A sequential algorithm for robust parameter estimation and enhancement of noisy speech