Subband-based Spectrogram Fusion for Speech Enhancement by Combining Mapping and Masking Approaches