Yi Liang
The quality of real-time voice communication over best-effort networks is mainly determined by the delay and loss characteristics observed along the network path. High packet loss rates lead to unacceptable impairment of the perceived speech quality and excessive delay impedes interactivity. Significant playout buffering at the receiver is prohibitive and strongly delayed packets have to be discarded and considered as late loss. We propose to extend traditional single-path voice transmission to multi-path transmission, where multiple copies of the voice stream are sent over multiple independent network paths. Multi-stream adaptive playout scheduling is employed at the receiver to improve the tradeoffs among delay, loss rate, and voice quality. Path diversity gain is obtained from the uncorrelated statistical behavior of the independent channels, which results in improved quality of the communication. Experiments over the Internet show significant reductions in mean end-to-end latency and loss rates compared to FEC protected single-path transmission at the same rate. The experimental results also underline our claim that the statistical behavior of loss and delay jitter is largely uncorrelated for different paths, which can be exploited for improved real-time voice communication over the Internet.
Keywords: Real-time speech communication, multi-stream transmission, multi-path transmission, adaptive playout, multiple description coding, forward error correction, voice over IP.