Silence is golden?-the effects of silence deletion on the CPU load of an audio conference

This paper seeks to identify, improvements that reduce audioconference CPU load. A major contribution is the comparison of the performance benefits of five potential audioconference improvements: faster CPU, faster communication, better compression, digital signal processing (DSP) hardware, and silence deletion. To compare audioconference CPU load, we develop a model that identifies components of a typical audioconference. We hypothesize that silence deletion will improve the scalability of audio more than any of the above four improvements. We parameterize our model with measurements of the actual component performance. Overall, we find audioconference CPU loads with silence deletion scale better than audioconference CPU loads with any of the other four improvements. Techniques based on DSP hardware alone do not scale as well as silence deletion alone. However, DSP based silence deletion and compression together scale better than any other technique. These results hold even when using compression and even for ten times faster processors, networks and DSP hardware.<<ETX>>