Howdya' Make Them Mixes?
Well dear reader, you'll be learning the greatest long-kept secret of mine within this post, but first--let's get a brief idea of what's currently some of the most popular tech for making what's old (and mono) sound new--and in stereo: stem-separation, a rather sophisticated piece of tech usually involving models that have undergone a lot of training to properly identify and isolate specific parts of a tune.
One of the most noted usages of stem-separation tech is what Peter Jackson's team at WingNut Films created, with its first major usage being for Get Back. Much of the production audio WingNut had to work with for Get Back was mono, meaning that if, say, a bit of dialogue was buried underneath an instrument or some other noise, there traditionally wasn't much that could be done to properly separate the two. With WingNut's "M.A.L." technology, elements within these mono sources could suddenly be isolated from one another--individual voices, instruments, vocals and so on--to a degree that simply wasn't possible before. While certainly not the first tech of its kind, it gave many an idea of what the future could hold.
There were many sites aplenty for this sort of thing, but back in the day they were usually limited to much more specific separations (vocals & instrumental was what I remember them being best at, IIRC), so you certainly didn't have the same capabilities at home as what a huge company and dedicated team could pull off. Moving from 2021 to now, however, and we can practically do at home a good chunk of what technology like "M.A.L." demonstrated could be possible. WingNut's de-mixing technology itself has since been utilized for projects such as the 2022 Revolver remix--where elements from the original four-track recordings could be separated--and "Now And Then," where John Lennon's vocal could be isolated from the piano on his original demo.
MVSep is one such site that has helped create what I would dub a "homebrew mix" revolution--a spectacle, and an incredibly great leap compared to what we had about 5 or 6 years ago. Rather than being one particular separation model itself, MVSep gives users access to a variety of different models and algorithms for separating vocals, drums, bass, guitar, piano and plenty of other elements from a recording.
I do indeed work with MVSep within my mixing workflow, alongside Adobe Audition to mix, pan and EQ just about any track I can work with. It's strongest when the recording itself is of at least passable quality, so unfortunately, with some very minor things I've experimented with (crummy audience recordings of live shows, etc etc), I have had very little luck there currently. There's only so much you can pull out of a source when there wasn't much quality there to begin with.
Prior to Audition, I used to work with Audacity (sheesh, ya think I was maybe a little nutty for using that?) which, while it was fine in some ways, certainly couldn't handle everything I was wanting to accomplish with my mixes. What I generally strive for is a good combination of what was originally in mind when the track was mixed and what also works best to the modern ear. I tend to make sure I have a greater balance of both, unless I'm doing a mix in which I need to be very, very faithful to its original sound.
In my years of mixing, I have been very pleased with MVSep and the folk who work their asses off to make these models work as well as they do. I genuinely don't think I would've made nearly as much on YouTube (much less gotten any recognition from it) without the incredible folk involved and the tools they've helped make available to people like me.
I thank them greatly for what we currently have.