Acoustic Noise Reduction by Two Dimensional Spectral Smoothing and Enhancement.

Abstract

This paper proposes an acoustic noise reduction method which can recover and enhance speech formant structure as well as diffuse and suppress the noise component. A Gaussian filter is applied over a time sequence of the spectral envelope to diffuse the noise component and to recover the formant structure. The filter is called TDSS according to its function of Two Dimensional Spectral (a time sequence of the spectral envelope) Smoothing. After the application of the TDSS operator, Non-linear Spectral Amplitude Transformation (NSAT) is carried out to further suppress the noise component and enhance the formant structure. The effectiveness of this noise reduction method is shown by auditory testing and word recognition experiment.