Model architecture of the deployed model. Training data are generated by taking observed data and adding a pulse without noise onto the original observation scaled by a desired S/N. The data then gets preprocessed into three dynamic ranges, which introduces three additional ones. Before entering the model, the data is spliced into chunks of eight in time, which gets appended to the filter dimension. This data finally enters both the auxiliary channel to the automasking layer along with the RESNET34 model.