Accuracy analysis on call quality assessments in voice over IP

Voice over IP (VoIP) now has tremendous influence on the telecommunication market with its flexibility and price advantage. Users of VoIP expect call quality to be as good as, if not better than the traditional Public Switched Telephone Network (PSTN). However in VoIP, factors that are related to the IP transport network such as packet loss, delay, bandwidth, jitter, and voice encoding (codec) all affect call quality. Call quality assessment in VoIP systems is mainly conducted with off-line tests using the Perceptual Evaluation of Speech Quality (PESQ) [1] methodology. Another method that can be utilised is an on-line approach using the E-Model, which can be used in real time. However, these two methods have limits and inaccuracy, and often do not give the same results. Call quality assessment is often used to adjust system and codec parameters. Therefore, given inaccurate results, the system would decrease the adjustment efficiency or even inadvertently decrease call quality. The primary contribution of this paper is a comparison between the accuracy of PESQ and the E-Model investigated by conducting an extensive set of experiments in a real enterprise network using a widely deployed Voice over IP (VoIP) product. Experiments were conducted under varying controlled network conditions. The results show that under various conditions, loss rates, codecs and across a range of languages that there can be significant differences between the call quality measurement obtained when using the E-model versus a PESQ analysis.