Discriminative Learning for Speech Recognition