• Journal of KIISE:Computing Practices and Letters
• /
• v.6 no.4
• /
• pp.403-410
• /
• 2000

In this paper, the development of a stock information retrieval system using speech recognition and its features are described. The system is based on DHMM (discrete hidden Markov model) and PLUs (phonelike units) are used as the basic unit for recognition. End-point detection and echo cancellation are included to facilitate speech input. Continuous speech recognizer is implemented to allow multi-word speech. Data collected over several months are analyzed.

• Proceedings of the Acoustical Society of Korea Conference
• /
• autumn
• /
• pp.119-122
• /
• 1999

In stereo acoustic echo canceller scheme, adaptive filters coefficients converge very slowly or misconverge to real acoustic echo path in receiving room. This is due to cross-correlation between channel signals. In this paper, new preprocessor using absolute difference factor between stereo signals is proposed to reduce cross-correlation between signals. Computer simulations demonstrate that this preprocessor perform well to the half wave rectifier which has simple and good performance of the several preprocessors in recent papers. When the paths in transmitting room change, performance does not degrade in proposed preprocessor.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.56 no.3
• /
• pp.629-631
• /
• 2007

In hands-free full-duplex communication systems, acoustic signals picked up by the microphones can be mixed with echo signals as well as noises, which may result in poor performance of the corresponding communication system. Also, the system performance may decrease further if the reverberation occurs since it is harder to estimate the impulse response of the demixing system. For blind source separation (BSS) in such cases, a time-frequency masking approach can be employed to separate undesired echo signals and noises, but, permutation ambiguities also should be solved for the echo cancellation. In this paper, we propose a new acoustic echo cancellation (AEC) approach utilizing the time-frequency masking and higher-order statistics, whereby a desired signal selection, based on coherence and third-order statistics (i.e., kurtosis), is introduced along with output signal normalization. Simulation results demonstrate that the proposed approach yields better echo and noise cancellation performances than the conventional AEC approaches.

• Proceedings of the KIEE Conference
• /
• 2006.10c
• /
• pp.416-418
• /
• 2006

In this paper, a new acoustic echo cancellation approach based on the DUET algorithm and scaling factor estimation is proposed to solve the scaling ambiguity in case of blind separation based acoustic echo cancellation in a noisy environment. In hands-free full-duplex communication system. acoustic noises picked up by the microphone are mixed with echo signal. For this reason, the echo cancellation system may provide poor performance. For that purpose, a degenerate unmixing estimation technique, adjusted in the time-frequency domain, is employed to separate undesired echo signals and noises. Also, since scaling and permutation ambiguities have not been solved in the blind source separation algorithm, kurtosis for the desired signal selection and a scaling factor estimation algorithm are utilized in this rarer for the separation of an echo signal. Simulation results demonstrate that the proposed approach yields better echo cancellation and noise reduction performances, compared with conventional methods.

• Proceedings of the KIEE Conference
• /
• 2006.04a
• /
• pp.24-26
• /
• 2006

In this paper, a nonlinear echo cancellation approach, based on a Gauss-Seidel pseudo affine projection algorithm and Volterra filtering, is proposed to compensate for echo path nonlinearity in the telephone network. Simulation results demonstrate that the proposed approach yields reduction of computational complexity and improved convergence speed than the conventional nonlinear echo cancellation methods (NLMS, ECLMS, FAP, RLS).

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.56 no.2
• /
• pp.420-423
• /
• 2007

In this paper, an efficient nonlinear echo cancellation method is proposed, whereby the fast block exact Gauss-Seidel pseudo affine projection (FBEGS-PAP) is further utilized for adaptive Volterra filtering. In particular, the proposed nonlinear echo cancellation approach requires lower computational complexity as in the conventional linear adaptive echo cancellation methods based on NLMS and GS-PAP, and still provides nonlinear echo cancellation performance similar to the GS-PAP method. Finally, echo cancellation performance of the proposed approach is demonstrated by providing some simulation results.

• Proceedings of the Korean Institute of Communication Sciences Conference
• /
• 1991.10a
• /
• pp.111-114
• /
• 1991

This paper proposes an efficient adaptive echo canceller using pilot filter approach to achieve improved convergence speed. The pilot filter is an adaptive filter with only a few filter coefficients to filter the received signal for the purpose of whitening the signal. Thus the convergence speed of the main LMS-TDL filter combined with the pilot filter is improved. In the proposed echo canceller, an adaptive lattice predictor as the pilot filter is used and its inverse filter is used to equalize the distorted near end talker signal. Simulation results for colored signal show that the convergence speed of the proposed echo cancellation algorithm is faster than that of the conventional LMS-TDL echo cancellation algorithm.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.21 no.8
• /
• pp.2033-2042
• /
• 1996

In this papre, we describe a recently implemented echo canceller for digital cellular communication of Code Division Multiple Access(CDMA) that features time sharing of digital signal processor(DSP) over four channels in one DSP to reduce per channel costs. In the Public Land Mobile Network(PLMN), it is important to cancel the echo reflected from the Public Switched Telephone Network(PSTN) side. In case of digital mobile system, the round-trip delay of the echo is in excess of about 180 milliseconds due to frame-by-frame voice coding. It is necessary to cancel the echo in PLMN. We have developed a multi-channel echo canceller tht operates with Time Switch Module in a Mobile Switching Center(MSC). The general echo canceller needs PCM trunk interface circuits and the tone detection and disabling circuits, but the multi-channel echo canceller linked with Time Switch Module does not need them. Therefore we could develop the effective and economical echo canceller.

• Journal of the Korean Institute of Telematics and Electronics A
• /
• v.33A no.4
• /
• pp.18-26
• /
• 1996

In this paper we propose a new dual structured tap selection algorithm for voice echo canceller in digital cellular communication system, investigating adaptive filtering algorithms for echo cancellation in long distance telephony or mobile communication system. The proposed algorithm has a two-stage processing structure that after a dispersive region in an impulse response of an echo path is found out, the tap coefficients of a short length filter are adjusted adaptively for the region, because the impuse response has a very little portion of the dispersion. Simulation results show that the proposed algorithm with 256 taps gives a performance of convergence speed superior to both full-tap normalized least mean with 256 taps gives a performance of convergence speed superior to both full-tap normalized least mean square (NLMS) and a scrub taps waiting in a queue (STWQ) algorithms by about eighty per cent, also to a tap selection algorithm by about twenty per cent. And the resutls diplay that if the more tap coefficients are used due to a long delayed dispersive zone, the proposed algorithm produces the better performance.

• Proceedings of the IEEK Conference
• /
• 2006.06a
• /
• pp.383-384
• /
• 2006

The P-ECLMS algorithm adapted the existing Post-Processing method occurs the distortion of the near-end signal at the double-talk situationt. To solve this problem, we propose the SP-ECLMS(Selective Post-Processing ECLMS) algorithm which makes the Post-Processing coefficient differently at the case of the single-talk and the double-talk. When the correlation level is not output less than 30%, the proposed algorithm output the original signal to prevent the signal's distortion.