Adventures in AI

[Frankencat]

First a little context: Higher Ground Band is my old originals band, we had a great following in the early 2000s, made several records and were even nominated for a Dove Award in 2010. Live At The Rock is a live album we released around 2009ish.

So I am working in my home studio this weekend on a project that started out as a simple stem mix of an old Higher Ground Band tune I want to remix for backing tracks. One thing led to another and now I find myself remixing the Live At The Rock CD. There is a really cool story behind how this originally came about and it was a great show we did with Mr Lou Angelwolf presiding and all of our friends and even family in attendance. I never thought much about it afterwards besides being happy with the show in general and then I got a call a couple of years later and learned that there was a multi-track recording of the show and "would I like it?". Well yeah! So that is is how the original HGB Live At The Rock CD came about. I worked hard on it and the recording was not the greatest. Lots of bleed and the sound in the room as I remember was atrocious. It was in a gymnasium all super echoey and whatnot. Anyway, I was never really happy with the sound of the finished product but the show itself is very cool so I let it go out for posterity and for the folks who were interested and followed the band. So now I found my original mixes and started going through them. It occurred to me that we have some really amazing AI powered tools for audio cleanup available now that hadn't even been imagined back when I was mixing this. I have been working with them and cleaning up some of the stems and remixing some of it and I have to say it is sounding waaaaay better and much more honest and raw, they way it really was in the moment. I don't know how far I am going to go, or of anyone else besides me is even interested anymore, I may just do it for myself to have something cool to listen to. Back then I had to make some decisions about the sound of things due to the techincal limitations at the time. When I listen to the CD I can hear all of that and even though I dig the set and the gig in general, the sound of it kind of bums me out. But now, using computer based tools including some of the AI powered tools that were developed for restoration etc, I am able to clean up a lot of the nasty bits very effectively and I am really pleased with the result. Anyway, I just wanted to share what I am up to. It's a journey, as always. Bless you all and long live Higher Ground Band!

 

[Frankencat]

BTW I used FADR Stem separator to remove the background noise and bleed on the vocal tracks which was a huge help. The software worked great and they have good support and are quick to answer questions, even on the weekend.

 

[FuzzyAce]

Cool man, sounds like a fun project!
I would love to go back and remix some older albums we did years ago. Never was totally happy either, far from it in many cases. Most were on ADAT though, and not even sure those original tapes exist anymore. That's a whole other story in itself....
But, even getting the ADATs would mean I'd need at least 1 machine to drop them into a DAW. It's a lot of work. Then I think, once I got them into the system I'd want to start redoing parts, etc. At that point, why not just rerecord? LOL

I've tried some separation software for gits and shiggles, although it's fun and exciting to try, there are just too many digital artifacts to deal with.
Using AI for that kind of thing, and for noise cleanup is really cool and will only get better though.

 

[Frankencat]

I recorded and produced a bunch of records for other people and I have a lot of real old master stuff on 1/2" tape that I dumped to DAT and then to digital a while back but those are locked in now. There are a few I would love to remix but unfortunately the multitracks are long gone, some ADAT and some on 4-track. Luckily when I got more serious I got into Pro Tools and made lots of copies and backups.

 

[Sascha Franck]

Stem separation is working pretty great these days in case you only want to do some optimizing.

For instance, Logic by now comes with its own stem separation. It's only splitting things into vocals (the sum of them), bass, drums and the "rest". But even that can take you quite far already. I just separated the stems of an IEM mix a bass player recorded on a recent gig, for pretty obvious reasons the bass was pretty loud, but I could nicely and easily tame it a bit. I was also able to very slightly correct some out of tune vocals and give the drums a bit more width, punch and clarity.

What all that doesn't do too well yet is to actually create stems that you could use for, say, remixes or rerecordings, simply because there's still some artefacts (obviously depending a lot on the source material). But as long as you mix them together and just slightly alter the individual stems, those artefacts will vanish.

Fwiw, there's some tools allowing you to further separate extracted stems. Steinberg's Spectral Layers by now has an option to further extract a drum stem track into individual drumset elements, which seems to work not too bad (I don't own it but may check the demo one day). And if you happen to own any version of Melodyne DNA, it'll perhaps allow you to finetune (or maybe even extract) some of the voices inside, say, a choire stem.

I think these are very valid reasons for AI usage (I absolutely *loathe* some of the other recent popular things...).

 

[ctreitzell]

Stem separation is working pretty great these days in case you only want to do some optimizing.

For instance, Logic by now comes with its own stem separation. It's only splitting things into vocals (the sum of them), bass, drums and the "rest". But even that can take you quite far already. I just separated the stems of an IEM mix a bass player recorded on a recent gig, for pretty obvious reasons the bass was pretty loud, but I could nicely and easily tame it a bit. I was also able to very slightly correct some out of tune vocals and give the drums a bit more width, punch and clarity.

What all that doesn't do too well yet is to actually create stems that you could use for, say, remixes or rerecordings, simply because there's still some artefacts (obviously depending a lot on the source material). But as long as you mix them together and just slightly alter the individual stems, those artefacts will vanish.

Fwiw, there's some tools allowing you to further separate extracted stems. Steinberg's Spectral Layers by now has an option to further extract a drum stem track into individual drumset elements, which seems to work not too bad (I don't own it but may check the demo one day). And if you happen to own any version of Melodyne DNA, it'll perhaps allow you to finetune (or maybe even extract) some of the voices inside, say, a choire stem.

I think these are very valid reasons for AI usage (I absolutely *loathe* some of the other recent popular things...).

I've been working with SpectraLayers extensively since the release of SL11. I'm talking full work weeks for months. The great majority of my work with SL has been NR for film audio (interviews and fx). For my purposes SL is simply awesome; there's no other tool as amazing, IMO/E.

But music stem separation is in its infancy and SL is pretty good at vocal separation. That said, the SL unmix modules are somewhat capable to completely unable. My only reasonably successful unmixing of music is getting drums and bass reasonably well separated, BUT only with utterly traditional sounds in a classic rock mix; five pc acoustic kit and a fender P Bass.

SL unmix modules will not be able to separate a string or horn section; and certainly not multi-layered guitars and/or synths without possibly months of manual work. Electronic instruments or synth instruments based on their real counterparts really will not separate.

SL music stem separation is going to allow for some re-balancing, albeit with a many, many hours of manual work. Getting clean stems is not currently possible in my experience.

I'm not saying I don't like SL, it is awesome. Anyone considering using SL seriously best get prepared to learn how to read the spectrogram well; it's deep waters.