Polyphonic audio tagging with sequentially labelled data using CRNN with learnable gated linear units.