We begin with two basic approaches for processing the noisy speech signals in the time-domain. The first method forms a direct mapping from noisy speech to clean speech, while the second approach uses a predictive model in conjunction with an extended Kalman filtering approach.