Speech recognition in noisy environments using discrete hidden Markov models