Kemper Profiler MK 2

[Sascha Franck]

Some hours late to the party, but well, the Kemper staff denying there were certain issues with aliasing - that is just as embarrassing as it gets.
And remember, this is coming from someone who could easily get along with the sound quality of the old Kemper revision - yet, this is completely inacceptable.

 

[Sascha Franck]

Sure, it can null nicely but there are other variables.

Such as?

 

[Dom Mcsweeney]

Such as?

perceived folding and what not.

 

[Orvillain]

Until the juice runs down my leg.

 

[OneEng]

From here:

The null test being perfect may insure that the capture is identical to the source and that there is NO sonic difference.

The null test being less than perfect DOES NOT mean that there IS a sonic difference from the capture to the source.... but it COULD mean that there is.

I still maintain that it doesn't matter. Good tone is good tone. I have heard profiles that I like better than the original amp (as an example).

Still, I wouldn't be surprised to see Kemper's new algorithm get much closer to a perfect null test. It seems that doing such a thing makes a certain part of the market very happy.

 

[ian_dissonance]

I don’t think anyone expects a null test to be exactly canceled, but he’s saying some fine hash and an occasional low end peak means a null test is useless. That’s like saying the loudness of something is only the highest momentary LUFS. If one thing nulls to -25db and one nulls to -65db average, one of those is more accurate, measurably. Whether the limits of hearing catch it or not isn’t the point of the test. The point of the test is to see which thing is CLOSEST to the original constant.

 

[LP59]

From here:

The null test being perfect may insure that the capture is identical to the source and that there is NO sonic difference.

The null test being less than perfect DOES NOT mean that there IS a sonic difference from the capture to the source.... but it COULD mean that there is.

I still maintain that it doesn't matter. Good tone is good tone. I have heard profiles that I like better than the original amp (as an example).

Still, I wouldn't be surprised to see Kemper's new algorithm get much closer to a perfect null test. It seems that doing such a thing makes a certain part of the market very happy.

I remember when he posted that. He actually did two videos about null tests and he completely botched the test and interpretation of the results. Many people explained that to him in the video comments and he deleted every one of them. There are hardly any comments remaining on those two videos .

 

[norminal]

I remember when he posted that. He actually did two videos about null tests and he completely botched the test and interpretation of the results. Many people explained that to him in the video comments and he deleted every one of them. There are hardly any comments remaining on those two videos .

 

[DLC86]

The null test being less than perfect DOES NOT mean that there IS a sonic difference from the capture to the source.... but it COULD mean that there is.

While that is true in theory it's rarely true in practice.
E.g. if I take an audio file and rotate the phase by 90° on all frequencies, that will produce a very bad null test result but the difference to our ears will be negligible.

But IME, specific cases like that never happen when comparing amp captures, in every test I've done and seen, the one producing the worst null was also the one sounding the most different compared to the source.

Also, amp captures based on neural networks make models by trying to match the waveforms, so the chance of them adding something that skews a null test is pretty much inexistent.
Btw, in that regard, some time ago I made a tool that automatically aligns two files and performs a null test, also giving you measurements based on the same parameters (ESR, MAE, MSE and MRSTFT) that NAM uses during the training.

TTS Null Tester | The Tone Scientist

Free and premium presets for Fractal Audio Axe FX III, FM9, FM3 and Line 6 Helix. Amp and pedal captures for NAM and Aida-X. Speaker and cab impulse responses.

shop.thetonescientist.com

 

[laxu]

Let me be more blunt. @Orvillain @laxu @la szum @norminal or anyone. Please consider taking 5 minutes to read this white paper and give you thoughts about the discussion of combining white and black box modeling. Link again:

https://res.hotoneaudio.com/prod/support/CDCM%20HD_White%20Paper_V01_200323.1584955064392.pdf

I've read this before, and own a Hotone Ampero 2 so I'm in the perfect position to comment.

I think it works pretty well, but has its own share of issues.

Hotone currently has two grades of modeling on the A2. They have standard and a few "HQ" models that they call "Hotone's next gen modeling" in the amp model's description. The only HQ models in the unit are all channels of Soldano SLO and JP2C. There's SLO models available as both standard and HQ versions, and the HQ ones do sound a bit better.

The main difference in the HQ models seems to be a much more configurable poweramp model where you can pick tube types, adjust some params etc.

Overall either model types do a pretty good job, but I've noticed a bunch of problems:

• Not all amps have authentic controls. E.g Mark IIC model is missing the 2nd gain knob.
• Master volume amps do not go into poweramp distortion.
  • Non-master volume amps do go into poweramp drive.
  • This is a big accuracy issue, but realistically not a dealbreaker because I run pretty much all my real MV amps without poweramp drive.
• I found a bright switch on a Dumble model that seems just broken. It makes the model much quieter and bad sounding. I reported it to Hotone but so far haven't seen fixes.
• Models seem to be darker than e.g Helix in the 5KHz+ frequency range.

I've tried the JP2C into a Fryette PS and compared it to my Mesa Mark V 90. I also tried the JP2C preamp into the Mesa poweramp. Both cases did a pretty good job, and the JP2C preamp into the Mark sounded very similar to the real ch3 preamp.

But for a relative newcomer in the modeling game, at a cheaper price point no less, I think Hotone does a pretty good job with its models. Despite inaccuracies, they sound good and feel good to play. If they keep figuring out how to upgrade them in future products (or better yet...current ones!) then they'll have a very competitive product. That's a pretty good result considering they don't have like 20+ years of experience like Fractal or Line6 do, and have only "a bit more than HX Stomp" level processing power.

So what about Kemper? I think they could get a better result if they adopt this sort of system. "Liquid profiling" is pretty similar to the Hotone system in concept, the difference being that with Kemper you could apply this to any amp. So if their upcoming new algorithms will do a better job at it, maybe it will work pretty well.

 

[Orvillain]

The null test being less than perfect DOES NOT mean that there IS a sonic difference from the capture to the source

That is precisely what it means.

 

[MirrorProfiles]

The thing with Kemper’s accuracy is that it doesn’t need null tests or any kind of visual representation to hear how far off it is, because you can hear it when listening without much trouble. Those kinds of things are more helpful when it’s impossible to distinguish and you still need to split hairs. Kemper just tends to have audible giveaways in comparisons, ToneX and NAM also have differences but they can be harder to notice.

 

[Sascha Franck]

I don’t think anyone expects a null test to be exactly canceled, but he’s saying some fine hash and an occasional low end peak means a null test is useless. That’s like saying the loudness of something is only the highest momentary LUFS. If one thing nulls to -25db and one nulls to -65db average, one of those is more accurate, measurably. Whether the limits of hearing catch it or not isn’t the point of the test. The point of the test is to see which thing is CLOSEST to the original constant.

Well, there's one thing to consider, though. Note: What I'm saying is largely hypothetical (as in "made up"), it might yet illustrate the point.

When you're doing a nulltest and compare the results by plain dB measurements, it could very well be that the "losing" candidate could still be more accurate (and I'm not talking about "sounding better to ones ears" but scientifically more accurate) in the important frequency range.

A nulltest covers the entire frequency spectrum. So, what if one of the candidates is horribly off in the lower end but pretty damn near to perfect in the entire rest of the frequency spectrum whereas the other contestant is mildly off all over the frequency range? The one with the "wrong" low end will likely lose in a plain dB based nulltest - but it might still be the better one for any real life application because it's doing better in the relevant frequency range.

As said, this is purely hypothetical, but it does sort of illustrate why a nulltest isn't always the be-end-it-all means to demonstrate accuracy (or the lack thereof). And it's not even all *that* unlikely that it happens sometimes.

 

[Lysander]

When you're doing a nulltest and compare the results by plain dB measurements, it could very well be that the "losing" candidate could still be more accurate (and I'm not talking about "sounding better to ones ears" but scientifically more accurate) in the important frequency range.

You're not wrong. In general, take any null test result as single numbers with a huge grain of salt. They will tell you nothing about how a profiler/capturer behaves across the entire frequency range.

Leo Gibson gets it closer than most when he measures null diffs in LUFS (normalized for loudness) but it still drives me crazy when he does things like this:

 

[DLC86]

A nulltest covers the entire frequency spectrum. So, what if one of the candidates is horribly off in the lower end but pretty damn near to perfect in the entire rest of the frequency spectrum whereas the other contestant is mildly off all over the frequency range? The one with the "wrong" low end will likely lose in a plain dB based nulltest - but it might still be the better one for any real life application because it's doing better in the relevant frequency range.

LUFS measurements are K-weighted, so each frequency is weighted to roughly correspond to the human hearing, a low frequency signal will be measured at a much lower dB LUFS level than a mid frequency of the same dBFS level... so that's a non-issue

 

[Sascha Franck]

Leo Gibson gets it closer than most when he measures null diffs in LUFS (normalized for loudness) but it still drives me crazy when he does things like this:

Well, it's still just covering the entire frequency range.
One could possibly separate all test files by cutting them in "frequency slices", using EQs with a brickwall high/low cut (such as the Nova EQ, IIRC) and get a more meaningful outcome.
But then, in general I have not too much of a doubt that in general, the capture device accuracy rankings are largely correct because most people in the knows seem to just agree on them.

 

[Sascha Franck]

LUFS measurements are K-weighted, so each frequency is weighted to roughly correspond to the human hearing, a low frequency signal will be measured at a much lower dB LUFS level than a mid frequency of the same dBFS level... so that's a non-issue

Well, corresponding to the human hearing still doesn't necessarily mean too much in terms of "relevant for guitar amplification".

 

[Orvillain]

1+2 = 3
2+1 = 3
6 / 2 = 3
1+1 = 2

The first three are the same. The last one doesn't match.

When comparing a capture to the real source, without any variables changing you would hope for X or Y to be exactly the same. If they're not, then this shows difference. Difference results in a sonic different, at whatever scale you choose to analyse.

Now whether that is important to you or not is a completely different question, and is almost not even within the realms of the science of empiricism; it is more akin to feelings, emotion. We're not making any claims about that when we do null tests.

Null tests are simply a way of comparing two signals to see if they match, and if they don't, to reveal where they don't match. That's it.

Another way to say this is that when you do a null test and the bass region is off, it is you who determines whether that is worse or better. Not the null test. The null test simply shows you what is true.

Essentially there are two questions at play:

1. Are there differences?
2. Do they matter?

Null tests answer the first one.
They cannot answer the second one. Only a human can do that.

Additionally, I agree with Sascha. Analysing a null test based on your metering in the DAW is really roughshod, and isn't the appropriate way to do it. The appropriate way is to perform the null within the spectral domain, just as I did here:


This test very literally tells you exactly which frequency areas had the largest differences. To this day, no one has ever replicated my comparison, or even talked about it very much at all. Most I've had is a few guys just laugh at it. Very strange.

But to my mind, when it comes to replicating the sound of a real amp, the hierarchy is:

NAM
ToneX
QC
Kemper

 

[Orvillain]

Also I really don't know how Leo has gotten such a following. His methods are extremely basic.

 

[Lysander]

LUFS measurements are K-weighted, so each frequency is weighted to roughly correspond to the human hearing, a low frequency signal will be measured at a much lower dB LUFS level than a mid frequency of the same dBFS level... so that's a non-issue

It's still an issue. You're correct that K-weighting de-emphasizes lower frequencies... which means i can construct a null-test result with very audible garbage on the low end, and still have it measure better in LUFS than a NAM plugin.

Doing null tests the Leo Gibson way is an entirely valid approach, don't get me wrong, but ranking devices based on those results is flawed AF. Maybe one can boil null-test results to a single number for comparisons sake, but LUFS on audio deltas is definitely not it.