Noise removal in single channel Hindi speech patterns by using binary mask thresholding function in various mother wavelets

Abstract

This paper presents a comparative analysis for enhancement of noisy single channel Hindi speech patterns', using a binary mask threshold function in mother wavelet transforms. In this wavelet transform a three level of wavelet decomposition is used and all three levels are given individually to binary mask threshold for removing noise and enhancing the speech patterns. The suitability of the binary mask thresholding function in different mother wavelet transforms (like Daubechies 13, Daubechies 40, Symlet 13, Symlet 21 and Biorthogonal 3.1) is analyzed and performance is compared in terms of SNR, PESQ, and Cepstrum analysis. The given binary mask thresholding function in Coiflet5 mother wavelet transform is more effective for removal of noises. The algorithms were implemented in MATLAB 7.1. © 2013 IEEE.