This paper introduces a Perceptual Matching Pursuit (PMP) algorithm for audio coding. A masking model has been developed and integrated into the matching pursuit algorithm to account for the characteristics of the hearing system. By doing so, only an audible kernel is extracted at each iteration. Moreover, contrary to the matching pursuit algorithm, PMP will stop decomposing an audio signal once there is no audible part left in the residual. We have used ITU-R PEAQ to compare audio materials decomposed by PMP and by matching pursuit. Objective scores for PMP increase by up to 1 unit. A semi-formal listening test has verified the objective scores and shown the perceptual superiority of PMP over the matching pursuit algorithm.
This paper costs $33 for non-members and is free for AES members and E-Library subscribers.