Marek Pawelczyk, Silesian Technical University (Poland)
In the paper a new approach to stability problem of feedforward control with LMS identification is presented. A modification improving speed of LMS identification is introduced. Then new narrowband noise cancellation algorithms are described and their results are compared to results obtained by FIR and IIR filters. The idea of the algorithms proposed is extended to broadband noise cancellation. In the last part of the paper sampling with various frequencies is considered and concept of multirate signal processing is proposed as a solution for extending of the attenuation band.
Marc Ihle, Universitaet Karlsruhe (Germany)
K. Kroschel, Universitaet Karlsruhe (Germany)
In this paper a front-end for handset-free telecommunication is presented, which combines noise reduction and echo attenuation, exploiting phenomena of hearing physiology. Simple and robust techniques for the adaptation to the speaker render a system which can be used in extremely noisy environments around 0 dB, which are typical in small vans. The implementation in hardware is highly eco- nomic because some functions are used commonly for both tasks.
Henrik Sahlin, Chalmers University of Technology (Sweden)
Holger Broman, Chalmers University of Technology (Sweden)
A signal separation algorithm is used in the present paper in order to improve the Signal to Noise Ratio (M) of a signal disturbed with noise. The algorithm uses a criterion of squared crosscorrelations between separated signals, and is thus based on second order statistics. Leaking is introduced in order to improve the performance. The signals used axe real world signals measured with a modified mobile unit with two microphones. Signal to noise ratios, before and after separation, axe presented. Furthermore, it is shown how to compute the SNR for signals in scenarios when both the mixing system and the noise signals are unknown.
Martin Drews, Technical University of Berlin (Germany)
Martin Streckfuss, Siemens AG Semiconductors (Germany)
A multi-channel speech enhancement system with 16 microphones is presented which consists of a conventional delay-and-sum beamformer and an adaptive postfilter. The post-filter adaptation is performed by Wiener filter analysis. The adaptation scheme is improved by a method for frequency-dependent channel selection, a modified method for speech spectrum estimation, and auditory constraints. The selective processing yields small speech spectrum estimation errors, thus providing a high noise reduction. The application of the improved post-filter to the delay-and-sum beamformer results in a clear improvement of the speech signal quality even if only 4 microphones are used.
Walter Kellermann, Fachhochschule Regensburg (Germany)
For a recently proposed concept combining acoustic echo cancellation (AEC) and adaptive beamforming microphone arrays (ABMAs), crucial design and control issues are discussed. For ABMAs, data-independent and data-dependent beamforming algorithms are considered. While the actual signal processing of ABMA and AEC can be largely decoupled, efficient implementations benefit from control mechanisms overviewing the entire system. Key design parameters for typical microphone array applications are discussed.
Rainer Martin, Aachen University of Technology (Germany)
Stefan Gustafsson, Aachen University of Technology (Germany)
Mario Moser, Aachen University of Technology (Germany)
In this contribution we propose and experimentally verify an algorithm for the cancellation of acoustic echoes for signals derived from a beamforming microphone array. Almost independent of the number of microphones this algorithm has the computational complexity of only a single echo canceller. It requires, however, additional memory to store several coefficient vectors for this echo canceller. Experimental results show that after a brief training period the new approach can outperform the computationally expensive conventional approach, where one echo canceller is used for each of the microphones.
Luc De Coster, Katholieke Universiteit Leuven (Belgium)
Rudy Lauwereins, Katholieke Universiteit Leuven (Belgium)
J.A. Peperstraete, Katholieke Universiteit Leuven (Belgium)
In this paper we describe the rapid prototyping of an adaptive noise canceler using the Graphical RApid Prototyping Environment (GRAPE), developed at our laboratory. It is an environment which facilitates the realtime emulation and implementation of synchronous DSP applications on heterogeneous target platforms consisting of DSPs and FPGAs. Our work demonstrates the feasibility of real-time prototyping using a multiprocessor and a powerful programming framework.
Fabrice Amand, Cefriel (Italy)
Ivan Bourmeyster, Alcatel Mobile Phone (France)
Giorgio Parladori, Alcatel Advanced Technologies (Italy)
This paper deals with a speech recognition system in a car facing a car radio. The speech recogniser can be used either for a mobile phone or for a board computer. So, it will be possible to activate the mobile phone together with the car radio. It will also be possible to voice control the board computer in the same context (car radio active). The sound diffused by the car radio loud-speakers disturbs the speech recogniser [11. To cancel the loud-speakers echo picked up by the microphone, we show and evaluate the use of an Acoustic Echo Cancellation technique in the Stereophonic case (AECS).
F. Berthault, Matra Communication (France)
C. Glorion, Matra Communication (France)
Francois Capman, Matra Communication (France)
Jerome Boudy, Matra Communication (France)
Philip Lockwood, Matra Communication (France)
During the last decades, several adaptive filtering algorithms have been optimised in the context of speech echo cancellation. The use of such algorithms for the realisation of a hands-free function Is now technically feasible for GSM mobile telephony, In car environment. Recent works now focus on the multi-channel case, and investigate the use of stereophonic echo cancellers for high-quality visioconference systems. In this paper, we address the Issue of stereophonic car radio noise compensation, for speech recognition.
Matthias Dorbecker, Aachen University of Technology (Germany)
A common approach to enhance the quality of speech signals disturbed by acoustic background noise is the application of adaptive filtering techniques aiming at the reduction of the noise (e.g. ). However, in case of low SNR most of the currently known adaptive techniques result in a poor quality of the processed signal, which is caused by time-variant distortions of the speech signal and by the unnatural character of the remaining noise (e.g. in form of "musical tones"). An alternative approach, which does not affect the speech signal by time-variant distortions, is the application of a microphone array with a fixed directivity pattern aligned to the speaker's position, resulting in a suppression of spatial distributed noise sources (e.g. [41). Within the scope of this paper it will be shown that owing to the proposed optimization of the directivity pattern even an array consisting of only two microphones may offer a performance comparable to state-of-the-art adaptive filtering techniques. Acoustic measurements related to electronic hearing aids confirm that the improvement of the SNR predicted by theory also holds for signals recorded in a real acoustic environment.
S. Bourennane, CMCS URA CNRS (France)
In the context of the narrow-band or wideband array processing problem, in this paper we develop a robust algorithm to improve the accuracy of the estimation of the direction of arrival of the sources. It is well known that when the noise cross-spectral matrix is unknown, these estimates may be grossly inaccurate Using both a propagation operator and noneigenvector algorithm to estimate the noise subspace projection matrix we develop a new robust algorithm for the source characterisation problem in the presence of noise with an unknown cross-spectral matrix. When shall show that the performance of bearing estimation algorithms improves substantially when our robust algorithm is used. Simulation results are presented for the band noise spectral matrix.
Amir Hussain, University of Paisley Scotland (U.K.)
Douglas R. Campbell, University of Paisley Scotland (U.K.)
Thomas Moir, University of Paisley Scotland (U.K.)
A multi-microphone sub-band adaptive speech enhancement scheme using a human cochlear model is presented. The effect of distributing the sub-bands nonlinearly as in humans is investigated. A new robust metric is developed in order to automatically select the best form of diverse processing within each sub-band. Comparative results achieved in simulation experiments demonstrate that the proposed scheme employing diverse processing in cochlear spaced sub-bands is capable of significantly outperforming conventional noise cancellation schemes.
Y. Loke, Imperial College of Science Technology and Medicine (U.K.)
J. Chambers, Imperial College of Science Technology and Medicine (U.K.)
Recent analysis on the leaky Least Mean Square (LMS) adaptive filter has justified the use of leakage factor in numerous applications. In this work, a similar leakage factor is introduced in the two-channel LMS, and the eXtended LMS (XLMS) algorithms for use in stereophonic acoustic echo cancellation. This is compared to the alternative of adding random white noise to the input stereo signals. Simulations and experimental result., indicate that the leakage factor is superior compared to the direct addition of random white noise. Performance measures used are based on output error and weight error vector norms.
Djamila Mahmoudi, Swiss Federal Institute of Technology (Switzerland)
Andrzej Drygajlo, Swiss Federal Institute of Technology (Switzerland)
This paper presents a new method of localization and tracking of a moving speech source in an adverse environment using a microphone array. To provide a good quality of communication, the problem of enhancing the acquired speech signal is also addressed. The source localization is applied to the sub-band decomposition and is achieved in two stages. First, the location of a coarse region where the speech source is present is detected. Then the multi-beamforming operation together with the discrimination function are used to pinpoint the speaker's location. Both stages are based upon the concept of examining the beam signal energy and its variation. Then, a postfiltering based noise reduction system is applied to attenuate the noise effect. The complete algorithm provides high noise reduction and relatively small errors in the source localization estimate. The system was developed in the multiresolution wavelet transform domain using a fast algorithm with short prototype filters. This results in a significant reduction in computational load and gives a minimum delay in sub-band processing.
Chuijie Yi, TU-Clausthal (Germany)
P. Dietz, TU-Clausthal (Germany)
X.M. Lian, Tsinghua University (China)
X.Y. Jiang, Tsinghua University (China)
This paper represents a flexible sound intensity measurement system known as SIMS. It has developed successfully and applied to noise measurement and analysis in vehicular engineering. Comparison with these sound intensity instruments in commercial market, SIMS has two main strengths: (1) it is indeed inexpensive relative to the instruments mentioned above and (2) it is more flexible to adapt to the noise measurement and analysis of vehicles in motion. In the latter half of the paper, SIMS have been used to identify noise sources of luxurious buses and analyze the sound isolation behavior of the fuselage of helicopters. SIMS was found to be sufficient for tackling problems associated with the noise measurement and analysis in vehicular engineering.