Block-online multi-channel speech enhancement using deep neural network-supported relative transfer function estimates