Support and discussion for all of Akai’s modern standalone MPCs including the MPC X / X SE, MPC Live 1, 2 & 3, MPC One / One+, MPC Key 37/61.
By sparky Thu Jul 11, 2024 11:05 pm
I was actually just going to post earlier today whether anyone had a tip for a software version of the EMU transform/multiply function, I loved how it messed things up. Foolishly gave my ESI4000 away for a combi organ a few years back. Also the filters on that thing, and the random parameters - so good.

Re stems: I’m not surprised it wasn’t available as a trial. They’ll make plenty of money out of curious (again, read foolish) people like me who would never have used it if they’d heard the results before buying. I can see myself maybe building tracks out of random effected grabs off this where things are very mangled - but if you had a track in mind and wanted to isolate something for it… good luck!
User avatar
By Juxx Fri Jul 12, 2024 12:56 am
So Far it's great for me... :worthy:

I've never used stems on anything before...

So in my mind Im comparing it to how I i've been making "stems" for the last few years.... trying to Isolate Lows, MIDs and Highs of a Sample with Kill EQ and filters....10 Bucks Hell Jea :smoker:
By J.O.BEATS Fri Jul 12, 2024 5:04 am
LivePsy wrote:Y'all complaining. What do you guys expect it to do? I'd be disappointed if it DIDN'T munch the sound up. Because stems was removed from Acid 11, I'm glad to have a lo fi drums seperator.



Your definition of “lo-fi” is clearly different than mine. But that’s fine, if it works for you , awesome! Bang out those beats. Would love to hear what you make

For me, stem separation isn’t about getting textures or degrading the quality. I use it for when I want to isolate or remove a sound from a sample. In that case, I need it to be “clean” and transparent.

But I also don’t really need to use it a lot. And so when I need it it’s no problem to spend the extra couple of minutes and steps to use a higher quality tool.
By LivePsy Fri Jul 12, 2024 6:45 am
J.O.BEATS wrote:
LivePsy wrote:Y'all complaining. What do you guys expect it to do? I'd be disappointed if it DIDN'T munch the sound up. Because stems was removed from Acid 11, I'm glad to have a lo fi drums seperator.



Your definition of “lo-fi” is clearly different than mine. But that’s fine, if it works for you , awesome! Bang out those beats. Would love to hear what you make

For me, stem separation isn’t about getting textures or degrading the quality. I use it for when I want to isolate or remove a sound from a sample. In that case, I need it to be “clean” and transparent.

But I also don’t really need to use it a lot. And so when I need it it’s no problem to spend the extra couple of minutes and steps to use a higher quality tool.


AI can only do so much, it cannot magically separate the sounds back to their isolated condition. I've done some tests where removing the vocals left a horrible smeared and phased mess but then I've isolated drum hits and gotten technically terrible fidelity but amazing vibe hits.
User avatar
By Monotremata Fri Jul 12, 2024 3:37 pm
sparky wrote:I was actually just going to post earlier today whether anyone had a tip for a software version of the EMU transform/multiply function, I loved how it messed things up. Foolishly gave my ESI4000 away for a combi organ a few years back. Also the filters on that thing, and the random parameters - so good.


Yeah there is a reason I will never get rid of mine (I have two maxed out E6400 Ultras). I don't use them much anymore (I do pretty much everything in Cubase with Halion nowadays) but when it comes to Jungle/DnB basses, that thing has THE sound. I think the 'Z plane' filters are way too overhyped though. They were great in the Morpheus/Proteus 2000 but there's only 3 of them in the EIV and they're not really that usable heh.

The transform/multiply function is basically convolution, just a 25 year old version of it. You can do the same things with any convolution reverb or modeler, but you lose out on the artifacts and 'glitches' the old school software gave you.
By Legotron Fri Jul 12, 2024 4:14 pm
All of you who are complaining about the quality, do you understand how AI works?
For a machine like MPC´s, you need to train new model, because of the limitation of hardware. It will get better as it´s trained, but it can also be "overtrained, if not monitored, which leads to worse quality".
If they keep training it with quality source and monitor the model, you´ll be surprised what can be done with even small models in few months, sad thing is it requires lots of processing power and time.
By J.O.BEATS Fri Jul 12, 2024 7:26 pm
AI can only do so much, it cannot magically separate the sounds back to their isolated condition. I've done some tests where removing the vocals left a horrible smeared and phased mess but then I've isolated drum hits and gotten technically terrible fidelity but amazing vibe hits.
[/quote]

Disagree with this. There are several separation tools out there that are so good you wouldn’t be able to tell it was a separated stem. Night and day compared to the mpc even in controller mode
By Korzo Fri Jul 12, 2024 7:29 pm
Legotron wrote:All of you who are complaining about the quality, do you understand how AI works?
For a machine like MPC´s, you need to train new model, because of the limitation of hardware. It will get better as it´s trained, but it can also be "overtrained, if not monitored, which leads to worse quality".
If they keep training it with quality source and monitor the model, you´ll be surprised what can be done with even small models in few months, sad thing is it requires lots of processing power and time.

Are you suggesting that Akai is going to release new versions of the algorithm ? :lol:
They have a bunch of bugs that are not going to be solved..
The Stems standalone version is awful. It was just a marketing strategy.
User avatar
By Lampdog Fri Jul 12, 2024 8:27 pm
LivePsy wrote:AI can only do so much.

Cannot take out the egg from a baked cake.

People expect a perfect farm fresh egg.
You already know and accept the difference. Many don’t.
By HouseWithoutMouse Fri Jul 12, 2024 9:20 pm
The neural network in my head is able to separate instruments very cleanly. ;) So why couldn't an artificial neural network do the same. In time the separation fidelity will get better. It's already fairly good. It needs someone to donate lots of money for training better models, and someone to work on the algorithms.

If there's still someone who doesn't know, Ultimate Vocal Remover has incredibly good stem separation facilities, and more, such as de-reverbing, and you don't have to buy it and you don't have to install it on your MPC and you don't have to online-authorize it.
By LivePsy Fri Jul 12, 2024 9:34 pm
Legotron wrote:All of you who are complaining about the quality, do you understand how AI works?
For a machine like MPC´s, you need to train new model, because of the limitation of hardware. It will get better as it´s trained, but it can also be "overtrained, if not monitored, which leads to worse quality".
If they keep training it with quality source and monitor the model, you´ll be surprised what can be done with even small models in few months, sad thing is it requires lots of processing power and time.


Its not large language learning models and not true AI, the algorithim is programmed to recognize a bass or percussion sound and remove what it thinks does not match that. It has to assume a whole lot of things. I split a CR78 beat and it only put kicks on Percussion and splatty hats and congas in Other.
User avatar
By Ultros Fri Jul 12, 2024 9:36 pm
Lampdog wrote:
LivePsy wrote:AI can only do so much.

Cannot take out the egg from a baked cake.

People expect a perfect farm fresh egg.
You already know and accept the difference. Many don’t.


Its easiest to liken it to color removal (color to alpha) or isolation in graphic imagery. the pixels only clean up so nice and have that jaggidy edge glyphed out artifacty look. you can soften it and filter as much as ya want but in the end you can only extrapolate so much information from the digital information supplied.

A lot of its ability to extrapolate signal will hinge on the fidelity of audio you feed it. You cant extrapolate high quality signal from a low-resolution format or compressed format. you will extrapolate a compressed signal into a pcm file. We cant add signal where signal does not exist so feed it shit in THE HIGHEST RESOLUTION PCM you have. 32bit 192k or whatever, the name of the game is FREQUENCY and DYNAMIC RANGE.

Extrapolating high resolution from a low resolution source is called "Up-scaling" and mpc / the software doesnt do that anywhere. Another loss for the youtube biters.

As an example try this, render 2 versions of a sample one to pcm and one to mp3. save both then reload both into your sample pool, then run the stem seperation on both versions and listen to the resulting stems.
By Legotron Sat Jul 13, 2024 11:33 am
LivePsy wrote:
Legotron wrote:All of you who are complaining about the quality, do you understand how AI works?
For a machine like MPC´s, you need to train new model, because of the limitation of hardware. It will get better as it´s trained, but it can also be "overtrained, if not monitored, which leads to worse quality".
If they keep training it with quality source and monitor the model, you´ll be surprised what can be done with even small models in few months, sad thing is it requires lots of processing power and time.


Its not large language learning models and not true AI, the algorithim is programmed to recognize a bass or percussion sound and remove what it thinks does not match that. It has to assume a whole lot of things. I split a CR78 beat and it only put kicks on Percussion and splatty hats and congas in Other.

It states their site that it uses neural network for processing, so neural network always needs an model