MODIFIED LMS ALGORITHMS FOR SPEECH PROCESSING WITH AN ADAPTIVE NOISE CANCELER

A desired signal corrupted by additive noise can often be recovered by an adaptive noise canceller using the least mean squares (LMS) algori thm. A major disadvantage of the LMS algorithm is its excess mean-squa red error, or misadjustment, which increases linearly with the desired signal power. This leads to degrading performance when the desired si gnal exhibits large power fluctuations and is a serious problem in man y speech processing applications. This work considers two modified LMS algorithms, the weighted sum and sum methods, designed to solve this problem by reducing the size of the steps in the weight update equatio n when the desired signal is strong. The weighted sum method is derive d;from an optimal method (also developed in this work), which is not g enerally applicable because it requires quantities unavailable in a pr actical system. The previously proposed, but ad hoc, sum method is ana lyzed and compared to the weighted sum method, Analysis of the two mod ified LMS algorithms indicates that either one provides substantial im provements in the presence of strong desired signals and similar perfo rmance in the presence of weak desired signals, relative to the unmodi fied LMS algorithm. Computer simulations with both uncorrelated Gaussi an noise and speech signals confirm the results of the analysis and de monstrate the effectiveness of the modified algorithms. The modified L MS algorithms are particularly suited for signals (such as speech) tha t exhibit large fluctuations in short-time power levels.