Font Size: a A A

Speech Enhancement Based On An Improved Noise Estimation Method

Posted on:2013-03-08Degree:MasterType:Thesis
Country:ChinaCandidate:Y YuFull Text:PDF
GTID:2248330371994340Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Speech is for the most a natural means of communication. Communication can begreatly hindered by noise. We are surrounded by noise wherever we go. Noise is present,for instance, in the street(e.g., car passing by, street construction work), the car(e.g., enginenoise, wind), the office(e.g., PC fan noise, air ducts), restaurants(e.g., people talking innearby tables), and department stores(e.g., telephone ringing, sales representatives talking).In order to improve speech processing performance, speech enhancement system is tosuppress or completely remove the unwanted noise while maintaining the quality andintelligibility of the speech.Firstly, this paper introduces the background and meaning of speech enhancementand current research status. The second chapter expounds the basic concept about speechenhancement and introduces the classification of speech enhancement algorithms incommon. Chapter three focus on the noise power spectrum estimation. In addition, thisarticle also focused on improved noise estimation with spectral subtraction and wienerfilter, respectively. The spectral-subtractive algorithm is historically one of the firstalgorithms proposed for noise reduction. It is based on a simple principle. Assumingadditive noise, one can obtain an estimate of the clean signal spectrum by subtracting anestimate of the noise spectrum from the noisy speech spectrum. The noise spectrum can beestimated, and updated, during periods when the signal is absent. However, a largedifference between the estimated value and the actual value. This paper proposed a newmethod for noise estimation. The method considers the correlation between frequencybands. In contrast to using a fixed window for tracking the minimum of noisy speech inother methods, the noise estimate is updated continuously by smoothing the noisy speechpower spectra in each frequency bin using a nonlinear smoothing rule. Compute the time-frequency-dependent smoothing factor according to speech presence probability. Thesubtraction of the noise spectral from the noisy spectral introduces a distortion in the signalknown as musical noise. So, we turn our attention to a wiener filtering. Based on the newmethod for noise estimation and harmonic regeneration, we proposed a novel method. Thismethod can mask the residual musical noise effectively with the regenerated speech components. The last chapter analyzes all kinds of speech enhancement algorithmperformance described in this paper.
Keywords/Search Tags:speech enhancement, noise estimation, spectral subtraction, harmonicregeneration, wiener filter
PDF Full Text Request
Related items