The Design of A Multi-party VoIP Conferencing System

Multi-party VoIP;play-out scheduling;loss concealment;SVM;subjective perceptual quality

Huang, Zixia ; Wah ; Benjamin W.

关键词: Multi-party VoIP;    play-out scheduling;    loss concealment;    SVM;    subjective perceptual quality;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/11972/Huang_Zixia.pdf?sequence=2&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This thesis identifies the problems and trade-offs of a multi-party VoIP conferencing system implemented over the Internet and proposes approaches to solve these problems. The current Internet is unreliable, and it degrades the conversational quality of real-time multi-party conferencing. Delay disparities may cause unbalanced silence periods, and losses and jitters may affect the intelligibility of speech segments received. We collect real Internet traces from the PlanetLab and classify them into different categories according to the traffic behavior. After studying the conversational dynamics in the multi-party system, we identify user-observable metrics that affect the perception of conversational quality and study their trade-offs. Based on the dynamics and the Internet traces, we design the transmission topology to reduce delay variations and to avoid links with high losses and jitters. We propose loss concealment schemes for reducing the packet drop rate and play-out scheduling algorithms for equalizing silence periods and smooth jitters. We also discuss issues and solutions in a practical multi-party VoIP system design. We compare the performance of our system and that of Skype (Version 3.5.0.214) using repeatable experiments that simulate human participants and network conditions in a multi-party conferencing scenario. Our limited, subjective tests show that we can improve the perceptual quality when network connections are lossy and have large delay disparities. Because it is impossible to conduct subjective tests under all possible conditions, we have developed a classifier that learns to select the best equalization algorithm using learning examples derived from subjective tests under limited network and conversational conditions. Experimental results show that our classifier can consistently pick the best algorithm with the highest subjective conversational quality under unseen conditions.