An adaptive approach for contextual audio denoising using local Fisher information

This paper presents a novel adaptive technique for audio signal denoising based on the extension of the pointwise Wiener filter through the minimization of the local Fisher information. The solution obtained by the resulting filter can be interpreted as a tradeoff between the minimization of the local mean square error and the minimization of the local observed Fisher information, since it is a linear convex combination of the two conditions. To evaluate the performance of the denoising task we use four quantitative metrics: the Peak Signal-To-Noise Ratio (PSNR), the Improvement in Signal-To-Noise Ratio (ISNR), the Noise-Reduction Ratio (NRR) and the Signal-To-Distortion Ratio (SDR). The obtained results show the effectiveness of the proposed generalized filter.