Traffic characteristics of packet voice

Packet voice is used for audio communication in multimedia applications which are expected to dominate bandwidth requirement in future communication networks. Existing performance studies to date, however, are all based on a traffic model that was developed for digitized voice. This paper presents the characteristics of packet voice traffic, and studies the suitability of the traditional model for packet voice traffic. Statistical analysis of the actual packet voice traffic indicates that not only the average talkspurt and silence periods are much longer than the values traditionally used, but also the traditional model of exponential distribution is statistically rejected in many cases. The traditional model further fails as a candidate for approximation of packet voice traffic. This paper also shows that audio contents can have a direct impact on traffic characteristics. The traffic data reported suggests that factors such as speakers, topics and situations of a voice call may cause traffic patterns too diverse to be described by a single model.