Discriminative training of HMMs for automatic speech recognition: A survey