This paper proposes a restoration mechanism based on convolution neural network (CNN) for electronic camouflage audio. Since there are certain change rules in the process of converting original audio into electronic camouflage audio and audio is short-time stationary, convolution and nonlinear mapping are performed on the historical sampling acoustic information and restoring factors of the electronic camouflage audio. After companding transformation, the reduction audio is outputted. In the experiment, the voiceprint features comparison, LPC analysis and human ear identity judgement are made between restoring audio and original audio. The results show the validity of the proposed mechanism. It is of great theoretical and practical significance to the restoration of electronic camouflage audio in judicial expertise.