Improving a Parametric Model for Speech Quality Assessment in Wireless Communication Systems

Demóstenes Z. Rodríguez, Gabriel F. Pívaro, Renata L. Rosa, Gabriel Mittag, Sebastian Möller

In: Proceedings of 26th International Conference on Software, Telecommunications and Computer Networks (SoftCOM). International Conference on Software, Telecommunications and Computer Networks (SoftCom-2018) September 13-15 Split Croatia Seiten 1-5 SoftCOM ISBN 978-9-5329-0087-3 IEEE 2018.


Telecommunication operators need tools to evaluate and improve the quality of the their services offered. In the case of telephony services, the speech quality perceived by the users is relevant. There are different speech quality assessment methods, one of them is the parametric method that is generally used in network planning. The ITU-T Rec. G107, and its recent updates, is the most representative parametric method. However, it does not consider the impairments occurred in a wireless channel. In this context, we propose the inclusion of parameters, such as fading channel model, signal-to-noise ratio (SNR), and maximum Doppler shift using different modulation schemes. Several wireless communication scenarios were simulated using actual speech samples as input. Then, the impairment speech samples were evaluated by an intrusive signal-based method described in ITU-T Rec. P.862.2, which results are used as ground-truth. Experimental results demonstrated that there is a high correlation between wireless channel parameters and mean opinion score (MOS) index. Thus, an impairment wireless function, named Iw is proposed and added to the E-model wideband algorithm. Performance validation test results demonstrated that Iw represents a reliable wireless degradation model, reaching a Pearson correlation coefficient (PCC) and a normalized RMSE (NRMSE) of 0:973 and 0:085, respectively.

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence