An improved adaptive predictor for uncorrelated noise reduction in hearing aids

This work presents a new low-cost technique to reduce uncorrelated acoustical noise in digital hearing aids. The clean speech is estimated by a convex combination of the conventional adaptive linear predictor output and the contaminated input signal. The convex combination coefficient is adjusted to control the attenuation of the uncorrelated noise while avoiding significant unvoiced utterance distortion. The proposed adaptive architecture is presented and designed to minimize the mean square prediction error. Comparisons with the conventional adaptive predictor indicate a superior performance of the new solution. The proposed technique can be easily implemented in existing hearing aids systems that incorporate the adaptive predictor, with a minimum amount of extra computational resources.