Kemper Profiler MK 2

Thanks again !

I have question I truly don't understand.

I know that for example, Fractal and Helix have very good "low aliasing" but from my reading and seeing some examples, NN / ML formats like Tonex and NAM suffer from [relatively] much higher oversampling.

Is there a specific reason for this ?

Cant oversampling or some other "process" be used to reduce the aliasing in these NN / ML formats ?

It's very possible, but it needs to be approached in a different way.

The reason it's alot easier in component based model such as how Helix works, in very basic terms, is because you can filter at a component level, in the circuits you are designing. It can be added where ever you need it to be.

In a black box model, it needs to be approached in a different manner, which is not so straight forward.
It's mostly then going to be about how the software which loads the model needs to be interacting with the models, adding in necessary filtering in a more appropriate way - one that is not attached to individual circuit parameters.
 
It's very possible, but it needs to be approached in a different way.

The reason it's alot easier in component based model such as how Helix works, in very basic terms, is because you can filter at a component level, in the circuits you are designing. It can be added where ever you need it to be.

In a black box model, it needs to be approached in a different manner, which is not so straight forward.
It's mostly then going to be about how the software which loads the model needs to be interacting with the models, adding in necessary filtering in a more appropriate way - one that is not attached to individual circuit parameters.

Flying blind here - but because the NN / ML has no way of knowing which individual components are producing aliasing and what amount of aliasing each component is introducing, that sounds like its going to be a -very much- harder thing to do in NN / ML processes as opposed to doing it at an individual component level - possibly even not doable in a good or accurate way with NN/ML processes - if I'm right ?

Again I could be %100 totally wrong / off base.
 
Last edited:
Here's a quick comparison I made to illustrate how oversampling can reduce aliasing with a component based plugin modeling a tubescreamer.

The project is 48 kHz. I use Cytomic's The Scream (great plugin!), which lets me set the internal oversampling manually. So the only thing changed between the two cases shown, is that the oversampling in the plugin is manually changed from 2x to 8x.

2 x oversampling:

Scream_2xOS.jpg


8 x oversampling:

Scream_8xOS.jpg
 
Here's a quick comparison I made to illustrate how oversampling can reduce aliasing with a component based plugin modeling a tubescreamer.

The project is 48 kHz. I use Cytomic's The Scream (great plugin!), which lets me set the internal oversampling manually. So the only thing changed between the two cases shown, is that the oversampling in the plugin is manually changed from 2x to 8x.

2 x oversampling:

View attachment 45492

8 x oversampling:

View attachment 45493
Nice work.

So what this tells us is that:

1 - No, running the entire thing at 48kHz is not enough to clear up the aliasing. At 48kHz that puts Nyquist at 24kHz. But non-linear distortion will generate frequencies much higher than that.

Theorhetically speaking, non-linear distortion generates infinite harmonics. These extend far past nyquist, and this will happen at any core sampling rate. If you don't handle these frequencies, then they bounce back off the Nyquist frequencies, and change direction - this is why it is called foldback distortion.

A 30kHz harmonic at 48kHz sample rate will reflect back to 18kHz; and this is true for any frequency. It isn't a random process. There is a true mathematical nature to where a frequency will fold back, and sine tones reveal the nature of the system very well. But all of this is equally applicable to any input signal.

2 - Oversampling at 8 times massively cleans it up. BUT IT CLEARLY ISN'T JUST OVERSAMPLING THE SIGNAL. You can clearly see the presence of an anti-aliasing filter in the chart; a tell tale sign of this is where the line goes up, reflects down a tiny amount, and then disappears.

Why does oversampling work? It pushes the Nyquist frequency much higher. So 48kHz * 8 = 384kHz, which gives the system space to represent those frequencies that the distortion generates; even though we cannot hear them. But this alone won't cure the problem, because at some point the signal needs to be downsampled. You're not running your DAW at 384kHz; you're running it at 48kHz. So the plugin needs to downsample the audio to match what the DAW is expecting. Which is an opportunity for all of those harmonics to reflect back; even though you've given them plenty of space earlier on.

This is where the anti-aliasing filter comes in. The fact that frequencies are filtered out when the signal descends in the graph, tells us that the anti-aliasing filter is being applied during the downsampling phase of the plugin.

An anti-aliasing filter is easiest to understand as a brick wall filter, that blocks frequencies above the cutoff. There's more to it than that, because you also need to choose a design that isn't going to destroy the phase of your signal, while not blowing up your DSP/CPU usage; because this filter is going to be running on every single sample that gets processed. Can be quite a heavy operation.

PS: I used to work alongside Andy at Cytomic, right when I got into the industry. We were both at FXpansion. Great guy! Super genius at DSP. Makes great plugins.
 
Thanks again !

I have question I truly don't understand.

I know that for example, Fractal and Helix have very good "low aliasing" but from my reading and seeing some examples, NN / ML formats like Tonex and NAM suffer from [relatively] much higher oversampling.

Is there a specific reason for this ?

Cant oversampling or some other "process" be used to reduce the aliasing in these NN / ML formats ?
Hmm, this isn't an area I am too familiar with. But I believe it comes down to:

- The 'model' gets trained and validated on data at a specific sample-rate.
- This means the model can generate data designed for the same sample-rate.
- User comes along and wants to increase the base sample-rate in their DAW.
- Now none of the training or validation data is relevant.
- Model doesn't know how to interpolate for the new sample-rate.

And actually, this would happen even when going from a high to a low sample-rate - so for example, if the model was trained on 96kHz data, and the user was running it at 44.1kHz.

If you imagine this entire thing as a pedalboard....

In component modelling land, you've got:

Pedal 1 > Pedal 2 > Pedal 3

Then dealing with aliasing becomes like putting other pedals in between those pedals.

But in NN/ML land, you've only got 1 pedal, and putting other pedals either side of it won't fix the issue. You actually need to change the design of the pedal.

That's quite a contrived metaphor; but hopefully it works.
 
Hmm, this isn't an area I am too familiar with. But I believe it comes down to:

- The 'model' gets trained and validated on data at a specific sample-rate.
- This means the model can generate data designed for the same sample-rate.
- User comes along and wants to increase the base sample-rate in their DAW.
- Now none of the training or validation data is relevant.
- Model doesn't know how to interpolate for the new sample-rate.

And actually, this would happen even when going from a high to a low sample-rate - so for example, if the model was trained on 96kHz data, and the user was running it at 44.1kHz.

If you imagine this entire thing as a pedalboard....

In component modelling land, you've got:

Pedal 1 > Pedal 2 > Pedal 3

Then dealing with aliasing becomes like putting other pedals in between those pedals.

But in NN/ML land, you've only got 1 pedal, and putting other pedals either side of it won't fix the issue. You actually need to change the design of the pedal.

That's quite a contrived metaphor; but hopefully it works.

Thanks. Yep actually got it ... will miracles never cease :)
 
I was told that and it didn’t really help. And I can’t see

That’s exactly why i sold my kemper to buy an fm3

I gotta be really honest. I bought the kemper to replace my tubes amps as I was getting hired by a band that required silent stage, stereo effects etc.

I struggled like never to get a cardboarded non dynamic kind of plexi crunch. I just coud not play the kemper it was nothing like an amp.

First rehearsal, the other guitarist in the band had a ax8….
It was like I was playing on a peavey bandit and he had a full Bradshaw rig…

And don’t talk about the dynamic, if kemper can’t replicate the dynamic of a plexi then it just can’t replicate dynamic, just faking it.
Works for everyone else. Hmmm.....
 
Here's a quick comparison I made to illustrate how oversampling can reduce aliasing with a component based plugin modeling a tubescreamer.

The project is 48 kHz. I use Cytomic's The Scream (great plugin!), which lets me set the internal oversampling manually. So the only thing changed between the two cases shown, is that the oversampling in the plugin is manually changed from 2x to 8x.

2 x oversampling:

View attachment 45492

8 x oversampling:

View attachment 45493

Nice work.

So what this tells us is that:

1 - No, running the entire thing at 48kHz is not enough to clear up the aliasing. At 48kHz that puts Nyquist at 24kHz. But non-linear distortion will generate frequencies much higher than that.

Theorhetically speaking, non-linear distortion generates infinite harmonics. These extend far past nyquist, and this will happen at any core sampling rate. If you don't handle these frequencies, then they bounce back off the Nyquist frequencies, and change direction - this is why it is called foldback distortion.

A 30kHz harmonic at 48kHz sample rate will reflect back to 18kHz; and this is true for any frequency. It isn't a random process. There is a true mathematical nature to where a frequency will fold back, and sine tones reveal the nature of the system very well. But all of this is equally applicable to any input signal.

2 - Oversampling at 8 times massively cleans it up. BUT IT CLEARLY ISN'T JUST OVERSAMPLING THE SIGNAL. You can clearly see the presence of an anti-aliasing filter in the chart; a tell tale sign of this is where the line goes up, reflects down a tiny amount, and then disappears.

Why does oversampling work? It pushes the Nyquist frequency much higher. So 48kHz * 8 = 384kHz, which gives the system space to represent those frequencies that the distortion generates; even though we cannot hear them. But this alone won't cure the problem, because at some point the signal needs to be downsampled. You're not running your DAW at 384kHz; you're running it at 48kHz. So the plugin needs to downsample the audio to match what the DAW is expecting. Which is an opportunity for all of those harmonics to reflect back; even though you've given them plenty of space earlier on.

This is where the anti-aliasing filter comes in. The fact that frequencies are filtered out when the signal descends in the graph, tells us that the anti-aliasing filter is being applied during the downsampling phase of the plugin.

An anti-aliasing filter is easiest to understand as a brick wall filter, that blocks frequencies above the cutoff. There's more to it than that, because you also need to choose a design that isn't going to destroy the phase of your signal, while not blowing up your DSP/CPU usage; because this filter is going to be running on every single sample that gets processed. Can be quite a heavy operation.

PS: I used to work alongside Andy at Cytomic, right when I got into the industry. We were both at FXpansion. Great guy! Super genius at DSP. Makes great plugins.

When I was doing a lot of recording back in the day, I did everything in the box, and I used " The Glue" Compressor which was from memory, a Cytomic product - it was glorious on Bass and Guitar and pretty great on the Drum Buss too - it was a dream to use.

Man is was so good on everything and so easy to use.
 
Kemper banned me from their Forum for a week:grin.
Some one asked if the new profiles would be backward compatible. Something that was poorly explained, if at all in their press release.
Some posters just mentioned that the press release could have had this info to avoid the confusion,
One of their moderators told the poster, they could have reached out to the support line.
I called out the ongoing arrogance and sarcasm of the support staff over it's poor communication and BOOM, gone for a week.

Boy, cultural differences aside common sense business practices dictate that you should not treat your customers and prospective customers like this. "We were clear, you're just stupid."

They're not the only ones. Cliff at FAS can be pretty brutal on his own forum, over defending anything he is challenged on, but that usually involves an esoteric engineering explanation of his point of view. Kemper's staff, just holds a middle finger up and quietly fixes issues w/o admitting there's an issue.
 
Last edited:
So what does the aliasing sound like? Is it extra high end content, almost like a white noise layered into the sound?

I've noticed a fair amount of noise around some notes when picking very hard, it's like an over the top distortion "whoosh" sound. The Axe 3 does this as well, maybe to a lesser extent. I'll need to fire up my real amp and compare as well.
 
So what does the aliasing sound like? Is it extra high end content, almost like a white noise layered into the sound?

I've noticed a fair amount of noise around some notes when picking very hard, it's like an over the top distortion "whoosh" sound. The Axe 3 does this as well, maybe to a lesser extent. I'll need to fire up my real amp and compare as well.
I think what you're describing is what Cliff calls "clutter" that is inherent to some amp circuits.
 
So what does the aliasing sound like? Is it extra high end content, almost like a white noise layered into the sound?

I've noticed a fair amount of noise around some notes when picking very hard, it's like an over the top distortion "whoosh" sound. The Axe 3 does this as well, maybe to a lesser extent. I'll need to fire up my real amp and compare as well.

Imagine every time you played a Cmajor chord, right there in the distant background there is a bminor 17th chord at the same time. Imagine how shit it would read if you wrote that down in a score and asked 300 violinists to play it. Then Hannibal Lecter cuts your heart out right in front of your wife and kids as punishment. Then he eats it with a McDonald's milkshake; the vintage kind. Proper thick and fluffy, not the watered down shite they sell today.

That's what aliasing can potentially be.

<chatgpt takes over>

Your speakers start whispering secrets they were never supposed to know. Not in words — no, in flickers of ghostly harmonics and jittering transients, like the sound of a broken modem trying to summon the dead.

Each note you play becomes two, then four, then a crawling swarm of warped spectral imposters — glassy, sharp-edged frequencies that weren’t invited to the party but showed up drunk anyway. The further up the neck you go, the more they multiply, until your solo sounds like a metal scrapyard being sucked into a black hole.

Every time you play a note — just one clean note — aliasing slips in like a hex, whispering backwards Latin through your signal chain. Not loud. Just enough for the dog to start growling at your amp. Just enough to make you wonder if it really sounded like that when you recorded it.

You're on stage. The lights dim. The crowd roars. You hit that first glorious open C chord — and aliasing, that unholy glitch-ghoul, rides shotgun through your amp. It isn’t loud enough to notice at first, just a thin, metallic shimmer — like rusted tinsel on your tone. But it spreads.


Your delay pedal, once lush and angelic, now repeats with a cold, robotic sneer. The reverb trails curl inward, souring into digital rot. Every harmonic above the 5th is no longer yours — they’ve been possessed. The top end becomes a nest of broken glass and razorwire.

You crank your solo — finger poised for that searing high note. You strike — and instead of piercing the air like a war cry, it tears sideways, splitting into bent, dissonant shrieks that no one invited. The front row winces. Your drummer glares. You feel it slip — not just pitch, but soul.


You try to roll back the treble — too late. The PA system feeds back with a scream that seems to know your childhood fear. The monitors speak in tongues. You swear you hear laughter through the mids.

Your in-ears lie to you. The crowd hears one version of your tone — shrill, chaotic, like a swarm of cursed circuit boards melting in real time. You're hearing another — hollow, haunted, like your guitar's ghost playing through a broken radio at the bottom of a well.


Your tone is gone. Your soul is forfeit. All because you didn’t lowpass at Nyquist and thought,
"Maybe I don’t need oversampling."


That’s aliasing in a live rig.
The devil on your pedalboard.
Smiling. Always smiling.
 
Back
Top