Audio quality is impaired by packet loss on the network and lack of real-time support in general-purpose operating systems. Acoustic aspects of packet network audio systems also produce problems, since one channel of restricted bandwidth is not the natural means of human audio communication. In this article, we describe state-of-the-art Internet audio communication and its applications, and outline impending technical developments and applications that will be available in the short-and medium-term. We describe first-generation multicast audio tools and how problems observed in their use have led to the conception and development of a Robust-Audio Tool (RAT). Subsequent user trials demonstrated a notable improvement in perceived audio quality. The origins of speech transmission over long-haul packet networks stem from the ARPANET and SATNET experiments, which spawned packet-based multimedia conferencing research on both sides of the Atlantic. Indeed, the Mbone is used by a number of research and teaching collaborations. The existence of an international infrastructure, and the use of generalpurpose computer workstations to access it, has created a critical mass of users who are piloting the technology. Thus, first-generation multimedia conferencing tools and the effect of real-time multiway conferences on the network have been studied with a range of diverse user groups: