Sounds like a cool idea and there may just be enough separated material to pull it off!
I like Frankenstein projects like this that uncover interesting stuff!
Suggestions?
I'd bring all the sources into Reaper DAW. Reaper has Elastique Pro time/pitch stretch algorithm which is genuinely lossless in veri-speed mode. (Classic linked speed/pitch like how an analog tape deck worked. And that's what this job specifically calls for.) You would be able to sync up all the different sources which will all have unique speed profiles from the different transfers of different analog generations. You'd probably be able to nail all this in just a couple or few hours and land with everything in perfect sync with zero loss or artifacts. This part is 98% of the difficult/critical work! And the part that would have literally been impossible to pull off before lossless linked veri-speed correction became available in recent times.
Then you'd be free to roll up your sleeves and get creative! Any further separation beyond what's on the surface of those mixes could very likely be pulled off with some targeted nulling with artifact free results. You would have a genuine separated (at least mostly) multitrack to work from. You could end up with a really genuine intentional surround mix! (As opposed to the "I guess this random result from the upmix is cool" approach.)
Very cool idea!
Thank you for the suggestions and encouragement !
I finished a 5.1 mix of Little umbrellas following what I tried to explain. It was a "proof of concept" and of course it can be improved, but to my very subjective standards 2 or 3 of the different mixes are a sonic improvement over the official stereo mixes.
Next I listened to the other tracks and reviewed the available material. Below my comments, but in general I observe that Little umbrellas has the most striking instrument separation front / rear, making it the ideal candidate for the treatment. The underlying concept for the album remix being along the lines used in several if not most studio surround releases, keeping keyboards on the back along any extras / overdubs.
Peaches in regalia: Here the separation front / rear is not as clear, the center of the backing track could be positioned in the center channel, its panned piano mixed in the rear channels.
Willie the pimp: the exception where I tend to prefer the 1987 remix. The bass is removed at the start entering at the 3rd beat of the 4th bar. Not sure what I would position front / rear, maybe I as in the others the 1987 mix at the rear channels. The highlight here would be to use the isolated vocals from the Hot Rats Sessions in the center channel. The available backing track is unedited and much longer, not much worth using it in my opinion. Instead I'd feed the LFE with a low pass of a mono mix. There are also 2 tracks with guitar overdubs available, intriguing how these could be used.
Son of Mr Green genes: there are enough differences in the overdubs of the 1969/1987 mixes to create a quite interesting surround mix. The rhythm track is over a minute longer, requiring some detective work if it merits being used in the center and LFE channels.
The gumbo variations: here it gets complicated, sonically not much difference between the 1969 and 1987 mixes, the 1987 remix being 4 minutes longer (which I never really cared for). There is no rhythm track, just the unedited 32 minute jam. A lot of work for not much audio enhancement.
It must be a camel: less differences than in Son of Mr Green genes but still interesting enough. The rhythm tracks could be used similarly as suggested in Peaches in regalia to compensate, but as it is split in two sections some detective /extra work is necessary.
In general I think that a somewhat "intentional" surround mix with the available material is entirely possible. Some tracks are relatively straightforward and potentially very rewarding (Little umbrellas, Peaches in regalia, son of Mr Greens). Willie the pimp has even more potential but requires much more work. It must be a camel less rewarding and some work. The gumbo variations offers the least potential and requires a fair amount work and patience.
I have worked with Reaper in the past and became quite proficient, but it is a use or lose thing and I find it very frustrating to have to re-learn everything. I will follow your suggestions and do some trials with the "low hanging fruit".
One quick question: what do you mean exactly by "some targeted nulling with artifact free results"? Any specific ideas?