Regarding voice call bandwidth usage post compresion –
All voice codecs like G 728, 729, 729A etc and compresion algorithms like CS ACELP use call state analysis to compress (and de compress) voice calls. Simply put, the software analysis the state of call before “truncating” useless and/or repetitive part of it before reconstructing the call at the other end.
Its important to note here that this logic works only when the call is active i.e. a human voice on both sides. Before a call has actually matured, i.e. while dtmf is being sent, silence, ringing etc, the compression algorithm dosen’t work to the fullest and compresses only a part of the bandwidth usage. This is normally between 13-14 kbps in G.729A CS ACELP. The bandwidth “dies” down to 8 (or 6.4 KBPS in some modes) after some milli seconds post ative connection.
Hence while doing BW calculation in IPLC + compression scenarios one needs to take about 2 odd % over and above standard calculations.
I hope that answers few questions. More? Write again!