Wavelet packet based multiple noises suppression in single channel speech using binary mask threshold

Abstract

This paper presents a comparative evaluation of various mother wavelets for reduction of mixed highly non-stationary noises to increase the quality and intelligibility in single channel speech. A binary mask function is used for thresholding in noisy wavelet packet coefficients. For speech signal decomposition, various mother wavelets (such as Daubechies13 and 40, Symlet2 and 8, Coiflet1 and 5, Bior 1.1 and 6.8, rbio 1.1 and 6.8) are used and results are compared in terms of performance measure parameters like SNR, PESQ, and Cepstral Distance. The performance comparison of above mentioned mother wavelets, the binary mask thresholding function with Coiflet 1 mother wavelet transform is showing good results. The algorithms were implemented in MATLAB 7.1. © 2014 IEEE.