The incidence of attacks on audio signal self-recovery schemes has been growing. One of the more severe attacks is the discordant size content replacement attack. This attack replaces a set of samples of an audio signal with another set of samples of different sizes. This paper proposes a mathematical model of a communication channel subject to the discordant size content replacement attack. The goal is obtain a mathematical description and input-output relation, to detect how the attack tampered with an audio signal, to analyse its behaviour, and to use the model to design new schemes for audio signal self-recovery that are robust against this attack. Models of content replacement attacks with equal, larger, and smaller sizes are presented. Tests show that these models function correctly.