Fig. 7From: A noise PSD estimation algorithm using derivative-based high-pass filter in non-stationary noise conditionsAverage PESQ score comparison on the NOIZEUS corpus corrupted with Mod. WGN, babble, traffic, passing car, and passing train noises at multiple SNR levels (from −10 to +10dB, in 5 dB increments), where the estimated noise PSD by the competing methods are used in the DD approach to compute the a priori SNR for utilizing the a SRWF ([14], Section 6.5.1 of Chapter 6), b MMSE-STSA [4], and c MMSE-LSA [5] gain functions in speech enhancement contextBack to article page