Discriminative training for speech recognition is compensating for statistical dependence in the HMM framework | IEEE Conference Publication | IEEE Xplore