Abstract: In this study, an adaptive averaging periodogram scheme between adjacent frequency bins, based on charac-teristics of the short-time spectral density, is proposed to optimize the bias-to-variance trade-off of the local periodogram. The proposed method is applied to the magnitude spectral subtraction in speech enhancement. Simulation results verify that the performance of the proposed algorithm is better than those conventional algorithms that used other periodogram estimation methods in terms of the segmental SNR and the log-spectral distance.