Elimination of Musical Noise Phenomenon with Burg-Based A Priori SNR Estimator

In order to eliminate the musical noise remained in the results enhanced by short-time spectral attenuation techniques, this paper proposes a novel a priori SNR estimator. The proposed estimator is built on Burg-based power spectral estimation, which takes into account the properties in estimator of smoothness, accuracy and resolution, with the emphasis on their relationships and influences on de-noising. The experimental results demonstrate that: if the proposed estimator is incorporated into the noise suppressor using minimum mean-square error (MMSE) method or log-spectral minimum mean-square error (LSA-MMSE) method, better results will be achieved with the musical noise greatly eliminated and distortion reduced further.