CHARACTERISTICS OF MULTILAYER PERCEPTRON MODELS IN ENHANCING DEGRADEDSPEECH

A multi-layer perceptron (MLP) acting directly in the time-domain is a pplied as a speech signal enhancer, and the performance examined in th e context of three common classes of degradation, namely low bit-race CELP degradation ie nonlinear system degradation, additive noise, and convolution by a linear system. The investigation focuses on two topic s: (i) the influence of non-linearities within the network and (ii) ne twork topology, comparing single and multiple output structures. The o bjective is to examine how these characteristics influence network per formance and whether this depends on the class of degradation. Experim ental results show the importance of matching the enhancer to the clas s of degradation. In the case of the CELP coder the standard MLP with its inherently non-linear characteristics is shown to be consistently better than any equivalent linear structure (up to 3.2 dB compared wit h 1.6 dB SNR improvement). In contrast, when the degradation is from a dditive noise, a linear enhancer is always superior.