Modified Wiener Filter based Speech Enhancement Technique

Abstract

In order to achieve effective information exchange the quality and intelligibility newlineof speech are of prime importance in any speech communication system. Basically a newlinespeech processing system is designed considering noise free environment. However in actual usage the system performance is affected due to the noise present. The result of this noise is to affect adversely the pleasantness and intelligibility of the speech. There are several speech enhancement techniques to improve speech quality in such noisy environment. The existing enhancement techniques are investigated and a new newlinemethod of speech enhancement is proposed. Short-Time Spectral Amplitude (STSA) newlineestimation approach is one of the most powerful estimation techniques widely used in newlinespeech enhancement scheme. Several enhancement schemes developed assume a flat newlinepower spectrum for noise, though the assumption is not true for colored noise. Hence a new multi-band speech enhancement system is developed using a different estimator factor for each frequency sub-band. However with this scheme some harmonics newlinepresent in the clean signal are suppressed as with other conventional methods. To newlineovercome this problem, harmonic regeneration is proposed. But some residual musical noise remained in the resulting speech signal. A psycho acoustically motivated constrained perceptual weighting filter is designed and used as a post processing technique to reduce musical noise. While improving the speech quality with the scheme it was observed that there is degradation of unvoiced speech or high frequency bands containing low energy. For further improvement of speech quality newlineand reduce degradation of unvoiced speech a new method is proposed. In the proposed scheme during noise estimation the noise is updated in all the frames irrespective of the condition that speech signal is present or absent.