the overhead b/w (dimensioning: kbps) for voice data carried over IP can’t be calculated in general. Amongst others it depends on the codec used and on how many bytes of voice data you carry per packet (this number is not codec-specific, so it can vary even for the same codec).
The number of overhead bytes that are transported with each voice-carrying packet can be calculated if you know the header length of all protocols that are used. RTP header length is minimum 12 bytes, UDP header is 8 Bytes, and so on.
Either way, “16” is not a number that I would say is typical for voip overhead as far as I know.
I had a look for the ISDN standards in the last days but unfortunately I couldn’t find the right passages by the time that I could affort. In fact on copper-wire based ISDN BRA (and as far as I know even on PRA) subscriber lines that I know definetely no FDM is used to provide 2 independent full duplex B-channels. Each subscriber NT adds up his outgoing voice data on the wire and filters out (using echo cancellation) these data in real time to make sure the TE only sees the voice of the opposite subscriber. As far as I found it this method is not necessarily true for all countries, so there might be other solutions around the world.