Good to know! I’ve only used Spleeter a little for vocal samples and generally just mangle them.
So I’ve messed around with it for an hour or so. Definitely overkill if all you want to do is grab some audio an extract the stems. But you can do a lot of crazy shit with it if you want a pretty advanced tool. I loaded in a song I did and thats about 2.5 minutes and it took about 3 minutes to extract the stems. Its not an extremely complex song, just a drum line, vocal, lead synth and bass synth. I’d say it still has that “isolator” sound I’m used to with the artifacts and such but in general it sounded quite good to me, that being said I love distortion and aliasing and artifacts so its up my alley.
The thing thats really crazy is you can repitch sections in real time and even replace whole sections with different instruments so if you have a harp you can replace that sound with a guitar or some shit. I know technically you can do this with Ableton and converting to MIDI but this sounds a bit different. You can also easily turn a sound into a chord and it essentially lays out the audio as if its a midi note and you can edit in real time.
Really crazy sound design tool. I’m gonna mess around with it more during the 21 day trial I have but I may end up picking this up.
This is on the developers official youtube channel but its Benn Jordan trying some shit out and that dude is definitely into more left field and complex sounding stuff so its cool to see him mess with it.
I wonder if it works like shaperbox 3 and it’s envelope follower.
It can mask the transients with other sounds.
Thanks for checking it out.
EDIT. I may check out Serato Stems. It comes with DJ lite for free.
I’ve used Koala which is sortof ridiculous for the cost. That said I’m interested to throw a couple of things into lalal to see if that’s any better.
Let me know!
Just trying out StemRoller on the Bar-Kays tune that’s on the bi-weekly sampling extravaganza thread and my god I have to say it’s pretty damn good!
Alright, so this is based on a single (very awkward) sample to process. But I’ve just run a quick trial on Lalal and it was I would say maybe slightly tighter and more refined than Koala. It removed more artefacts in my sample that were left in on Koala. That said, this specific tune is a bit tricky because the vocals are drenched in reveb and sat deep in a mix. And so overall, the actual file is cleaner, but it’s certainly not night and day in terms of the end result, and how that would sound when put into a new mix. That said, what did look interesting on Lalal was the different types of split you can do. For example, I ran my sample (full band vox, guitars, bass, drums) through the synth splitter and it spit out a weird mix of the vocals and a pad sortof sound which I think was coming from a guitar reverb. So there is that to consider, but that will vary hugely depending on the tune, and I should imagine for most purposes Koala probably does the job.
Thank you for testing it out.
I was thinking the same thing. If I need stems, Koala can get me close, and I’ll just process and use as they are.
But I kinda changed my tune on splitting stems. I’m trying to use samples in a way that chopping is the only thing I do. Keeping drums in but EQing them out or keeping them in and layering on top of them.
Just like our ancestors did.
Stemroller seems very good and free.
I do love me a lazy chop. Stem extract can be useful, but I think the main thing with sampling is just how many ways there are to do it. You can extract the stems yourself, use a site like Tracklib, chop it oldschool - it kinda doesn’t matter.
Seratos stems is definitely the best I’ve heard so far in terms of quality, none of this stuff is perfect but theirs is definitely the best algorithm I’ve heard, I’ve been using it in serato DJ, and had brief try of it inside serato studio but I really didn’t like the serato studio software…
Thanks for the suggestions here. I did try running dumucs under Python but had issues with dependencies.
Just trying StemRoller on an M1 MBP…it’s the first time I’ve felt it get hot…I’ve still had to hold the laptop up to my ear to hear the fans though…damn that machine is quiet.
Yeah. My M1 was going full bore. Last time I tried using it.
By last time, I mean I’m not using it again. For it to stress the machine like that, means that it has poorly optimized code.
And I think @Symian is right about Serrato being pretty good.
Plus it comes with the free version of its DJ software if I’m not mistaken.
I think you’d need a serato controller to take advantage of it though, the free version of serato studio comes with stems too and I don’t think you need one for that… seems like you can use a midi keyboard etc…
I think I tried lalal once, but it didn’t work very well, at least enough for me to want to use again even for free.
I can’t think of anything “AI” related I’d be willing to pay for at the moment, I’ve tried text and audio and video upscaling.
Perhaps generative code if my company bought an account, but that I can tangibly understand and fix around a framework.
Ripx seemed good , used it to make my nirvana / whigfield track - smells like Saturday night.
And ‘phat doves’ … leftfield and prince.
Stem separation is also on roadmap for traktor - using izotope tech … though it’ll probably be part of their new subs service.
Oh my god. I just tried it on The Pet Shop Boys - West End Girls. I now have the vocals, bass, synths and drums separated!! They sound really great, not perfect obviously but I’ve never heard these individual elements before. The bass is surprising!
YOOOOOO…
tag me in whatever you use the stems for… Thats one of my favorite songs!
Same here, they really knew how to make a mood! I’m going to use this to come up with stuff for the best battles, it’s insanely useful!