Finetune pretrained image recognition model with kapre melspectrogram

Question

I am trying to finetune a resnet-50 with keras while also using a melspectrogram layer from kapre. Where I fail is that resnet expects an RGB image as input (e.g. width x he