Enhanced Signal-Dependent Rank-Ordered Mean-Based Audio Noise Removal

This study focuses on restoring audio signals corrupted by impulse noise added to them in a random manner, either when generated or when recorded, transferred or processed. More specifically, we investigate the application of a variant of the well-known Signal-Dependent Rank-Ordered Mean (SDROM) method using simulated addition of impulse noise to clean signals. Our study describes an automatic exploration of the threshold space used in this method and selection of optimal thresholds based on global minima for the mean energy of the error in noise identification. Experimental results showed that the method performs well for noise levels up to 10% of the signal.