Investigating speaker gender using rhythm metrics in Arabic dialects

This paper investigates the relationship between speakers' gender and rhythm metrics (i.e., %V, ΔC, and ΔV) for three Arabic dialects, namely Modern Standard Arabic, Saudi Arabic, and Levantine. To carry out this study, three corpora have been used: the West Point, Saudi Accented Arabic Voice Bank (SAAVB), and BBN/AUB DARPA Babylon, for Modern Standard Arabic, Saudi Arabic, and Levantine, respectively. Because of differences (e.g., in the language of the spoken content) among the corpora, we compare genders within each specific corpus. Variations are found in speech rate between the three dialects and even within the same dialect between males and females. Rhythm metrics can be used to identify the gender and dialect.