1、PDF外文：http:/ Computing andCommunication Journal附 录 ADAPTIVE WIENER FILTERING APPROACH FOR SPEECH ENHANCEMENT M. A. Abd El-Fattah*, M. I. Dessouky , S. M. Diab and F. E. Abd El-samie # Department of Electronics and Electrical communications, Faculty ofElectronicEngineeringMenoufia
2、University, Menouf, Egypt E-mails: * maro_ , # fathi_ ABSTRACT This paper proposes the application of the Wiener filter in an adaptive manner in speech enhancement. The proposed adaptive Wiener filter depends on the adaptation of thefilter transfer function from sample to sample based on the s
3、peech signal statistics(meanand variance). The adaptive Wiener filter is implemented in time domain rather than infrequency domain to accommodate for the varying nature of the speech signal. Theproposed method is compared to the traditional Wiener filter and spectral subtractionmethods and the resul
4、ts reveal its superiority. Keywords: Speech Enhancement, Spectral Subtraction, Adaptive Wiener Filter 1 INTRODUCTION Speech enhancement is one of the mostimportant topics in speech signal processing.Several techniques have been proposed for thispurpose like the spectral subtraction
5、 approach, thesignal subspace approach, adaptive noise cancelingand the iterative Wiener filter1-5 . Theperformances of these techniques depend onquality and intelligibility of the processed speechsignal. The improvement of the speech signal-tonoiseratio (SNR) is the target of most techniques. Spect
6、ral subtraction is the earliest method forenhancing speech degraded by additive noise1.This technique estimates the spectrum of the clean(noise-free) signal by the subtraction of theestimated noise magnitude spectrum from the noisysignal magnitude spectrum while keeping the phasespectrum of the nois
7、y signal. The drawback of thistechnique is the residual noise. Another technique is a signal subspaceapproach 3. It is used for enhancing a speechsignal degraded by uncorrelated additive noise orcolored noise 6,7. The idea of this algorithm isbased on the fact that the vector space of the noisysigna
8、l can be decomposed into a signal plus noisesubspace and an orthogonal noise subspace.Processing is performed on the vectors in the signalplus noise subspace only, while the noise subspaceis removed first. Decomposition of the vector spaceof the noisy signal is performed by applying aneigenvalue or
9、singular value decomposition or byapplying the Karhunen-Loeve transform (KLT)8.Mi. et. al. have proposed the signal / noise KLTbased approach for colored noise removal9. Theidea of this approach is that noisy speech framesare classified into speech-dominated frames andnoise-dominated frames. In the
10、speech-dominatedframes, the signal KLT matrix is used and in thenoise-dominated frames, the noise KLT matrix isused. In this paper, we present a new technique toimprove the signal-to-noise ratio in the enhancedspeech signal by using an adaptive implementationof the Wiener filter. This implementation
11、 isperformed in time domain to accommodate for thevarying nature of the signal. The paper is organized as follows: in sectionII, a review of the spectral subtraction technique ispresented. In section III, the traditional Wienerfilter in frequency domain is revisited. Section IV,proposes the adaptive
12、 Wiener filtering approach forspeech enhancement. In section V, a comparativestudy between the proposed adaptive Wiener filter,the Wiener filter in frequency domain and thespectral subtraction approach is presented. 2 SPECTRAL SUBTRACTION Spectral subtraction can be categorized as anon-p
13、arametric approach, which simply needs anestimate of the noise spectrum. It is assume thatthere is an estimate of the noise spectrum that istypically estimated during periods of speakersilence. Let x(n) be a noisy speech signal : x(n) = s(n) + v(n) &n
14、bsp; (1) where s(n) is the clean (the noise-free) signal, andv(n) is the white gaussian noise. Assume that thenoise and the clean signals are uncorrelated. Byapplying the spectral subtraction approach thatestimates the short term magnitude spectrum of thenoise-free
15、 signal S by subtraction of theestimated noise magnitude spectrum )(V fromthe noisy signal magnitude spectrum X It issufficient to use the noisy signal phase spectrum asan estimate of the clean speech phasespectrum,10: XjNXS e x p (2) The
16、 estimated time-domain speech signal isobtained as the inverse Fourier transform ofS . Another way to recover a clean signal s(n)from the noisy signal x(n) using the spectralsubtraction approach is performed by assumingthat there is an the estimate of the power spectrumof the noisePv( ) , that is ob
17、tained by averagingover multiple frames of a known noise segment.An estimate of the clean signal short-time squaredmagnitude spectrum can be obtained as follow 8: o t h e r w i s e vPXifvPXS ,0 0, 222 (3) It is possible combine this magnitude spectrumestimate with the measured phase and then get theShort Time Fourier Transform (STFT) estimate asfollows:
