For line level audio is OK to have a ~10k resistor in series at the input, then a back to back zenner from each input to ground would do the trick. Using 15V 0.5V zenners would take overload way bigger than you may expect in the real world, only limited by the resistors power handling. Using 1W resistors (or two 0.5W 1%MF) you would be able to handle up to ±100V on each input, RMS and long enough to realize, it might degrade the resistance value taken to that limit which would have an impact on CMRR for both, temporary (not important) and permanent if the over voltage is presented for a long enough time.
The problem with that approach is, of corse, you already have the nicely laser trimmed resistors inside your IC and you can't access the needed end of them, CMRR could be trimmed to a decent level, provided a proper and stable trimming solution. Using this ICs solutions takes away the flexibility to handle many things, as protection in this case. Still it can handle over 100W 4? power amp connected directly. Personally I'm not a lover of this ICs line driver/receivers but they do provide a solutions for reasonable cases. Usually really good for DIY line management or specific products as studio tools and such.
When you need to go out into the wild you need serious protection so the reliability isn't compromised, big live desks isn't a place for this to go to die for example, at least not as in the data-sheet examples. Still there are work arounds, for 12? miss match in the input resistors you are looking 64dB CMRR (I'm assuming 0.5 gain config) which is something you should totally be able to accept, as the source might have an even bigger miss match, even 100? is not rare. Then you can have two polyswitches in there ±12? between each one and a diode configuration that can handle the expected input transient depending on the selected poly. With that approach you may be able to handle hundreds of watts while keeping a decent CMRR. Distortion shouldn't be a problem I guess, the polys should be stable enough under normal signal levels.
If you don't care about the actual gain the amp has you could use 10k 0.1% input resistors and the diodes or something like that. 18k in series with the 6k would give the same 0.5 gain, you could replace the 0.1% resistors with matched 1%, your choice, would depend on the scale of the project.
Even the transformer approach isn't perfect, it has many advantages but they probably can't handle a high power amp because two things, burning the primary and depending on the particular circuit they might also feed a harmful over voltage to the next stage.
JS