New Semicausal and Noncausal Techniques for Detection of Impulsive Disturbances in Multivariate Signals With Audio Applications

摘要

This paper deals with the problem of localization of impulsive disturbances in nonstationary multivariate signals. Both unidirectional and bidirectional (noncausal) detection schemes are proposed. It is shown that the strengthened pulse detection rule, which combines analysis of one-step-ahead signal prediction errors with critical evaluation of leave-one-out signal interpolation errors, allows one to noticeably improve detection results compared to the prediction-only based solutions. The proposed general purpose approach is illustrated with two examples of practical applications-elimination of impulsive disturbances (such as clicks, pops, and record scratches) from archive audio files and robust parametric spectrum estimation.