Tt. Le et Js. Mason, ARTIFICIAL NEURAL NETWORKS FOR NONLINEAR TIME-DOMAIN FILTERING OF SPEECH, IEE proceedings. Vision, image and signal processing, 143(3), 1996, pp. 149-154
A multilayer perceptron (MLP) is applied as a time domain noalinear fi
lter to two classes of degraded speech, namely gaussian white noise an
d nonlinear system degradation introduced by a low bit-rate CELP coder
. The goal of the study is to examine the influence of the inherent no
nlinearity within the MLP, and this is achieved by varying the levels
of nonlinearity within the structure. Direct comparisons of MLPs and l
inear filters show that with CELP degradation the SNR improvements ach
ieved by the MLP is measurably better than with an equivalent linear s
tructure (3dB cf 1.5 dB) but when the degradation is additive noise th
e two structures perform equally well, The study highlights the import
ance of scaling to achieve optimum performance, and of matching the en
hancer to the degradation.