D-STAR will go into "R2D2", when the vocoder is struggling to get any intelligence it can from a highly damaged bitstream. As the BER increases from multipath and other propagation related events, and the internal forward correction (FEC) is unable to provide usable bits you will hear this.
On the other hand, some systems like DMR, cut off the bits to the vocoder at certain BER which makes for less garble, but more 'gaps'. Very short gaps can be overcome by bit stuffing previous samples into the stream - which sounds more like a long syllable, or they simply drop the audio altogether.
D-Star transfers AMBE audio over UDP port 40000 for remote linking. The port is open to the world by default. So sometimes due to random internet probes, AMBE audio from over the network goes "R2D2" before it even hits the transmitter. Realize that when you compress speech down to 2400 bps, this leaves really only the most important parts of the speech left. So if even a few mangled bits becomes a catastrophic thing. When you stream other HD audio over the internet in UDP a few bits lost or damaged isn't a big thing as it will still be intelligible if its even noticeable at all.
I don't know the specifics of how DMR signaling works over the internet*. I do know that since it really doesn't require one to open firewall ports to host a repeater, the voice port would only be open to the host that it is maintaining a statefull connection to.
*To-do: review the dmrlink IPSC client software project for a better idea how it works