Recognizing Emotionally Coloured Dialogue Speech Using Speaker-Adapted DNN-CNN Bottleneck Features.