A GA-NN based wavelet method of speech signal denoising

The denoising of speech signals is performed via filtering of its components and reconstructing them to original signal. The segmentation of signal into components through wavelet denoising has an additional benefit of temporal resolution over Fourier transform. The filtering of wavelets in DWT process is a concern of thresholding technique and the overall performance of system is studied in terms of SNR and EER generally. The optimization of wavelets in terms of its number of levels and value of threshold is also an important factor in defining the excellency of this method. In this paper, we review the sincere efforts of researchers to modify DWT in search of the best denoising approach. Following it, the paper shows that Neural Network based thresholding of wavelets have better SNR and lower EER compared to other methods. Also, an enhanced version of NN through GA is proposed for selecting number of wavelets and threshold. At the end of paper, the results of algorithms are compared in a symmetric fashion to understand their limitations and search the scope of improvement to meet the requirement of present and near-future necessities of digital communication.