• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.89-92
• /
• 1999

인식기 훈련과정에서 발생하지 않았던 잡음이 인식과정에서 신호를 손상할 경우 인식률의 저하가 발생한다. 본 논문에서는 음성의 질을 떨어뜨리는 이러한 잡음을 Wavelet Packets을 이용하여 전처리함으로서 인식률을 향상시키는 방법을 제안한다. 인식기로는 Hidden Markov Model을 사용하였고, 시스템에 사용된 특징 파라미터로는 15차 Cepstrum을 사용하였다. 11 kHz로 샘플링된 숫자음에 Additive White Gaussian Noise를 첨가한 손상된 음성신호를 인식실험에 사용하였다. 화자독립으로 진행된 실험에서 잡음에 의해 손상된 SNR 20dB의 음성신호에 대하여 Wavelet Packets로 잡음을 제거한 후 복원된 음성신호 의 인식률은 약 $10\%$ 향상됨을 확인하였다.

• Smart Structures and Systems
• /
• v.1 no.2
• /
• pp.185-215
• /
• 2005

Advanced signal processing techniques have been long introduced and widely used in structural health monitoring (SHM) and nondestructive evaluation (NDE). In our research, we applied several signal processing approaches for our embedded ultrasonic structural radar (EUSR) system to obtain improved damage detection results. The EUSR algorithm was developed to detect defects within a large area of a thin-plate specimen using a piezoelectric wafer active sensor (PWAS) array. In the EUSR, the discrete wavelet transform (DWT) was first applied for signal de-noising. Secondly, after constructing the EUSR data, the short-time Fourier transform (STFT) and continuous wavelet transform (CWT) were used for the time-frequency analysis. Then the results were compared thereafter. We eventually chose continuous wavelet transform to filter out from the original signal the component with the excitation signal's frequency. Third, cross correlation method and Hilbert transform were applied to A-scan signals to extract the time of flight (TOF) of the wave packets from the crack. Finally, the Hilbert transform was again applied to the EUSR data to extract the envelopes for final inspection result visualization. The EUSR system was implemented in LabVIEW. Several laboratory experiments have been conducted and have verified that, with the advanced signal processing approaches, the EUSR has enhanced damage detection ability.

• Journal of Information Processing Systems
• /
• v.13 no.5
• /
• pp.1135-1148
• /
• 2017

Speech recognition is one of the fascinating fields in the area of Computer science. Accuracy of speech recognition system may reduce due to the presence of noise present in speech signal. Therefore noise removal is an essential step in Automatic Speech Recognition (ASR) system and this paper proposes a new technique called combined thresholding for noise removal. Feature extraction is process of converting acoustic signal into most valuable set of parameters. This paper also concentrates on improving Mel Frequency Cepstral Coefficients (MFCC) features by introducing Discrete Wavelet Packet Transform (DWPT) in the place of Discrete Fourier Transformation (DFT) block to provide an efficient signal analysis. The feature vector is varied in size, for choosing the correct length of feature vector Self Organizing Map (SOM) is used. As a single classifier does not provide enough accuracy, so this research proposes an Ensemble Support Vector Machine (ESVM) classifier where the fixed length feature vector from SOM is given as input, termed as ESVM_SOM. The experimental results showed that the proposed methods provide better results than the existing methods.

• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.425-428
• /
• 2002

본 논문에서는 해양에서 다중경로를 통하여 수신되는 음파의 도달시간을 정확히 파악하기 위한 방법을 제시하였다. 음파 도달 시간을 파악하기 위한 방법으로는 정합 필터 방법 및 웨이브렛 방법을 도입하였으며 각각 모의 수신신호 및 실관측 수신신호에 대해 적용하여 식별 성능을 분석하였다.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.35S no.6
• /
• pp.72-80
• /
• 1998

A method is proposed for directional vector quantization using the wavelet packets. After partitioning the wavelet packet coefficients into 9 edges according to the corresponding directions, it encodes and transmits locally dominant edges. The directions of the edges are encoded with a variable length coding and conditional switching of codebooks, while the contents of them with the vector quantization followed by the variable length index coding. The proposed algorithm is superior to various conventional image coding algorithms in the sense of PSNR, which is relatively more significant at very low bit rate such as 0.1~0.3 bpp. As the proposed algorithm preserves the edges which is the most important for the human visual system, it also provides the reconstructed images of good subjective quality even at very low bit rate.

• Proceedings of the KIEE Conference
• /
• 2006.07d
• /
• pp.1807-1808
• /
• 2006

In this paper, we present the wavelet neural network (WNN) controller as an active queue management(AQM) in end-to-end TCP network. AQM is important to regulate the queue length and short round trip time by passing or dropping the packets at the intermediate routers. As the role of AQM, the WNN controller adaptively controls the dropping probability of the TCP network and is trained by gradient-descent algorithm. We illustrate our result that WNN controller is superior to PI controller via simulations.

• Journal of the Korean Institute of Intelligent Systems
• /
• v.25 no.1
• /
• pp.97-104
• /
• 2015

An 'associative cube', a class of auto-associative memories, is revisited here, in which training data and hidden orthogonal basis functions such as wavelet packets or Fourier kernels, are combined in the weight cube. This weight cube has hidden units in its depth, represented by a three dimensional cubic structure. We develop an unsupervised incremental learning mechanism based upon the adaptive least squares method. Training data are mapped into orthogonal basis vectors in a least-squares sense by updating the weights which minimize an energy function. Therefore, a prescribed orthogonal kernel is incrementally assigned to an incoming data. Next, we show how a decoding procedure finds the closest one with a competitive network in the hidden layer. As noisy test data are applied to an associative cube, the nearest one among the original training data are restored in an optimal sense. The simulation results confirm robustness of associative cubes even if test data are heavily distorted by various types of noise.

• Smart Structures and Systems
• /
• v.22 no.3
• /
• pp.315-326
• /
• 2018

This paper presents a novel time-domain method for the identification of plastic rotations and stiffness parameters of single-bay frames with nonlinear plastic hinges. Each plastic hinge is modelled as a pseudo-semi-rigid connection with nonlinear hysteretic moment-curvature characteristics at an element end. Through the comparison of the identified end rotations of members that are connected together, the plastic rotation that furnishes information of the locations and plasticity degrees of plastic hinges can be identified. The force consideration of the frame members may be used to relate the stiffness parameters to the elastic rotations and the excitation. The damped-least-squares method and damped-and-weighted-least-squares method are adopted to estimate the stiffness parameters of frames. A noise-removal strategy employing a de-noising technique based on wavelet packets with a smoothing process is used to filter out the noise for the parameter estimation. The numerical examples show that the proposed method can identify the plastic rotations and the stiffness parameters using measurements with reasonable level of noise. The unknown excitation can also be estimated with acceptable accuracy. The advantages and disadvantages of both parameter estimation methods are discussed.

• The Bulletin of The Korean Astronomical Society
• /
• v.43 no.2
• /
• pp.45.3-46
• /
• 2018

Oscillations in a sunspot are easily detected through the Doppler velocity observation. Although the sunspot oscillations look erratic, the wavelet analysis show that they consist of successive wave packets which have strong power near three or five minutes. Previous studies found that 3-min oscillation at the chromosphere is a visual pattern of upward propagating acoustic waves along the magnetic field lines. Resent multi-height observations help this like vertical study, however, we also focus on horizontal facet to extend three dimensional understand of sunspot waves. So, we investigate a fragmented sunspot expected to have complex wave profiles according to the positions in the sunspot observed by the Fast Imaging Solar Spectrograph. We choose 4 points at different umbral cores as sampling positions to determine coherence of oscillations. The sets of cross-correlation with three and five minutes bandpass filters during a single wave packet reveal interesting results. Na I line show weak correlations with some lags, but Fe I and Ni I have strong correlations with no phase difference over the sunspots. It is more remarkable at Ni I line with 3-min bandpass that all sets of cross-correlation look like the autocorrelation. We can interpret this as sunspot oscillations occur spontaneously over a sunspot at photosphere but not at chromosphere. It implies a larger or deeper origin of 3-min sunspot oscillation.