A Comparative Performance Analysis of Wavelets in Denoising of Speech Signals

It is known that data or signal obtained from the real world environment is corrupted by the noise. In most of the cases this noise is strong causing poor SNR and therefore, need to be removed from the desired signal before further processing of signal. Research in the area of wavelets showed that wavelet shrinkage method performs well and efficiently as compared to other methods of denoising. Here, we present a comparative analysis of performance of various types of wavelets i.e. Haar, Db10, Coif5, Bior3.3 and Sym5 in denoising of speech signals in the presence of White Gaussian noise. In the process of denoising, scale dependent Visu Shrink employing universal threshold selection criteria (square-root-log) method for deciding the threshold levels for truncating the wavelet coefficients with soft thresholding is used. Along with the performance evaluation of different types of wavelets, the effect of wavelet decomposition levels is also investigated. The quality of denoised speech signal is expressed in terms of Peak Signal to Noise Ratio (PSNR) as compared to original noiseless speech signal