In real-time interactive Internet multimedia conferencing, audio quality can be impaired by packet loss resulting from network congestion, lack of real-time process support in end-host operating systems, and acoustic problems. Furthermore, when audio is combined with other media, such as video, the lack of coordination in the presentation between the individual media streams is problematic. The thesis is that the quality of Internet multicast audio conferencing can be improved by a range of application techniques that focus on a combination of the system, the network and the user. A new architecture for a multicast conferencing audio-tool is presented and implemented in the Robust-Audio Tool (RAT). This software is used as a platform for most of the research in this thesis. Within RAT, a novel forward error correction technique is developed which successfully addresses the network packet loss problem with minimal network load and delay overheads. To address the operating system scheduling problems, a new adaptive scheme that analyses the real-time performance of the workstation and works around anomalies is developed and evaluated. The feasibility of synchronised presentation of audio and video in a multimedia conference to solve the lip-synchronisation problem is investigated. To this end the Mbone video conferencing program vie is modified to work together with RAT in stream presentation. The success of the approach is shown through a series of human subjective perception experiments. Finally, the problem of audio delivery to participants of multicast conferences with heterogeneous network connectivity is addressed. The proposed protocol is run between the audio applications in end hosts that cooperate to work around congested network links and requires no modification to the network infrastructure. The effectiveness of the approach is shown through simulation.