Personally all of my favorite PS1 soundtracks, mostly coming from RPGs, were not redbook. I can only think of one case that had a predominantly streamed soundtrack, but it was still compressed XA audio.
Redbook worked okay sometimes but had a lot of downsides. It wasn't space efficient, so you'd get at best 70 minutes but probably a lot worse depending on how much you need to use the space on other things. It also doesn't loop well, and has to be interrupted when you need to load other data (unless you have enough buffered)
If you were to use a Saturn as a synth tho, you could just as well dedicate main/all RAM to storing compressed samples, decompress on the SH2 CPUs in realtime and then dynamically swap data in and out of sound RAM.Complicated, sure. And why bother today...? Still, interesting idea, I thought.
Maybe you could store compressed samples in sound RAM and have the 68k decompress small chunks at a time. The 68k is pretty slow so there's a big performance question. Here's one decoder that says it does 50k/samples a second on an 8MHz 68k:
https://github.com/Kalmalyzer/adpcm-68k
Saturn's 68k is 11.3MHz so a little faster, but you do need some CPU time remaining to drive the sequencing (not a lot). ~60k/samples per second is perhaps enough for 2-3 compressed voices. Not great but better than nothing.
