Speech enhancement based on nonlinear spectral subtraction

Abstract

The spectral subtraction method is the most suitable technique for the elimination of stationary noise from degraded speech signal. In this article the basic principles of nonlinear spectral subtraction are described, which optimise the subtractive factor in dependence on input-degraded speech signal. In this way the undesirable musical noise remainder can be suppressed, which rises after a spectral subtraction. The obtained results are computed with the standard spectral subtraction method.