Detection and restoration of click degraded audio based on high-order sparse linear prediction

Abstract

Clicks are short-duration defects that affect most archived audio media. Linear prediction (LP) modeling for the representation and restoration of audio signals that have been corrupted by click degradation has been extensively studied. The use of high-order sparse linear prediction for the restoration of clickdegraded audio given the time location of samples affected by click degradation has been shown to lead to significant restoration improvement over conventional LP-based approaches. For the practical usage of such methods, the identification of the time location of samples affected by click degradation is critical. High-order sparse linear prediction has been shown to lead to better modeling of audio resulting in better restoration of click degraded archived audio. In this paper, the use of high-order sparse linear prediction for the detection and restoration of click degraded audio is proposed. Results in terms of click duration estimation, SNR improvement and perceptual audio quality show that the proposed approach based on high-order sparse linear prediction leads to better performance compared to state of the art LP-based approaches.