14th International Congress of Phonetic Sciences (ICPhS-14)

San Francisco, CA, USA
August 1-7, 1999

Investigating Dialectal Differences via Vowel System Modeling: Application to Arabic

François Pellegrino (1,2), Melissa Barkat (2)

(1) DDL, Lyon, France; (2) IRIT, Toulouse, France

This paper deals with the automatic discrimination of Arabic dialects. We focus on the relevancy of the vowel system scattering as a cue for Western vs. Middle-Eastern distinction of Arabic dialects. The former group indeed is characterized by a centralized distribution of the short vocalic segments although the latter favors a peripherical scattering. These two types of distribution in the acoustic space allowed us to consider this phonetic criterion as reliable for the discrimination of dialectal varieties in Arabic.
   The experiments are based on an automatic vowel detection algorithm and a statistical vowel system modeling. Experiments are performed using utterances recorded from a score of Arabic speakers of various geographical origin. Results show that Western dialects and Middle-Eastern dialects may be discriminated using cepstral features combined with a vowel duration cue since the rate of correct identification reaches 78 %.

Full Paper

Bibliographic reference.  Pellegrino, François / Barkat, Melissa (1999): "Investigating dialectal differences via vowel system modeling: application to Arabic", In ICPhS-14, 145-148.