Impact of Packet Loss Location on Perceived Speech Quality

In VoIP applications, packet loss can have a major impact on perceived speech quality. The impact is affected by factors such as packet loss size, loss pattern and loss locations. In this paper, we report an investigation into the impact of loss location on perceived speech quality and the relationships between convergence time and loss location for three different codecs (G.729, G.723.1 and AMR) using perceptual-based objective measurement methods (PSQM+, MNB and EMBSD). Our results show that loss location has a severe effect on perceived speech quality. The loss at unvoiced speech segments has little impact on perceived speech quality for all codecs. However, the loss at the beginning of voiced segments has the most severe impact on perceived speech quality. The convergence time depends on the speech content (voiced/unvoiced). For unvoiced segments, the convergence time is stable whereas for voiced segments it varies but has an upper bound at the end of the segment. Our method allows a more accurate measurement of the exact effect of packet loss on perceived speech quality. This could help in the development of a perceptually relevant packet loss metric, which could be valuable in non-intrusive VoIP measurements.