We propose in this study a way to integrate neural networks by an acoustic-phonetic decoder expert system. The goal of this coupling is to improve the recognition rate of plosive and fricative consonants. We first show that the work we carried out aimed at improving the efficiency of the phonetic knowledge base. In a second step we present a new method based of multi-layer perceptrons. This is being used to recognise plosives or fricatives and to be coupled to the expert system. We conclude by discussing how to make this cooperation possible in a hybrid system.