The damn auto spell thing keeps changing it each time I edit. -0
Hey, about some advice here . . . no matter what this confusing piece of software is called. ;-) Just saying . . .

 

PulseWave, continued thanks for your contributions to this thread.

 

I plan on extracting complete drum performances from songs, then drop them into my DAW (probably Logic Pro) as a full track, probably mono (any thoughts on mono vs. stereo for a drum loop stem? From there I will teach "at risk" kids about building tracks, starting with bass, then guitar/piano, etc. I realize that I might need to clean up the drum loop stem via iZotope or other, then possibly enhance the drum loop with EQ and a transient shaper.

My goal is to extract the cleanest drum loop from Rock/Indie Rock/Punk Rock songs and go from there. I'll probably start with The Beatles . . . I'm not sure if the recordings from each era and/or genre makes too much of a difference when stem splitting . . . does it?

In my situation, I sold almost every instrument I have as I can no longer physically do much . . . and used some of the $ (and trade in) to buy a Mac mini M4. So, I believe I have the processing power to get the high quality and I don't mind if the processing takes a while . . . I have more time than money.

 

Last edited by a moderator: Aug 4, 2025

I believe those drum models spits out stereo, so check phasing issues and align them if needed before converting to mono

 

;-0 What would you recommend in this case? Keep it stereo or . . . ? Not sure how to deal with phasing . . . no experience.

 

you over complicating things...

as i said in my first post:
for voc and inst, use becruily models...
use bs roformer sw in mvsep on your inst model output, to separate the stems...
that's it, enjoy your layers

you can find becruily in mvsep too... you dont need uvr, you dont need ensemble mode...

why i suggest becruily? because its good all around... there is no perfect model or perfect way to extract layers.

 

Voxengo PHA-979 can be very helpful. https://www.voxengo.com/product/pha979/

 

I appreciate your suggestion, but I don't need an all-around model . . . do you recommend anything specific for drum loops? For what it's worth, it's not me overcomplicating things . . . it's UVR. ;-) I'd be happy to press DRUM SEP and be done, but the only apps that do it that are subpar. UVR reminds me of Reaper . . . it can do it all, but takes time to get a workflow that works for your needs.

Do you mostly use the paid version of mvsep or UVR? If you don't mind . . . why did you choose what you chose?

 

Clone, I appreciate you as always, but yikes . . . things are getting more complicated. I wish I could simply jack into my brain and be done with it all. Maybe someday . . .

 

It's a lot easier than it might look at first glance. Anyway, there are "Easier" to use tools for fixing phase alignment like Sound Radix Auto-Align,
but there is no shared version of that one for Mac since 2015. There is Melda auto-align plugin called MAutoAlign and some other ways to do it.

Just remember that collapsing the tracks with the phase alignment issues to mono is not going to fix it.

 

You dont need uvr, use mvsep. Its free, just make an account and use it... The difference between paid and free is just some ensembles, that is their exclusive... You can not have those ensembles in uvr ether, and they are not that important.

Just make account so you can download wav and flac...

For drums, you can just use bs roformer sw on song, but its better to use an inst model first and extract just instrumental, and then something like bs roformer sw to separate other layers(like drums)...Because this way is cleaner...

Forgot about uvr.

 

How about using UVR to make instrumental (with the best model that seems to be MDX23C-InstVOC HQ) and then mvsep to extract other things from that, thinking about trying it

 

I signed up for a free account, but it's 16bit. The paid gets users up to 24 or 32bit. That's the other issue I have with UVR . . . 44.1. I'm working in 24/48 and really hope to keep all audio at that quality.

 

I used mvsep to sep out the drums from a 60s rock song. Here is the sep and parameters? Any thoughts and/or suggestions of how to improve these results? I think I hear some bass guitar in there, but maybe it's just a tuned kick or something.

Could the same quality . . . same models be used in UVR?
This is 16bit, but I'm impressed . . . way better than other apps and online freebies I tried. I'm wondering if I should get the premium version or finally get set up with UVR.

 

Go premium, it's worth it.
As a side note, If you liked AI answers from Pulsewave you can use it on your own, sometimes AI can be very helpful!

 

Have you tried the new Spectralayers 12? It's my favorite today. MVSEP is also an excellent tool.
Mostly I combine both apps for best results.

 

The above drum loop sep is via (mvsep). This one is from Logic Pro.

 

How do you combine? Do you process in one and then reprocess in the other? Did you choose Spectralayers because you have Cubase or did you get it specifically for editing etc.?

 

Im very sorry, i forgot and didn't realize you have logic pro... The bs roformer sw model is logic pro model... People extract it to use it outside of logic...
So basically you have best 6stem model in your daw... No need for uvr o mvsep...
Its best single model that you can find for drums...
But that model can sometimes have artifact, because of vocal... Because of that if your track have vocal i suggest to separate instrumental first with inst model like becruily.

You asked why you can hear something extra in your mvsep one... You used a drum ensemble, its because one of those model, sometimes they leak... In ensemble modes, you can enable one option and the mvsep give you results of those models individually... So if one of those are not usable you can delete it and with others make your own ensemble...

Now how to create ensemble? One way of making it to just mix the model together, in your daw, nothing special special... Like they are playing together... For every model added i think you must lower the volume of the master 6db.
This is average algorithm in uvr ensemble, You can also use uvr to use min max algorithm... I dont suggest it.

Uou dont need mvsep premium...
All the models are 44.1

Spectralayers, its nothing special... And also so much heavy on my system...

You have logic and mvsep, its enough for your usage...

 

Whoa, Logic has it! Insane stuff didn't know