We have developed a mobile audio space that supports multiple, simultaneous conversations.  In face-to-face interactions in gelled social groups, conversational floors change frequently, e.g., two participants split off to form a new conversational floor, a participant moves from one conversational floor to another, etc. To date, audio spaces have provided little support for such dynamic regroupings of participants, either requiring that the participants explicitly specify with whom they wish to talk or simply presenting all participants as though they are in a single floor. By contrast, our new audio space monitors participant behavior to identify conversational floors as they emerge. The system dynamically modifies the audio delivered to each participant to enhance the salience of the participants with whom they are currently conversing.

In face-to-face interaction or in a traditional audio space, participants may use selective listening to attend to different participants (Figure 1a).  In “The Mad Hatter’s Cocktail Party,” the speakers with whom you are not interacting seem to move far away from you, i.e., their audio attenuates. Such attenuation makes it much easier to understand the speaker of interest while maintaining the ability to monitor other speakers (Figure 1b).This happens automatically, unlike in other systems which support attenuation using explicit interface gestures.

Figure 1a, Cocktail party effect: Participant A uses selective listening to attend to the desired floor.
Figure 1b, Mad Hatter's cocktail party: As floors reconfigure, the speakers not participating in A’s floor appear to distance themselves spontaneously to assist her in floor separation. (The reference is to “A Mad Tea-Party” in Lewis Carroll’s Alice’s Adventures in Wonderland, in which the characters continually change places for the benefit of a single participant. Figure 1 is based on Tenniel’s illustrations, obtained from Project Gutenberg.)

The system includes a learning algorithm that uses the following principles from conversation analysis to identify which participants are speaking to each other: (1) if Participant A starts speaking when Participant B stops speaking, they are likely to be in conversation with each other; and (2) if Participant A and Participant B are speaking at the same time, they are unlikely to be in conversation with each other.  When the system has identified which participants are in conversation with each other, it performs audio adjustments to make it easier for people to hear the participants they are conversing with.

We have conducted a user study, comparing a traditional audio space with our system.  Conversation analysis revealed that in the traditional audio space environment with a single, static volume level, the participants frequently positioned utterances relative to multiple floors, a highly unnatural behavior. In our system in which volume levels were adjusted to deemphasize speakers from other floors, participants exhibited natural turn-taking practices within each floor.

For more details, please see our CHI 2003 paper.

      Modified: $Date: 2005/01/28 02:15:37 $ audiospace, audiospaces