In the case of external interfaces, external cables act like antennas and there may be a nugget of merit in using fully isolated transformers to prevent common-mode noise from getting into the motherboard's ground plane through the input signal transformer's center tap which is there to prevent lines from floating to arbitrarily high positive or negative voltages.
I don't normally go here but I have to call total bs on that. Ethernet signals on twisted pair are differential signals that cancel the induced noise due to wire twist. Because one wire has to go high and the other goes low an interference signal on the pair doesn't cause a false input. The twist makes the two wires act like one so it just becomes common mode noise. In noisy situations where the induced interference is bad enough the answer is shielded twisted pair instead of UTP. In the worst of cases optically isolated repeaters are used on both ends that isolate the ground entirely. But those devices are not really for signal cleanup, but for transient removal and lightening effects on systems with multiple ground stakes and propagating wave effects. There are also lightening and transient arrestors that use tube technology, but again they don't clean up signal loss, just protect equipment. Any worse than that and you need fiber optics, which had already become economically viable before the repeater installation. But in real world we are talking close proximity arc welder type interference to even need the STP. When you go to STP the ideal is that the shield acts like an antenna and shunts it to ground. Again because the ground shunts the signal any float up is negated as long as you have a good ground. Any ripple on the ground just becomes common mode noise again which doesn't really do anything. If there is measurable voltage flux on ground reference you have a bad ground network. In my past life I not only installed, but fully verified cables in commercial/industrial applications. Cable verification required zero packet loss over gigabytes of data being sent/received and verified. Unless the cable itself was messed up being made, zero packet loss is required for cable certification. Any packet loss on a short ethernet run means you have a bad cable or port. No additional devices should be needed for all effects zero data loss on a wired lan network and something is wrong if you are getting measurable amounts of it at all. The internet or wireless are another discussion entirely, but wired ethernet lan in spec is rock solid. Even if a problem occurs due to some rough solar flair or local EMP blast (that doesn't somehow blow up your computer) the TCP stack takes care of it automatically and any UDP implementations worth their salt will at least run some sort of CRC and dump the packet.
Magnetic isolation transformers aren't a bad idea, but as the are almost always a hardware implementation and usually only in industrial and/or medical devices where some level of EMI Immunity has to be certified.
For all given points, in a normal residential application, that level of EMI is first off not going to be present. If you are worried about it use verified STP cables on your ethernet and make sure that at least one side has a grounded port. Even then in most cases it probably won't improve your audio stream considering the data is buffered on the playing device and any bad packets would have been discarded and resent long before the music is being played.