Learning discriminative features in sequence training without requiring framewise labelled data

Re-assign community