Here's Dave Wise talking a little about how he did it:
Is there a specific timestamp we should be looking for in that video? I clicked in but it seems like it starts in the middle of a random conversation.
So would you have a composition tool that only played the compressed sounds? Or do you choose certain sounds to compress (like the strings shown in the previous post) and others not to?
I don't know exactly what program(s?) SNES composers would've been using, but they'd almost certainly have to be using the actual hardware for playback since there probably wouldn't have been another way to hear it as it would sound in-game. Every sound would definitely have to be compressed, take a look at this screenshot from the GSS guide Phantoon linked:
The yellow bit is the instrument samples, which comes in at a whopping 54 kB for all of them. For comparison, a minute of uncompressed CD-quality audio is about 5 MB for one channel, or 10 MB for stereo. One second of uncompressed mono audio is about 88 kB. The SNES audio system only had 64 kB of RAM to work with, so when you consider the amount of compression going on it's kind of a miracle any of it sounds like music at all!
Now that I think a little harder about it, it makes sense that the reverb on the original samples would've been lost when they got compressed that much -- they must've had to severely truncate the samples, which means any reverb tails would almost certainly be among the first information to go.