A new approach for rustle noise canceling in pen-type voice recorder

Abstract

This paper presents a new approach for rustle noise cancellation in pen-type voice recorder. Our objective it to enhance recorded speech signal corrupted by so-called rustle noise. The proposed method is based on the signal subspace approach. In general, a signal subspace-based approach has a better performance result than the one with a method in Fourier domain. In this paper, we estimate the subspace on each subband, and design the optimum subband filter to minimize the signal distribution while reducing residual noise. A saving on computational complexity is achieved by subband domain processing. Several experimental simulations and informal listening tests are performed with actual recorded speech and it shows that the proposed method provides better noise cancellation capability than other methods such as spectral subtraction and Wiener filtering method.