Underwater acoustic signal denoising Using multi-directionnals masks on time-frequency representation

This paper addresses the problem of underwater acoustic signal denoising. This field has been mainly investigated as it allows signal-to-noise ratio enhancement, a prerequisite to any data interpretation. Underwater acoustic signal denoising shares similarities with speech signal processing since both deals with acoustic signals, consequently it is possible under some manipulations to use speech processing top methods for underwater acoustic signal denoising. The acoustic underwater signal has different signatures than the speech signal, so it makes the usual well known speech denoising methods sub-optimal. In this paper, we present a new way to denoise the underwater acoustic signal, which is based on a statistical approach using multi-directional masks on the audio signal time-frequency representation. Compared to the well-known denoising methods, like Wiener filter and Ephraim and Malah algorithm, this approach results in less residual noise (still colorless) and better signal of interest enhancement from its noisy environment.