Experimental mappings and validation of the dependence on the language of objective speech quality scores in actual GSM network conditions

The measure of quality of service (QoS) is a priority for telecommunication companies, particularly the measure of speech quality. This measure is generally realized by objective quality rating models in GSM networks. In this paper, we address the dependence of such objective quality assessment algorithms on the language of the speech under test, especially the Perceptual Evaluation of Speech Quality (PESQ) model, standardized by the ITU-T Rec. P.862, and another commercial rating model. We then point out the dependence of the objective scores on the used evaluation model. The presented study is based on the analysis of an important measurement database, carried out in actual GSM network conditions. Since this work and the presented and discussed results were carried out in collaboration with a Tunisian mobile communication operator, Orascom Telecom Tunisie (Tunisiana), particular attention is paid to the adequacy of such objective quality rating models to the Arabic language.