Résumé
This paper presents a study of automatic identification of Arabic dialectal areas based on a prosodic automatic modelling. Inspired from Fujisaki's works, this modelling dissociates long term prosodic variations from short term micro-variations and exploits n-multigrams models. Experiments, achieved on semispontaneous recordings from 40 speakers, show that the system reaches 98% of correct identification of the three dialectal areas - Maghreb, Middle-East, and an intermediate area (Tunisia-Egypt) - with test excerpts of 7.6 seconds in average.