Speaker environment classification using rhythm metrics in Levantine Arabic dialect

This paper investigates the relationship between rhythm metrics and the ability to classify speakers depending on gender and/or social environments that may have been affected by factors such as second language effects and ways of living as expressed through speech. The BBN/AUB (BBN Technologies and American University of Beirut) corpus was used; it contains four subsets of native Levantine dialect speakers of both genders from different locations. Classification was conducted using rhythm metrics and artificial neural networks (ANNs). The ANN classifier results showed 65.22% accuracy using only the Interval Measures metrics. The ANN classifier was able to reach 70.79% accuracy when all 11 rhythm metrics were used.