An efficient algorithm for the restoration of audio signals corrupted with low-frequency pulses

Digital audio restoration of old recordings is addressed, with a focus on the removal of long pulses with low-frequency content. The main drawback of the state-of-the-art method, which is based on the separation of autoregressive (AR) processes, is its high computational complexity. A method is proposed in which the pulse tails are first estimated via a nonlinear scheme called two-pass split-window (TPSW) filtering, followed by a polynominal smoothing stage. After removing the tail of each pulse by subtraction, the remaining initial clicks are suppressed through a model-based declicking algorithm. The proposed procedure is as effective for pulse removal as the AR-based method but has a substantially lower computational complexity. Moreover, from a user point of view, its processing parameters are more intuitive and easier to adjust.