A modified a priori SNR for speech enhancement using spectral subtraction rules

The letter addresses the problem of single-channel speech enhancement using a spectral subtraction method. The proposed approach is directed toward finding a self-adaptive averaging factor to estimate the a priori SNR. Performance of the modified averaging factor is evaluated using conventional spectral-subtraction algorithms in which the averaging factor is otherwise kept as a constant. Improved results are obtained in terms of speech quality measures for various types of noise and at different SNR levels when the proposed time-frequency varying averaging factor is adapted in the conventional subtraction rules.