Topic: A.I. audio stem software

[checht]

Hiya Pat,

Except for the Mac Pro, Macs don't use graphics. cards these days.

I'm on the road w Los Lobos, so using Macboook Air and just going with no ensemble, just MXR Main. 1.25 hour show takes around 45 min to process. In this application, I'm recording omt4 with mk41s x/y PAS in the center, and mk22s 4' spread. I'm not too picky about extraction in this case, as I'm pulling the vocals from the 41s to mix back in in parallel with the 2 stereo tracks, a variety of vocal sweetening.

Here's a track from Thursday night:
https://archive.org/details/LosLobos2025-01-06/ll20250206.OMT4.2448-02.flac

This is my current state of the art, combining AI extraction with omt4 recording. Fun times.

Not very good picture attached showing omt4 boom at the triple door in Seattle.

[robgronotte]

SpectraLayers 12 was recently released.  It's supposed to have better stem separation and definitely has more individual stems it can create.  A small test with the Unmix Crowd Noise module showed no improvement for me, sadly, as that is all I've really used SL11 for.

Has anyone managed to successfully use the Unmix Multiple Voices module?  I've never been able to figure it out with any success.

[rastasean]

Here's another project you folks may like: https://github.com/CarlGao4/Demucs-Gui

You're able to separate out all kinds of stems and even make your own mix.

[phil_er_up]

Has anyone managed to successfully use the Unmix Multiple Voices module?  I've never been able to figure it out with any success.

Create vocal stem and open it in SL.

Select "unmix Multiple Voices" in the modules.

Use cursor and Highlight a 20 second piece or so for main vocalist in the vocal stem
Then in the  "unmix Multiple Voices" window Click "Register Voice" - In SL it creates "voice 1"  in the "unmix Multiple Voices" window.

Now you can select the second vocalist and do the same as above and the then this will create a "Voice 2" for the second vocalist.

Now click "apply" in the  "unmix Multiple Voices" window and SL creates layers for "Voice 1", "Voice 2", "Non_voice" and  "Non-Un-mixed".

File > Export > Layers

[hoserama]

Here's another project you folks may like: https://github.com/CarlGao4/Demucs-Gui

You're able to separate out all kinds of stems and even make your own mix.

Ultimate Vocal Remover can do everything that does, and more.

[hoserama]

As an aside, Apple's new models included in the latest version of Logic Pro are easily the best I've played with. It does a really good bass/drums/vocals/guitar/piano/other. In particular, the guitar and piano stems are far better than any others I've used.

[al w.]

As an aside, Apple's new models included in the latest version of Logic Pro are easily the best I've played with. It does a really good bass/drums/vocals/guitar/piano/other. In particular, the guitar and piano stems are far better than any others I've used.

+1, the new version is superb!

[robgronotte]

Here's another project you folks may like: https://github.com/CarlGao4/Demucs-Gui

You're able to separate out all kinds of stems and even make your own mix.

Ultimate Vocal Remover can do everything that does, and more.

What model / settings do you usually use in UVR5?

The one I've generally gotten the best results with is MDX-NET MDX23C-InstVoc HQ, but it's about two years old now.  I can't find any more recent recommendations anywhere.

[phil_er_up]

As an aside, Apple's new models included in the latest version of Logic Pro are easily the best I've played with. It does a really good bass/drums/vocals/guitar/piano/other. In particular, the guitar and piano stems are far better than any others I've used.

What Apple new models are you referring too?
Do you have a link?

[ballerusk]

https://support.apple.com/no-no/guide/logicpro/lgcp61bae908/mac

Stem Splitter is available on Macs with an M1 or later Apple silicon processor.

[hoserama]

Here's another project you folks may like: https://github.com/CarlGao4/Demucs-Gui

You're able to separate out all kinds of stems and even make your own mix.

Ultimate Vocal Remover can do everything that does, and more.

What model / settings do you usually use in UVR5?

The one I've generally gotten the best results with is MDX-NET MDX23C-InstVoc HQ, but it's about two years old now.  I can't find any more recent recommendations anywhere.

I was typically using the htdemucs_ft models, and would sometimes run it along side the mbf_kim_vocals (I forget exactly the name). Now I'm using the Logic models as my mains.

[hoserama]

As an aside, Apple's new models included in the latest version of Logic Pro are easily the best I've played with. It does a really good bass/drums/vocals/guitar/piano/other. In particular, the guitar and piano stems are far better than any others I've used.

What Apple new models are you referring too?
Do you have a link?

Right now its only thru the Apple Logic Pro program. Friend of mine was able to extract it and testing out a GUI for that in windows, but still very much in the beta form.

[guitard]

This guy was on Eddie Trunk recently.  Pretty cool stuff he's doing with stems.

https://www.youtube.com/@franKENstein-Creations

[ballerusk]

Has anybody seen any tutorials (videos or written) on mixing/mastering audience recordings? Everything I have found has been on soundboard or studio recordings where they have each instrument/element separated.

I played around with stems from UVR but it seems ..."wrong" working on something I feel is not 100% clean (some bleeding between stems, stems not getting all vocals, drums, etc) and that it would be better to mix/master on the recording as a "whole".

[robgronotte]

I don't know of anything like that, but I have posted fair amount about my process here in this thread and maybe another in this sub-forum. I'm happy to answer any questions as well.