Robust noise estimation applied to different speech estimators

In this paper we present robust noise estimation for speech enhancement algorithms. The robust noise estimation based on a modified minima controlled recursive averaging noise estimator was applied to different speech estimators. The investigated speech estimators were spectral subtraction (SS), log spectral amplitude speech estimator (LSA) and optimally modified log spectral amplitude estimator (OM-LSA). The performances of the different algorithms were measured both by the signal-to-noise ratio (SNR) and recognition accuracy of automatic speech recognition (ASR).