Tied-state based discriminative training of context-expanded region-dependent feature transforms for LVCSR | IEEE Conference Publication | IEEE Xplore