Digital Signal Processing Reference
In-Depth Information
Figure 10-8.
LAR-distances distances between the clean input and various output signals:
unprocessed output (Noisy), outputs of the SAF, STDWF and SAFWF techniques for
CarWClose case and different input SNRs.
since Wiener filters typically generate less speech distortion at high input
SNRs.
In order to examine the effect of input SNR, we have repeated the
objective assessments for the fourth set of noise recordings (CarWClose) at
input SNRs of 0, 5, and 10 dB. As depicted in Figure 10-8, the SAFWF
offers better performance in all cases. Considering LAR improvements of the
SAFWF (relative to the LARs for the unprocessed output) at various SNRs
(3, 2.5 and 2 for SNRs of 0, 5, and 10 dB, respectively) the superiority of the
SAFWF is more evident at low input SNRs. Also, notice that the SAF
performance improvement is almost the same for different SNRs. This is
expected as the adaptive noise canceller by design, removes the correlated
noise components.
