We propose in this paper a phonetic study of standard Arabic based essentielly on the spectrographic visions of 50 sentences of the DJOUMA corpus we have constituted. This study allowed us to determine the pertinent parameters of continuous speech recognition. For the recognition part, we present the algorithms developed and the results obtained for three important questions: The segmentation of speech into gross phonetic classes. This process is based* on methodes adapted from the APHODEX expert system developed in our group [Foh 86], [Car 86]. The recognition of vowels in continuous speech for multispeakers. The detection of parameters necessary for identification of plosives and fricatives. Results obtained on speech pronounced by 5 different speakers are presented and discussed, together with preliminary results for recognition of Arabic consonants.