed based on the prominence ratio, tonality impression function, and lower threshold level. It improves the accuracy and reliability of the sound quality index being used for the sound quality evaluation of the axle-gear whine sound.