Speech signal filtration using double-density dual-tree complex wavelet transform

We consider the task of increasing the quality of speech signal cleaning from additive noise by means of double-density dual-tree complex wavelet transform (DDCWT) as compared to the standard method of wavelet filtration based on a multiscale analysis using discrete wavelet transform (DWT) with real basis set functions such as Daubechies wavelets. It is shown that the use of DDCWT instead of DWT provides a significant increase in the mean opinion score (MOS) rating at a high additive noise and makes it possible to reduce the number of expansion levels for the subsequent correction of wavelet coefficients.