A double-threshold-based approach to impulsive noise detection in audio signals

In this paper we review threshold-based algorithms for detection of impulsive noise in audio signals, and propose a new detection strategy that can increase their capability to detect impulsive disturbances without any remarkable increase in computational complexity. The main features addressed by this paper are the use of two thresholds, one for detection and the other for estimation of impulse durations; the merging of disturbances detected as adjacent, under certain circumstances; and the alteration of the threshold level during the re-processing procedure of a given signal block. Experimental results assessed by human subjects as well as objective measurements confirm the improvements attained by the proposed detection scheme.