Bayesian VoIP Quality Evaluation Using the Time Between Retried Calls

This paper presents a new method for determining the voice quality of a VoIP carrier at a given time. The current methods rely on the voice stream analysis, their main drawback being the large computational power required and the cost for licenses. The method presented in this paper analyses the time difference between successive calls from the same source number to the same destination number, and identifies a connection between this calling pattern and the voice quality of a VoIP carrier. The advantage of this method relies in its low computational power requirement, as the voice stream analysis is no longer required. Bayesian networks are built to infer the exact a posteriori probability of poor voice quality. A case study was performed on a prepaid long distance carrier and the results show that a large percent (above 50%) of the customer complaints can be identified using this method.