The goal is to design a neural-based model that given a short window in the time domain, detects whether one or more devices have been subjected to a sample drop event. The relative results on CHIME-5 and artificial data are reported here: https://arxiv.org/pdf/1911.06713.pdf