Keunsung Bae
Kyungpook National University
Network
Latest external collaboration on country level. Dive into details by clicking on the dots.
Publication
Featured researches published by Keunsung Bae.
international workshop on digital watermarking | 2003
Siho Kim; Hong-Seok Kwon; Keunsung Bae
In this paper, we present a new echo kernel, which is a modification of polar echo kernel, to improve the detection performance and robustness against attacks. Polar echo kernel may have the advantage of large detection margin from the polarity of inserted echo signal, but its poor frequency response in low frequency band degrades sound quality. To solve this problem, we applied bipolar echo pulses to the polar echo kernel. Using the proposed echo kernel, the distributions of autocepstrum peaks for data ‘0’ and ‘1’ are located more distant and improvement of detection performance is achieved. It also makes the low frequency band flat so that the timbre difference in the polar echo kernel can be removed to reproduce the imperceptible sound quality. Informal listening tests as well as robustness test against attacks were performed to evaluate the proposed echo kernel. Experimental results demonstrated the superiority of the proposed echo kernel to both conventional unipolar and polar echo kernels.
international workshop on digital watermarking | 2004
Siho Kim; Keunsung Bae
Recently, informed watermarking schemes based on Costas dirty paper coding are drawing more attention than spread spectrum based techniques because these kinds of watermarking algorithms do not need an original host signal for watermark detection and the host signal does not affect the performance of watermark detection. For practical implementation, they mostly use uniform scalar quantizers, which are very vulnerable against amplitude modification. Hence, it is necessary to estimate the amplitude modification, i.e., a modified quantization step size, before watermark detection. In this paper, we propose a robust algorithm to estimate the modified quantization step size with an optimal search interval. It searches the quantization step size to minimize the quantization error of the received audio signal. It does not encroach the space for embedding watermark message because it just uses the received signal itself for estimation of the quantization step size. The optimal searching interval is determined to satisfy both detection performance and computational complexity. Experimental results show that the proposed algorithm can estimate the modified quantization step size accurately under amplitude modification attacks.
information sciences, signal processing and their applications | 2007
Tae-Gyun Lim; Keunsung Bae; Chan-Sik Hwang; Hyeonguk Lee
This paper presents a new method for classification of underwater transient signals, which employs frame-based decision with mel-frequency cepstral coefficients (MFCC). The MFCC feature vector is extracted frame-by-frame basis for an input signal that is detected as a transient signal, and Euclidean distances are calculated between this and all MFCC feature vectors in the reference database. Then each frame of the detected input signal is mapped to the class having minimum Euclidean distance in the reference database. Finally the input signal is classified as the class that has maximum mapping rate in the reference database. Experimental results demonstrate that the proposed method is very promising for classification of underwater transient signals.
IEICE Transactions on Fundamentals of Electronics, Communications and Computer Sciences | 2008
Tae-Gyun Lim; Keunsung Bae; Chan-Sik Hwang; Hyeonguk Lee
This paper presents a new method for classification of underwater transient signals, which employs a binary image pattern of the mel-frequency cepstral coefficients as a feature vector and a feed-forward neural network as a classifier. The feature vector is obtained by taking DCT and 1-bit quantization for the square matrix of the mel-frequency cepstral coefficients that is derived from the frame based cepstral analysis. The classifier is a feed-forward neural network having one hidden layer and one output layer, and a back propagation algorithm is used to update the weighting vector of each layer. Experimental results with underwater transient signals demonstrate that the proposed method is very promising for classification of underwater transient signals.
international conference on acoustics, speech, and signal processing | 2006
Siho Kim; Keunsung Bae
Scalar quantization-based watermarking schemes are very vulnerable to amplitude modification attack. To overcome this problem, we propose a novel and robust algorithm that estimates the modified quantization step size by searching QE (quantization error) function. For efficient searching of QE curve, we analyze the peak curve of QE analytically and derive the equations to determine an appropriate search interval. The search interval can be determined from the mean and variance of an audio signal regardless of its probability density function shape. The experimental results demonstrate that the proposed algorithm provides both exact estimation of the modified quantization step size and good watermark detection performance under AWGN attack
international conference on digital signal processing | 2002
Siho Kim; Keunsung Bae
This paper deals with the problem of double-talk (DT) detection in an acoustic echo canceller (AEC). In the DT detection algorithm with correlation coefficient, detection errors occasionally occur because it is hard to set the threshold to distinguish DT from echo path change (EPC). When EPC is erroneously considered as DT at the starting point of EPC, the adaptive filter can fall into the situation that it stops updating its filter coefficients. In addition, in case of echo path changing during the DT period, the end-point detection of DT period fails so that the adaptive filter cannot update its tap coefficients even after the DT period ends. To solve this problem, in this paper, we propose a new DT detection algorithm with an auxiliary filter. We use the idea that the error signal cannot be estimated using a reference signal in case of DT situation but it can be in case of EPC. The computer simulation verifies that the proposed method could solve the problem caused by the DT detection error or echo path change during the DT period.
signal processing systems | 2000
Hong-Seok Kwon; Siho Kim; Keunsung Bae
The goal of this research is the real-time implementation of an MPEG-1 Layer III audio decoder using the TMS320C6201 fixed-point digital signal processor (DSP). The main jobs for this work are twofold: one is to convert the floating-point operation in the decoder into a fixed-point operation while maintaining high resolution, and the other is to optimize the program to make it run in real time with a memory size that is as small as possible. The implemented decoder uses 6.77 kwords of program memory, 3.13 kwords of data ROM and 9.94 kwords of data RAM, respectively. It also uses about 26% of the maximum computation capacity of the TMS320C6201. Comparing the PCM (pulse code modulation) output of the fixed-point computation with that of the floating-point computation for test bit streams, we achieved a signal-to-noise ratio (SNR) of more than 60 dB.
information sciences, signal processing and their applications | 2007
Wonchul Heo; Taehwan Kim; Keunsung Bae
Double-talk detection errors can result in either large residual echo or distorting the near-end talkerpsilas input speech. Thus accurate double-talk detection is an important problem in the echo canceller to improve the speech quality. In the double-talk detection algorithm with the cross-correlation coefficient, detection errors can occur in the initial convergence period of an adaptive filter or in the noisy environment since the correlation becomes high in those situations. In this paper, we propose a new double-talk detector based on a correlation method that uses the normalized error signal power to reduce the double-talk detection errors. The experimental results show the performance improvement of the acoustic echo canceller as well as the robustness of the proposed double-talk detector.
international workshop on digital watermarking | 2006
Siho Kim; Keunsung Bae
The quantization-based watermarking schemes such as QIM or SCS are known to be very vulnerable to the amplitude modification attack. The amplitude modification attack results in the change of quantization step size so the estimation of a modified quantization step size is required before watermark detection. In this paper, we analyze the quantization error function of the audio signal having any shape of probability density function, and analytically determine the search interval that minimizes the quantization error considering both detection performance and computational complexity. It is shown that the appropriate search interval can be determined from the frame-based mean and variance of the input signal without regard to its shape of probability density function. Experimental results for real audio data verify that the derived search interval provides the accurate estimation of the modified quantization step size under amplitude modification attack.
australasian joint conference on artificial intelligence | 2004
Sungyun Jung; Jongmok Son; Keunsung Bae
We present a new feature extraction method for robust speech recognition in the presence of additive white Gaussian noise The proposed method is made up of two stages in cascade The first stage is denoising process based on the wavelet domain hidden Markov tree model, and the second one is reduction of the influence of the residual noise in the filter bank analysis To evaluate the performance of the proposed method, recognition experiments were carried out for noisy speech with signal-to-noise ratio from 25 dB to 0 dB Experiment results demonstrate the superiority of the proposed method to the conventional ones.