Speaker adaptation of deep neural networks using a hierarchy of output layers | IEEE Conference Publication | IEEE Xplore