Spectral-subtraction speech enhancement in multirate systems with and without non-uniform and adaptive bandwidths

摘要

The method of spectral subtraction is widely used for single-channel speech enhancement, if the speech signal is corrupted by additive noise. It is based on the manipulation of the magnitude of the noisy-speech spectrum. Most realizations use fixed, uniformly-spaced frequency transformations or, equivalently, filter banks with identical sampling rates in each frequency band. In this paper, we generalize the basic structure: Different filter-bank systems with non-uniform and, especially, non-constant, signal-adaptive spectral resolutions and, therefore, different sampling rates are examined. An efficient realization of a time-varying band allocation is proposed. The arising problems are discussed and the enhancement results are compared to each other and to those obtained with uniform spectral transformations.