Mpc stems out now

I want to seperate the stems after I chop the sample, so each pad in my program has stems in the sample layers. How can I do this?

HouseWithoutMouse wrote:Though it looks like they're not interested in fixing bugs and improving basic system features anymore.

This is patently false - I have no idea why people believe this let alone say it. I get that we're salty that there's still bugs and the Stem separation launch is clearly having growing pains, but there is NOTHING to suggest that they are not interested in fixing bugs and improving features.

Constantlyinclining wrote:I want to seperate the stems after I chop the sample, so each pad in my program has stems in the sample layers. How can I do this?

One way would be:

1. set slice points in sample (say, 16 regions)
2. process > stems, exporting to pad 1 in a blank program
3. In that program, copy pad 01 to the other 15 pads
4. program edit > samples , set 'Slice' on each pad accordingly (e.g. pad 2 = slice 2 etc)

Or if you only want to stem a specific pad, e.g. to remove a vocal on pad 6):

1. set slice regions
2. export slices to a program (either non destructively or extract samples).
3. select a pad, hit the 'process stems' icon at the top of the screen in program edit (make sure 'trim sample' is selected).
4. This will stem the audio on that pad only

Hopefully they'll eventually make the first idea an actual option on the stems processing screen so you don't have to manually assign stuff later.

With NVidia's H100 GPU, better open-source models will be trained.

Those improved models could be used by AIR Music.



But ... the models they're using might be reduced to make them more viable for the MPC.

Maybe they'll offer compatibility with a USB-powered AI accelerator for MPC Standalone.

I just tried Ultimate Vocal Remover GUI, and it is HUGE. It's "only" a collection of various algorithms and models developed elsewhere, but the amount of stuff that's available is pretty incredible, and it's all put together in a single user interface that understands what each algorithm and model is for, and the functions have been programmed into actual end-user features. In addition to vocal and instrument splitting, there are extra tools like separating lead vocals and backing vocals, and de-reverb/echo. Demucs is in there as one algorithm among others. And the buttons and options have actually helpful tooltips which explain what they do. For making a remix of a given tune, this allows lots of ways to try and get the best possible results for that particular song.

Re: Akai AI accelerator:

EnochLight wrote:

HouseWithoutMouse wrote:Though it looks like they're not interested in fixing bugs and improving basic system features anymore.

This is patently false - I have no idea why people believe this let alone say it. I get that we're salty that there's still bugs and the Stem separation launch is clearly having growing pains, but there is NOTHING to suggest that they are not interested in fixing bugs and improving features.

There are management and workplace culture issues.

It's a risk factor that is not going away.



But it's good enough. Just accept that we get something pretty awesome with some bugbears.



MPC Stems allows us to use the feature inside the same interface we're creating with.

It might help to clean the samples first. Maybe their model was trained with ideal samples that aren't noisy enough.

HouseWithoutMouse wrote:I just tried Ultimate Vocal Remover GUI, and it is HUGE. It's "only" a collection of various algorithms and models developed elsewhere, but the amount of stuff that's available is pretty incredible

If you click on the wrench icon, go to download center, hit refresh list, hit manual and then you can select from a ton of other models to download. You'll see there are currently almost 30 different demucs models to test out from.

The same goes for mdx and VR net as well. Check it out if you haven't done so already man. There really isn't anything out there that can compete with UVR as it has every separation model publicy available. Also, if a model doesn't show up in the list, you can manually download other models online and just unzip them into the model directory of the installation.

I started from cloning the repo on Linux, and initially it didn't have any models at all, so I first had to figure out how it all works and what the cryptic model names mean and what are some of the best ones.

MPC-Tutor wrote:Gave uvr a quick test last night, much better results than mpc, similar to ripx, perhaps slower but I’d need to spent more time with it as there’s a lot of options

It's a rabbit hole. I hope that you have a GPU to test with, otherwise you're in for punishment.

MPC-Tutor wrote:

Constantlyinclining wrote:I want to seperate the stems after I chop the sample, so each pad in my program has stems in the sample layers. How can I do this?

One way would be:

1. set slice points in sample (say, 16 regions)
2. process > stems, exporting to pad 1 in a blank program
3. In that program, copy pad 01 to the other 15 pads
4. program edit > samples , set 'Slice' on each pad accordingly (e.g. pad 2 = slice 2 etc)

Or if you only want to stem a specific pad, e.g. to remove a vocal on pad 6):

1. set slice regions
2. export slices to a program (either non destructively or extract samples).
3. select a pad, hit the 'process stems' icon at the top of the screen in program edit (make sure 'trim sample' is selected).
4. This will stem the audio on that pad only

Hopefully they'll eventually make the first idea an actual option on the stems processing screen so you don't have to manually assign stuff later.

Thank you so much bro

jaymack wrote:

jpeg wrote:perhaps its cos 3.0 is in the pipeline

Definitely. NAMM 2025. Everybody get ready for that 3.0 rollout! It's COMING!!!

Bolo said this year according to his insider info

richie wrote:
For all those that have not tried it out yet, download Ultimate Vocal Remover - You can actually mix and match different stemming algorithms to get impressive results. Far better than what these DAW implementations are in my testing.

I've blessed this post, I love you.

I've been using demucs-gui which let's me change algorithms as well but not mix and match them... this sounds like one level up from that.. downloading and will be trying..

Thanks!

richie wrote:Hey everyone,

How's your family doing? How are your kids doing? Tell your mother I said hello.

I'm curious of what algorithm Akai is going to use for stems in stand alone. As we're on an internet forum and all, Maybe some others can chime in here about the processing power of the stand alone units. *@ultros - since you dig around in Akai's OS updates, any insight to share?

Having had a lot of experience going through the rabbit hole of stemming stuff out, doing frequency analysis for many hours on end for weeks, using different and cpu intensive algorithms using the best thing in the world, Ultimate Vocal Remover, had it not been for me having a RTX 3080 GPU to stem out through, even my nice Ryzen 8 core 5800X was taking a while to stem full tracks.

As it stands now, the only way I can think of this working in stand alone mode is if you guys were say, stemming out small loops but it would take quite a long time to handle a full track unless they were using a shi*ttier algorithm or downgrade the quality of the open source algorithm everyone's using, consequently causing far more artifacts.

For all those that have not tried it out yet, download Ultimate Vocal Remover - You can actually mix and match different stemming algorithms to get impressive results. Far better than what these DAW implementations are in my testing.

I've blessed this post, I love you.

As a Force user I thank you Rich!

Unticking Rosetta mode doesn't solve the problem. I get the same message : unable initialize stems system

Need to be able to warp more than one stem layer…