طرح دو فيلتر جديد براي بهبود كيفيت گفتار مبتني بر توزيع احتمال پسين براي ضرايب موجك

Speech Enhancement Using Two New Filters Based on a Posteriori Probability Distribution of Wavelet Coefficients

ABSTRACT Wavelet-based denoinsing methods can be divided into thresholding and filtering based methods. The thresholding methods, which are based on the concentration of signal energy in a few number of wavelet coefficients, usually remove weak speech components. This tends to poor speech quality after removing colored noises due to speech distortion. Filtering based methods apply a filter to wavelet coefficients in order to denoise them and produce less speech distortion. In this paper, we first show the superiority of filtering methods in comparison to thresholding methods. For this purpose, we compare SURE soft thresholding method with Wiener filter. Then, we propose two new filters based on a posteriori distribution of wavelet coefficients. These new filters are compared to Wiener filter in the wavelet domain. Results show better performance of proposed filters. We also compare our filters with a maximum a posteriori filter in the frequency domain. Experimental results show that proposed filters improve signal to noise ratio and give a less distorted enhanced speech signal.