Pole-zero modeling of noisy speech and its application to vocoding.