Evaluating packet erasure recovery techniques for audio streaming

Perceived audio quality is an important metric in determining resilience of voice coding techniques against packet loss when streaming audio over Internet Protocol (IP) networks. For highly compressed audio, relatively small packet loss can dramatically reduce the quality of the reconstructed sound. A key strategy for decreasing the impact of packet loss on audio quality is the deployment of packet erasure coding which proactively sends the redundant packets used at the receiver to recover in part from lost data. In this paper, we present a testbed for voice quality monitoring with practical audio streaming applications. The paper provides architectural insights into the design and testing of erasure codes when combined with audio streaming applications. By adjusting packetization method and coding rates that maximize resistance to packet losses, we demonstrate that good voice quality can be obtained for a wide range of the raw packet loss rates in the IP networks for a variety of voice codes like G.711 and G.729.