Kemper Profiler MK 2

[Orvillain]

This is interesting and, of course, miles over my head. I considered laughing, but instead I'd like to ask if there's a "...for idiots" guide available somewhere?

Sound == Lego.

Lego breaks down into loads of different lickle blocks. We call those frequencies.

Those frequencies enter our brain holes and make nice feels.

These graphs show you how much nice feels are happening inside your panties when using various capture platforms.

The gold standard of panty feels is a real amp. Always will be.

 

[Orvillain]

Imagine having lego in your ears. Or your pants.

Awful.

 

[DLC86]

LUFS gives you a normalized value as you say. But it doesn't give you specifics.

It's like the difference between using a flawthrower to solder a capacitor into a pedal, versus using a soldering iron.... that's your dBFS versus LUFS.

Converting to the frequency domain is like having a magnifying glass to assist your soldering iron.

LUFS == perceptual loudness
Spectral analysis == adding a billion eyes to your face, each one with x-ray vision, so you can be the biggest, baddest, hard as fuck Dark Souls end of level boss.

Yes, that was clear to me... my question was: what's the added benefit of this magnifying glass? Do you really get additional useful information by knowing which frequencies are more/less accurate?

 

[Orvillain]

Yes, that was clear to me... my question was: what's the added benefit of this magnifying glass? Do you really get additional useful information by knowing which frequencies are more/less accurate?

You don't want to know in which areas of the frequency spectrum your capture is more/less accurate than the source? You don't think that information is useful?

 

[Sascha Franck]

Sound == Lego.

Lego breaks down into loads of different lickle blocks. We call those frequencies.

Can you step on them as well?

The gold standard of panty feels is a real amp. Always will be.

Now, that is irrelevant to this discussion. And it's pretty different for me. Any amp would have to compete with my current setup. And it'd have a hard time.

 

[DLC86]

You don't want to know in which areas of the frequency spectrum your capture is more/less accurate than the source? You don't think that information is useful?

Just for the sake of knowing it, maybe. But with LUFS I already know that a certain value corresponds to a certain difference for my perception, regardless of the area where that difference exists. That's enough for me personally for that scope.

 

[Orvillain]

Just for the sake of knowing it, maybe. But with LUFS I already know that a certain value corresponds to a certain difference for my perception, regardless of the area where that difference exists. That's enough for me personally for that scope.

“I don’t care where the engine is broken, I just know the car feels slower — and that’s enough for me.”

 

[DLC86]

“I don’t care where the engine is broken, I just know the car feels slower — and that’s enough for me.”

Exactly, I'm not a mechanic

 

[MirrorProfiles]

Point is they are all imperfect. The difficulty in telling where the differences are does not determine if those differences matter to the player or even if the differences are less or More desirable.

I don't think the discussion is whether or not any capture method is perfect. The discussion is about HOW accurate they are, and Kemper simply falls audibly far short of the other popular options in 2025. This is no matter what metric you want to use - be it listening in a blind test, null test, ESR, or whatever. It is simply less accurate than a quad cortex, less accurate than ToneX and significantly less accurate than NAM at high training settings.

Whether or not the differences matter to each individual player is not really relevant. Someone listening on a terrible playback system isn't going to care as much as someone in a more critical listening environment. Again, it's moot because that has no bearing on whether or not a capture method is accurate. That's down to the listeners own environment and needs which is a totally different discussion to this.

 

[DLC86]

“I don’t care where the engine is broken, I just know the car feels slower — and that’s enough for me.”

Btw, that's a great analogy to clarify my point too.
Knowing which part of the engine is broken is useful for those that have to fix it, not for those who drive the car.
Just like knowing where the difference lies in the frequency spectrum is only useful if you want to find out where those differences come from to maybe fix them. It's not more useful to just find out which one is a more accurate model in general.

 

[Orvillain]

Exactly, I'm not a mechanic

Right, well I'm at the very least a mechanic in training. So....

 

[Orvillain]

Btw, that's a great analogy to clarify my point too.
Knowing which part of the engine is broken is useful for those that have to fix it, not for those who drive the car.
Just like knowing where the difference lies in the frequency spectrum is only useful if you want to find out where those differences come from to maybe fix them. It's not more useful to just find out which one is a more accurate model in general.

I think it is useful for a customer to know how likely their car is to break.

 

[Sascha Franck]

Car analogies, yay, finally!
It might not matter how a caterpillar might show some horrible steering inaccuracies at 100mph.
By the same token it might not matter whether a capture shows some horrible inaccuracies at 10Hz.

 

[DLC86]

I think it is useful for a customer to know how likely their car is to break.

Don't "drift" away with your analogy

 

[Lysander]

You mean like human hearing?

Yes. But Leo Gibson is using LUFS to measure capturing accuracy and not perceived results, which means that his results favor mid- and high-frequencies. So this...

...is 100% wrong. Again, there is value to this type of measurements, but you need to understand what you're trying to get out of them.

Otherwise you end up ranking the Sonulab StompStation the same as the Ampero II Stage, even though the null test for the former sounds like complete crap.

 

[DLC86]

By the same token it might not matter whether a capture shows some horrible inaccuracies at 10Hz.

You seem to not be getting that this is already accounted for by the weighting in the LUFS measurement and by the absolute level of the amp's output.

 

[Orvillain]

Don't "drift" away with your analogy

 

[Orvillain]

Car analogies, yay, finally!
It might not matter how a caterpillar might show some horrible steering inaccuracies at 100mph.
By the same token it might not matter whether a capture shows some horrible inaccuracies at 10Hz.

For a brief moment I thought you meant....

 

[DLC86]

Yes. But Leo Gibson is using LUFS to measure capturing accuracy and not perceived results

Could you explain me what's the difference between those two things according to you?

Otherwise you end up ranking the Sonulab StompStation the same as the Ampero II Stage, even though the null test for the former sounds like complete crap.

I'm curious to hear that now...

 

[Orvillain]

Leo Gibson is using LUFS to measure capturing accuracy and not perceived results

Yep. This is the wrong approach if you want objective accuracy measurements.

It simply needs to be done at the spectral level. There is no other way. Basic "level meter" analysis (whatever the weighting is!) is just not fine enough detail to give any kind of realistic impression of how accurate something is.

You need to:

- Record the amp with a source signal. Sine tones or pink noise work best.
- Record the captured profile with a source signal. Sine tones or pink noise work best.
- Run an STFT over the source.
- Run an STFT over the capture.
- Account for any FFT smearing artifacts due to windowing or overlapping.
- Perform a diff across both data sets.
- The diff shows you what and where the two audio signals are different.
- At this stage, if you wanted some kind of weighting... you could do it all again, but apply a log conversion, or a weighting against the Bark/Mel scales to account for human perception. Both data sets are valid and can tell you things.


If anyone is interested in doing this, grab Python, Librosa, Matplotlib, and get to work!!