You can lower the sample rate of sound assets that you know don’t need the spectral bandwidth, for sure. The Engine does SRC on the fly because target devices can have all kinds of sample rates. You can also adjust the “Quality” level on the asset itself, which adjusts the compression scheme when the Asset is cooked.
Sound Groups basically allow you to adjust the file length thresholds which determine whether the Asset will be loaded fully in to Memory or will decompress in real-time.
It will always be a balance between CPU load vs. Mem load.
With that in mind, there is a console command AudioMemReport which allows you to log the current Audio Memory status–I believe it saves it as a CSV, so you should be able to view it in your SpreadSheet tool of choice.
In general, we have an initiative to look at ways we can improve profiling and performance reporting–this is definitely something that’s important to us.