Last N

Abstract

Multiparty conferencing has traditionally been a relatively expensive application that was only used in enterprise scenarios. Recently, however, the landscape has started to shift in ways that could change this. Ever-increasing bandwidth and processing capabilities make it possible for mobile endpoints and laptop computers to easily handle multiple incoming media streams (both audio and video). The development of Web Real-Time Communications (WebRTC) has also significantly simplified the development of video conferencing applications and made them mainstream. Both of these changes provide a way of replacing expensive video mixers (that produce composited videos) with light-weight video routers (that selectively forward streams). In this paper, we describe a Multipoint Control Unit (MCU) that identifies and selects the last N dominant speakers and forwards their streams to all the conference participants. We evaluate the performance of this Selective Forwarding Unit (SFU) against a simplistic everyone-to-everyone (full-star) MCU. Our results show that the SFU uses 45% less CPU and 63% less bandwidth when forwarding media for 10 of the endpoints in a 30-participant conference.

The development of healthcare mobile robot for helping medical personnel in dealing with COVID-19 patients

“…In addition, a server to support internal video conferencing was developed using Jitsi Meet. Jitsi Meet is an open source platform for video conferencing from www.jitsi.org [34], [35].…”

Dynamic video bitrate adaptation for WebRTC-based remote teaching applications

“…In this case, the amount of forwarded streams should be selected to avoid wasting bandwidth. Grozev et al develop a speaker identification algorithm to be deployed on an SFU, to identify the last N dominant speakers of the conference [10]. Only these N streams are forwarded to the conference participants.…”