Regional accents recognition based on i-vectors approach: The case of the Algerian linguistic environment

Abstract

Regional accents recognition is one of the most important research topics in speech processing. However, the development of such systems is time-consuming and an expensive process since it requires collecting and processing a large amount of training and evaluation data for each target accent. This work presents some preliminary results about the Algerian regional accents recognition using the i-vectors approach. The experiments have been carried out using a dataset collected from the east and the center of Algeria. The obtained results show the benefits of using the i-vectors approach to recognize both regional accents and to assess the impact of evaluation data quality on the accuracy of the proposed recognition method. However, it has been noted that the confusion rates obtained between both accents may be related to the evaluation data quality, the linguistic similarity existing between both accents and some languages contact phenomenon such as the code switching.