Standard Arabic is distinctive from other Indo-European languages by the articulation of sounds in the back part of the vocal track, by the feature of gemination and by the complexity of certain consonants from a velarisation. The stop consonants of Arabic do not escape these particularities and form the object of our study within the frame of speech recognition. With the help of a mixed system using Prolog rules and neural networks conjointly, we locate and identify the occlusives of Arabic as well as the nasal consonants in an ascendant phase of Acoustic-Phonetic Decoding. Keywords : Acoustic-Phonetic Decoding, standard Arabic, Prolog, neural network.