Application of wavelet transform and wavelet thresholding in robust sub-band speech recognition

Sub-band speech recognition approaches have been proposed for robust speech recognition, where full band speech are divided into several sub-bands and then likelihood or cepstral vectors of the sub-bands are merged depending on their reliability. In this paper, we use wavelet transform for splitting of input speech into sub-bands in two ways: equal bandwidths for sub-bands and dyadic bandwidths for sub-bands. We also use wavelet thresholding techniques to determine a criterion for reliability of sub-bands and use it as weight in sub-band recombination.