Unsupervised speaker adaptation of batch normalized acoustic models for robust ASR | IEEE Conference Publication | IEEE Xplore