American dialect identification using phonotactic and prosodic features

This work is concerned with employing both prosodic and phonotactic information in developing an automatic dialect identification system. Prosodic information is considered an important part of speech from which one can infer a speaker's affiliation, age, gender, emotion, and of course dialect. Prosodic information serves both linguistic and extra-linguistic functions. The linguistic functions deal with the lexical and post-lexical properties, while the extra-linguistic functions deal with age, gender, and attitude. We are interested in the extra-linguistic features, and believe that modeling the prosodic information together with the phonotactic information will help in providing a robust identification system. In this paper, the impact of employing prosodic features on the automatic dialect identification process was studied. Results show that prosodic features improve identification accuracy over using standalone phonotactic solutions.