Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing
Fig. 9 | EURASIP Journal on Audio, Speech, and Music Processing

Fig. 9

From: A noise PSD estimation algorithm using derivative-based high-pass filter in non-stationary noise conditions

Fig. 9

Comparing the spectrograms of a clean speech (concatenate sp05 and sp12) and b noisy speech (corrupt (a with 5 dB passing train noise (PESQ=1.80), to that of the enhanced speech produced by the SRWF-based SEA ([14], Section 6.5.1 of Chapter 6), where the noise PSD estimates are given by c MS [17] (PESQ=1.82), d MCRA [18] (PESQ=1.83), e MMSE-LC [21] (PESQ=1.85), f U-MMSE [22] (PESQ=1.93), g IMMSE [25] (PESQ=1.92), h NPP-DNN [27] (PESQ=2.01), and i Proposed (PESQ=2.13) methods

Back to article page