SDK: silence detection without DSPs

The Diva Server PRI-0 cannot do human voice detection/silence detection on board, because this is a function provided by the DSPs. 

However it is possible to work around this in some cases by processing the audio in the host (provided of course that there is enough CPU to process the audio of all the concurrent calls).

The principle is to stream all of the input audio into the application, and then the application itself can run thorough the sound samples itself, noting along the way the minimum and maximum audio levels it sees in all the inbound audio.  The attached file shows an example of how to do this.

The idea can be taken even further, as it is possible on the main CPU to run more sophisticated algorithms such as Goertzl (to recover DTMF and other common tones), or other DFTs to identify the frequencies present in the audio stream.  This of course takes more CPU than simply scanning for minima and maxima, but still this approach might be desirable in some cases where the PRI-0 fits the right price point.

 For more info on Goertzl see, http://www.embedded.com/story/OEG20020819S0057, or Google for "goertzl".