Media Management Research Lab

National University of Singapore

  • Increase font size
  • Default font size
  • Decrease font size

Project ACTIVE

Print PDF

There is evidence from learning and psychology research that indicates memory is enhanced by spatial associations. Such spatial information is lost with current telephone-based teleconferencing, which makes it difficult to hold audio teleconferences with large numbers of people. Since classroom discussions can involve large numbers of students,it is currently difficult to make practical use of audio teleconferencing in courses with large enrolments.

We propose to address this problem by developing a prototype Multiuser Audio Chat System with the features of rendering the source of users initially to several predetermined spatially arranged positions, as well as generating and archiving a searchable transcript of each session. This will make it possible for remote learners to participate more actively in on-line discussions, and give them more of a sense of being present in the discussion session, just as graphics-based virtual reality technology gives users the illusion of being present in a virtual scene. This same technology can provide students with the ability to "go back in time" to query and recall earlier discussion sessions.

The multiuser audio chat system involves numerous technical challenges that need to be addressed to build such an application. The number of participants in a chat session may be several dozens, with each student needing to hear and possibly talk to any other person in the session. Hence, one of the primary concerns is an efficient interconnection topology and architecture. An immediate first approach would be to connect each participant to a central server that merges incoming audio streams and then distributes the final mixed result to every listener that is connected. The advantage of such a star layout is that the sessions can be centrally managed and the delay for the sound streams to be relayed by the server depends mostly on the distance of the users from the server. The disadvantages of this architecture are that the central server requires a large amount of resources (for example memory and network bandwidth) that is proportional to the number of participants. Furthermore, the server can easily become a bottleneck and also is a single point of failure. Therefore, we plan to develop and implement a more distributed peer architecture where a newly joining user may be connecting to one of her peers who is already participating in an ongoing audio chat session. We envision that some central control is still necessary to manage the session (for example for floor control, i.e., who should be allowed to speak and at what time). However, the network resources that are required at the server side for this architecture will be greatly reduced.

One of the challenges with a distributed architecture is that the end-to-end audio latency may be more variable. From existing research we know that for an interactive conversation the delay from the microphone input through the transmission to the audio speaker output should not exceed 100 to 200 milliseconds for a natural conversation. If these limits are surpassed then the delay becomes distracting. The audio chat system will automatically select a peer connection that will enable audio transmissions to take place within these given limits.

People Involved
Roger Zimmermann
Leslie S. Liu
Beomjoo Seo
Min Qin
Last Updated on Wednesday, 20 January 2010 17:39