Denoising for improving time-frequency representations

Abstract

In this paper, we analyze properties of the noise in the time-frequency representations. We show that time-frequency representations can be improved by denoising. However, the task and conditions in which noise removal should be performed are not typical for filter design. We demonstrate that leaked noise has specific statistics and spatial correlation characteristics. Some opportunities of denoising application to the considered practical situation are discussed and tested.