Number of floors for a voice-only conference on packet networks – a conjecture

Abstract

Voice conferencing is an essential block of any multimedia system used for collaborative work, as voice is shared by all participants. Floor control is mission-critical here and has been investigated by many to ensure fair resource sharing; yet fixing the number of floors has remained an open problem. A conferee (participant in a conference) can speak only after acquiring the floor. To allow impromptu speech, floor allocation must be made for many concurrent speakers. However, too many concurrent speakers degrade voice intelligibility. Therefore, setting an upper bound for the number of streams (floors) that may be mixed is sine qua non for quality conferencing. The problem of setting an upper bound on the number of floors to support concurrent multi-party audio sessions is addressed. A conjecture based on conversational and qualitative analysis is proposed. A pseudo-measure termed 'loudness number' used to manage the number of floors is briefly outlined. The implementation at a functional level on Windowsr systems has yielded satisfactory performance.

Fixing Number of Floors for Virtual Voice-Only Conference - an Empirical Study

“…Therefore it is important to find an optimal number of floors, which we call N max , that can provide interactivity and spatialism, as well as keeping it to a minimum compared to complete mixing architectures [17], [16]. In this paper we present the results of an experiment to imply how to find this value, and briefly its impact on the design of application architectures taking off from previous studies [14].…”

“…• With multiple speakers allowed to address the conference, it is but imperative that streams from only a small set of selected speakers be broadcast/multicast to all participants as in a many-to-all fashion. Mixing arbitrarily many streams makes the resultant speech unintelligible [14]. • Giving users the ability to determine the weights for individual streams before mixing at their terminals results in increasing quality of mixed speech stream since, they can boost the volume of speaker(s) whom they want hear loudly.…”

“…Some work in this regard from ethnomethodological perspective has been done earlier [13], [14]. We take a completely different approach here when we try to fix N max by analyzing a few conferences that lacks visual aid.…”

“…This needs to be done dynamically, smoothly and in realtime without participants being able to notice it. In [14] this issue was settled by employing a pseudo-metric termed Loudness Number (LN). It quantified the following attributes of speakers: 1) current loudness of a speech, 2) loudness of a speech in the recent past duration, 3) level of activity in a reasonably past duration.…”

“…Since speech streams cross only two CSs, the delay is limited to transit time and queuing time at two nodes. The decision to select speakers is based on LN [14] at every packet instant (at most every 40 miliseconds). Thus it is dynamic enough to invoke the speaker change on the fly.…”