Prosody-based Spoken Algerian Arabic Dialect Identification

Abstract

Dialect is the most common way of communication in every day life. Automatically identifying a dialect is a challenging task, especially when we deal with close dialects within the same country. In this paper, we address the problem of Spoken Algerian Arabic Dialect Identification (SAADID). Indeed, we propose a new system based on prosodic speech information, namely intonation and rhythm. The rhythm features are got using a coarse-grained consonant/vowel segmentation. The performance of this approach is shown through experiments on six dialects of the departments of Adrar, Algiers, Bousaâda, Djelfa, Laghouat and Oran. The results prove the suitability of our prosody-based system for SAADID with more than 69% of precision using 2s test utterances.