Monkey Man
Groupie
- Messages
- 39
Only if you say Luvs with a German accent.Isn't that a diaper brand?
An integrated LUFS is a way to answer the accuracy question one wants answered: is the null test residual significant? In other words, how loud is it? The residual will have energy from any discrepancies against the original for things like dynamics, frequency response, and artifacts that vary over time. Integrated LUFS is a way to measure the magnitude of that difference over the duration of the test in a way that is meaningful for human hearing.
The problem usually isn't LUFS. The problem is usually the null test. Performing a good null test that handles things like normalization, baseline establishment, and phase alignment issues can be devilishly difficult. More importantly, if you're just using a single short DI, that's only going to analyze a single playing style. For that reason, it's hard to assign a single accuracy scalar value to a profiler. To thoroughly test profiler accuracy, you need multiple null tests with a variety of DI's that cover a range of styles. For example, many notes in one DI to analyze transients, or a long sustained chord in another to analyze decay.
It's an imperfect method, but if done properly, it does produce some useful information about profiler accuracy.
- Messages
- 10,669
By doing the above, you are testing signal similarity, not system equivalence. And a guitar amp is not a memoryless system.
An amp is a stateful nonlinear dynamic system.
The output depends on:
• the current input
• the recent past input
• the energy stored in capacitors
• bias point shift
• thermal drift
• supply sag recovery
• blocking distortion recovery
Two identical DI files played at different times within the same performance can produce different harmonic structures.
So a null test is actually answering:
“Did the model reproduce the waveform for this particular excitation history?”
Not:
“Did the model reproduce the amplifier?”
To actually test modelling accuracy you need to observe how distortion orders change over time and level (multi-order Hammerstein / harmonic IR analysis), not just the loudness of the residual.
I promise you, Line6 and Fractal are not building virtual circuits, and then performing a basic vanilla null residual to "guess" at how accurate they are.
A single LUFS value - and what's more, a hierarchy of them like Leo Idiotson provides - is about as useful as a chocolate teapot, when it comes to guitar amp comparison.
I think the thing you are missing in your ChatGPT prompt is the important word "integrated". That will indeed measure artifacts due to stateful influences. Integrated LUFS will measure similarity and it will measure it in a way that is meaningful to the human ear.
There are some alternative ways to measure the integrated magnitude of the residual, but it's debatable whether they would improve the measurement.
Again, it's not perfect, but it does provide some useful information about accuracy, particularly in a comparative analysis.
As I said, I agree with you there. You have to be careful about being so reductive that you assess accuracy with a single scalar value.
- Messages
- 10,669
Integrated LUFS is an energy metric. Amp tone and feel errors are structural - LUFS cannot represent them.
In order to even get close to comparing captures to a real amp, you want something like the above; which is my custom port of the NAM code over to JUCE, with my custom tri-linear IR blender on the output (no SIC connectivity)
I do use ChatGPT - but not to win arguments on forums. I use it to read DSP papers and to help write code. But thanks for the condescension!
The errors you are interested in are the ones you can hear. That's the value in measuring the magnitude in a scale that is designed to be meaningful for human hearing. As i said, there are alternative ways to measure the magnitude of the residual, and some of them are interesting to explore, but it's debatable whether they are an improvement over an integrated LUFS.
The irony here of course is discussing this in a Kemper thread, when Kemper will be going to great pains over the coming weeks to convince us we shouldn't be paying attention to accuracy
.
- Messages
- 10,669
I'm really not sure how else to explain it to you, and you're stubbornly refusing to get it.The errors you are interested in are the ones you can hear. That's the value in measuring the magnitude in a scale that is designed to be meaningful for human hearing. As i said, there are alternative ways to measure the magnitude of the residual, and some of them are interesting to explore, but it's debatable whether they are an improvement over an integrated LUFS.
The irony here of course is discussing this in a Kemper thread, when Kemper will be going to great pains over the coming weeks to convince us we shouldn't be paying attention to accuracy
LUFS answers a different question to the one that we are asking, when we are comparing an amp capture to the real amp. We are not asking about signal similarity when we ask that question. We are asking about system significance. LUFS cannot answer that question. Because it isn't a multi-dimensional measurement, and only tells you about the long-term energy residual. It doesn't tell you anything about the system.
If you want to know about loudness, sure. It is useful. If you want to know about harmonic structure, reactivity time, or anything else to do with the system, it won't answer that at all. You're at the complete whim of your input signal at that point.
Comparing recordings != validating a model.
Chocol8
Rock Star
- Messages
- 3,101
I hate to quote myself, but I think it's relevant. The man is a true believer in LUFS, and I don't think there is anyway we are going to convince him with facts.
We'll just have to agree to disagree on that. I'll just point out there's a school of thought that holds that what matters is what output you hear when you plug in your guitar and play through it, and how similar it sounds to the original amp.
Chocol8
Rock Star
- Messages
- 3,101
And LUFS doesn't measure that no matter how much faith you have that it does.
- Messages
- 10,669
Your whole argument essentially boils down to this:
"Using less information is a better approach than using more information, when it comes to comparing a source real amplifier to a digital recreation of it"
And to be honest, I just cannot agree to disagree on that. I think it is plain wrong.
No, that's an incorrect paraphrase. Just to be clear, I said the opposite of that. I said trying to boil down profiler accuracy to a single scalar value is overly reductive.
- Messages
- 10,669
You have been defending the combination of a null test and a LUFS measurement. So no, you have not said the opposite of that. Unless you are now changing your mind? You're now starting to make very much sense.
Again, I tell you... the big players in this field are not relying on this mechanism that Leo Gibson and others rely on. For precisely the reasons I've outlined.
- Messages
- 10,669
Posting this again:
And again, the methodology isn't 100%. But here was my thinking:
- I needed to scale the data against something that more accurately resembled human hearing; I chose the mel scale.
- I needed to be able to see where in this scale the accuracy was highest, and where it was lowest; hence converting to a spectrogram.
- I needed to be able to judge behaviour over time.
A plot like this:
Tells you much more about behaviour over time than a single LUFS measurement.
One thing I did not account for at the time however was spectral smear from the fftsize and hop size interaction. If I was to do this today, I'd probably take a multi-resolution approach, or even use wavelet analysis instead, for a more accurate picture of what is going on in terms of frequency response and behaviour.
And again, the methodology isn't 100%. But here was my thinking:
- I needed to scale the data against something that more accurately resembled human hearing; I chose the mel scale.
- I needed to be able to see where in this scale the accuracy was highest, and where it was lowest; hence converting to a spectrogram.
- I needed to be able to judge behaviour over time.
A plot like this:
Tells you much more about behaviour over time than a single LUFS measurement.
One thing I did not account for at the time however was spectral smear from the fftsize and hop size interaction. If I was to do this today, I'd probably take a multi-resolution approach, or even use wavelet analysis instead, for a more accurate picture of what is going on in terms of frequency response and behaviour.
Enchilada Jones
Roadie
- Messages
- 910
I think Leo isn’t going to give up being ‘the null test youtoober’ because that distinction is probably boosting his numbers.
I appreciate his desire to help but I’m thinking a Tuco Stop Helping is needed
I appreciate his desire to help but I’m thinking a Tuco Stop Helping is needed
I have frequently argued that there is much more to a live guitar rig than how accurately it captures.... much more. It is clear that for some people, the NULL test is "all important". This philosophy doesn't seem logical to me as it would imply that Axe III Fx and Helix are very poor since they don't capture at all. This philosophy would also imply that NDSP pedals are superior solutions.
As a live gig player, this doesn't make any sense at all; however, for people that have lots of amps that they want captured perfectly, and then want to create their own pedal boards to use those capture with ..... I can see how important it is for them.
It is my belief that LUFS has gone into the "tribal belief" realm.
My Kemper MK1 sounds glorious in the mix and is an outstanding live gig tool. The reliability is 100% over the last 13 years (not a single on-stage problem). Never had a tone change because of a firmware change. The unit integrated with my tablet setlist manager wirelessly using a CME WiDi MIDI adapter.
I applaud Kemper for (finally) acknowledging that they need to improve their profiling. I am still TOTALLY mystified why Kemper doesn't market their device for gigging musicians as I feel it is one of the best solutions for this purpose. They seem hell bent on winning the LUFS war ..... and I don't believe it is a war they can win.
Anyway, it will make for interesting forum reading when V2 FINALLY is released.
As a live gig player, this doesn't make any sense at all; however, for people that have lots of amps that they want captured perfectly, and then want to create their own pedal boards to use those capture with ..... I can see how important it is for them.
It is my belief that LUFS has gone into the "tribal belief" realm.
My Kemper MK1 sounds glorious in the mix and is an outstanding live gig tool. The reliability is 100% over the last 13 years (not a single on-stage problem). Never had a tone change because of a firmware change. The unit integrated with my tablet setlist manager wirelessly using a CME WiDi MIDI adapter.
I applaud Kemper for (finally) acknowledging that they need to improve their profiling. I am still TOTALLY mystified why Kemper doesn't market their device for gigging musicians as I feel it is one of the best solutions for this purpose. They seem hell bent on winning the LUFS war ..... and I don't believe it is a war they can win.
Anyway, it will make for interesting forum reading when V2 FINALLY is released.
![[Nathan]](/data/avatars/m/0/613.jpg?1726885163){kind=avatar}
[Nathan]
Shredder
- Messages
- 1,266
Yeah completely different discussions. How it sits in a mix, how many features it has, price point, everything else that makes it great for a guitarist has little to do with comparing its capturing accuracy. I think a lot of people will say it’s a good / good enough digital rig, but their legacy capturing approach has been overtaken by the newer wave net model stuff.
As a side note to this Misha recently posted his axefx ultra tones and they were 10/10 top tier. But it’s still an axefx ultra etc
Last edited:
ian_dissonance
Shredder
- Messages
- 2,144
There’s no point in marketing you profiling/capturing if you can’t claim accuracy. Unless you’re one of those OEM Chinese brands. Whether it sounds great or how easy it is to use in performance is totally something other than the profiling. The point of captures/profiles is to be as 1:1 to a verifiable source as possible. Once something fails the criteria here it’s biffed its main job. Most people don’t care, but that’s besides the point of the technology.
Lysander
Rock Star
- Messages
- 3,513
Problem is, that's not the question YouTubers are looking to answer. It's always "X has lower delta loudness than Y, so it's clearly a more accurate capture."
Similar threads
- Poll
