Dual-Tree Complex Wavelet Transform in the problem of speech enhancement

Abstract

The dual-tree complex wavelet transform is considered a relatively recent improvement for the discrete wavelet transform. In this paper, the applicability of such transform in the problem of speech enhancement is evaluated. For this purpose, a nonthreshold scheme is proposed. Two filters, one for the imaginary part and other for the real part of complex wavelet coefficients were designed. These two real filters were then averaged to obtain the final filter. A uniform noise reduction was performed for all wavelet scales. Simulation results show that, together, nonthresholding scheme and dual-tree complex wavelet transform obtained consistently results.