Ten billion profiles/captures = one amp

[metropolis_4]

There are a variety of reasons I'm opposed to all this profiling/ML/AI/etc. stuff:

- To fully sample an amp takes years/decades so in practice you only get a handful of snapshots.

- The data is opaque. You can't edit the data as there are no parametric relationships. It's just a bunch of data with no insight into what any of it means.

- Fundamental understanding of how an amp works is lost. Someone can make/sell a product that has samples without any understanding of why tube amps sound the way they do. The more people lean on this technology the more this knowledge will be lost to time.

- You can't make virtual amps. You can't design an amp completely in the virtual domain. You can only sample what already exists. So you can't make a virtual amp that does things that real tube amps can't do (i.e. FAS Modern).

- Guitar tone will never evolve. If we relegate ourselves to simply copying existing products we'll never evolve beyond that. We should be asking why did tube amps become the gold standard of guitar tone? Why did solid-state never gain widespread acceptance? What is it about tube amps that is pleasing? What can we improve upon? I have spent almost two decades now trying to answer those questions and I have some theories.

You know, I never thought of it that way before; it makes a lot of sense.

All of this new technology being used to simply copy the results of old technology. But it’s not actually capable of creating anything original or advancing guitar amplification. It can only imitate what already exists.

That would mean the only way to continue advancing guitar amplification would be to continue creating new physical amps so they could be copied in the digital realm. While modeling allows the ability to actually create new amps entirely within the digital realm.

 

[laxu]

- Guitar tone will never evolve. If we relegate ourselves to simply copying existing products we'll never evolve beyond that. We should be asking why did tube amps become the gold standard of guitar tone? Why did solid-state never gain widespread acceptance? What is it about tube amps that is pleasing? What can we improve upon? I have spent almost two decades now trying to answer those questions and I have some theories.

I'd love to hear your thoughts on this.

Since Fractal is ever closer to perfecting component modeling, once you get there, what would you like to do next? Is there another area that you'd like to look at replicating virtually (e.g cab modeling instead of IRs)?

Fractal is in many ways locked onto the paradigm of "ooh look at all these real amps we model" because that's what people want to buy. But let's say that a new strain of COVID alters guitarists' minds so they no longer crave digital replicas of real amps. What kind of virtual amps would you like to design?

 

[DrewJD82]

There are a variety of reasons I'm opposed to all this profiling/ML/AI/etc. stuff:

- Fundamental understanding of how an amp works is lost. Someone can make/sell a product that has samples without any understanding of why tube amps sound the way they do. The more people lean on this technology the more this knowledge will be lost to time.

There’s a lot of things that happen in a tube amplifier that are not really well understood. There’s a lot that we do know, what the tubes are doing, saturation curves and all that – that’s really well documented. But in the later stages of the amplifier, what happens with the interaction between the power supply, the pentodes, the output transformer and the reactive load of the speaker, there’s a lot of weird things that happen there – it’s not really well understood. - Doug Castro

We realised that we could spend years trying to understand what’s going on on a physics level,” Doug says, “or we could automate it. We just run signals through an amp – we don’t need to understand the effect of each part of an amplifier. If it has an impact on the signal, we should be able to recreate it. So that’s the advantage of black-box modelling like this.

What is responsible for that is the energy transfer from the amp’s power supply, through the power tubes, through the output transformer, out into the reactive load, the speaker, and the elements of the power supply itself – the capacitance of the power supply. All of these things have these super-weird effects on the guitar signal that are noticeable and perceivable. I don’t think this is very well understood to this day, and I don’t think we will ever have good science done on this because it is just not profitable for anyone to research the output transformer. The companies that have world-class modelling figure it out somehow and that is part of their secret sauce. Our approach does not care because it is just measuring things and based on what we measure the model writes itself. If you can hear it and feel it, our model will take that into account, so we don’t need to understand it to recreate it. - Doug Castro

There’s some “details” on NDSP’s ML/modeling/whatever in this interview. Guitar.com is trying to give my ipad herpes, so be prepared. https://guitar.com/features/interviews/neural-dsp-doug-castro-machine-learning-amplifiers/

 

[MirrorProfiles]

interview with one of the machine learning guys from NDSP, it’s quite interesting:

 

[Ed DeGenaro]

You know, I never thought of it that way before; it makes a lot of sense.

All of this new technology being used to simply copy the results of old technology. But it’s not actually capable of creating anything original or advancing guitar amplification. It can only imitate what already exists.

That would mean the only way to continue advancing guitar amplification would be to continue creating new physical amps so they could be copied in the digital realm. While modeling allows the ability to actually create new amps entirely within the digital realm.

The advancing thing strikes me as weird, look around here the majority of guys here want tones from 3-4 decades ago...period.

I mean how long did it take for guys to get used to Line6 or Fractal originals.

And while I'm at it the frontend whether hardware or plugins is constant irregardless of model with the only change on stuff like Fractal orUA Apollo being the input impedance.

Just as having an impedance curve for the power amp and speaker interaction is all fine and we'll except it's frequency dependant rather than a consistent curve.

 

[BenIfin]

I'm almost certainly in the minority here ... but I think the progressive and productive future is for more "modeler makers" to create their own versions of great Amps and ... to use a scientific-technical term ... "de-crap" them ...ie: as I see it, the goal for the leading companies has so far been to get their models as perfect as possible including the warts and all.

This is one of the reasons I genuinely admire Boss's and L6's approach with their "original" Amps in the GT and L6 series .. whilst they are loosely clearly based on real Amp designs, they have deliberately and upfront de-warted them ...... you may love or hate them, but they are a total no brainer to dial in.

The problem as I see it is .... there is much more money in selling a "supposedly perfect digital copy" of a real-old-classic Amp ... as opposed to offering something "different" or possibly "newer" or possibly "better in terms of ease of dialling in".

Tube Amps and old guitars .... is there a more intransigent, set in stone, everything new is sh%t "market" / ownership group (?)

My sole aim / wish is build a great tone that's quick and easy to dial in, and that sounds great and individual to me ..... I couldn't give a flying-f%ck if that's from a supposedly-perfect Amp model or a L6 or Boss "original" ... again .... hence why I am doubtlessly in the minority.

And to be transparent, .... I deliberately didn't mention, but am very well aware of the "FAS" models, from when I had all my Fractals ... and tried them all ..... great models no doubt, but they are a different kettle of fish to what I am talking about.

The future ?

This is hard / ugly.

If in 2023, Modeler X has an Amp that is %99 "there" ...... my guess is that when the next generation of Modeler X comes out, and that same Amp is then %99.2 "there", a lot / most non-TGP and non-TGF people are just going to be pretty non-plused - we should never forget that %99.99999999999 of the worlds guitar players have no idea what a TGP or TGF "is" ... and if asked .... they'd probably say they are acronyms for some sort medical conditions.

Factually, logically and objectively.....better is better because ... well .. its better ..... that is cool and everything ...... but we are and have been at the very pointy-point of hugely diminishing returns for a couple of years or more now.

Increasingly I think people are just going to say "I'm cool, my whites are white enough" / "I cant hear the difference" / "but your Helix Challenge [remember that] said and showed they were bang-on 6 years ago" or "Fractal said they were bang-on 4 Axe's ago" / "how f%cking much do you want for this miniscule improvement ?" / "don't you know I can buy Chinese brand Y for $600 and if you turn your back you the guitar player will never hear the difference" etc etc etc ......

Anyway .... rant over :)

Ben

 

[Lysander]

What is responsible for that is the energy transfer from the amp’s power supply, through the power tubes, through the output transformer, out into the reactive load, the speaker, and the elements of the power supply itself – the capacitance of the power supply. All of these things have these super-weird effects on the guitar signal that are noticeable and perceivable. I don’t think this is very well understood to this day, and I don’t think we will ever have good science done on this because it is just not profitable for anyone to research the output transformer. The companies that have world-class modelling figure it out somehow and that is part of their secret sauce. Our approach does not care because it is just measuring things and based on what we measure the model writes itself. If you can hear it and feel it, our model will take that into account, so we don’t need to understand it to recreate it. - Doug Castro

Pretty sure i noted this back when the interview came out, but, as someone with an EE background: this is absolute horseshit. ALL of the items mentioned here, and their interactions, have been exceedingly well understood for decades now.

Now, that doesn't necessarily mean those are easy to model, but stating that guitar amps work by arcane magic is... misleading.

 

[FractalAudio]

Pretty sure i noted this back when the interview came out, but, as someone with an EE background: this is absolute horseshit. ALL of the items mentioned here, and their interactions, have been exceedingly well understood for decades now.

Now, that doesn't necessarily mean those are easy to model, but stating that guitar amps work by arcane magic is... misleading.

Yup, pure unadulterated horseshit.

 

[Ed DeGenaro]

I'm almost certainly in the minority here ... but I think the progressive and productive future is for more "modeler makers" to create their own versions of great Amps and ... to use a scientific-technical term ... "de-crap" them ...ie: as I see it, the goal for the leading companies has so far been to get their models as perfect as possible including the warts and all.

This is one of the reasons I genuinely admire Boss's and L6's approach with their "original" Amps in the GT and L6 series .. whilst they are loosely clearly based on real Amp designs, they have deliberately and upfront de-warted them ...... you may love or hate them, but they are a total no brainer to dial in.

The problem as I see it is .... there is much more money in selling a "supposedly perfect digital copy" of a real-old-classic Amp ... as opposed to offering something "different" or possibly "newer" or possibly "better in terms of ease of dialling in".

Tube Amps and old guitars .... is there a more intransigent, set in stone, everything new is sh%t "market" / ownership group (?)

My sole aim / wish is build a great tone that's quick and easy to dial in, and that sounds great and individual to me ..... I couldn't give a flying-f%ck if that's from a supposedly-perfect Amp model or a L6 or Boss "original" ... again .... hence why I am doubtlessly in the minority.

And to be transparent, .... I deliberately didn't mention, but am very well aware of the "FAS" models, from when I had all my Fractals ... and tried them all ..... great models no doubt, but they are a different kettle of fish to what I am talking about.

The future ?

This is hard / ugly.

If in 2023, Modeler X has an Amp that is %99 "there" ...... my guess is that when the next generation of Modeler X comes out, and that same Amp is then %99.2 "there", a lot / most non-TGP and non-TGF people are just going to be pretty non-plused - we should never forget that %99.99999999999 of the worlds guitar players have no idea what a TGP or TGF "is" ... and if asked .... they'd probably say they are acronyms for some sort medical conditions.

Factually, logically and objectively.....better is better because ... well .. its better ..... that is cool and everything ...... but we are and have been at the very pointy-point of hugely diminishing returns for a couple of years or more now.

Increasingly I think people are just going to say "I'm cool, my whites are white enough" / "I cant hear the difference" / "but your Helix Challenge [remember that] said and showed they were bang-on 6 years ago" or "Fractal said they were bang-on 4 Axe's ago" / "how f%cking much do you want for this miniscule improvement ?" / "don't you know I can buy Chinese brand Y for $600 and if you turn your back you the guitar player will never hear the difference" etc etc etc ......

Anyway .... rant over :)

Ben

Yeah the sound thing hasn't been a thing in like15 years.

So I'm not sure that discussion happened outside Madison square bedroom.

But the warts and all, if i play a model that is supposedly a JMP then i want it to ghost, i want it to feel Like Wrestling a pig in mud when used without gain boxes in front.

But that has nothing to do with me using a purty overly compressed homogenized OTT gain tone or not. If need be I go there.

But i guranfuckingtee that this cottage industry will nit advance past what sells in terms of popular music.

Chugga chugga amps? An end result of Marshalls hit hard and we'd the snot out of in the studio created those amps.

And while I'm at it while Kemper spews idiotic statements about overtones in their manual it's clear that there's a major disconnect from guys like Red Rhoads, Dumble, Fisher, Rivera to guys twerking digital.

 

[DrewJD82]

Pretty sure i noted this back when the interview came out, but, as someone with an EE background: this is absolute horseshit. ALL of the items mentioned here, and their interactions, have been exceedingly well understood for decades now.

Now, that doesn't necessarily mean those are easy to model, but stating that guitar amps work by arcane magic is... misleading.

I know I had a field day with it, specifically the transformer line, because I was watching a ToneTalk and Friedman and I think George from Metropolis were geeking the fuck out on transformers for about 30 minutes as I was reading it. It was one of those "Ok, let's move on before we bore everyone" talks because they were getting deep enough into it even gear nerds stop caring at a certain point

 

[Foxmeister]

Despite my belief that if possible, algorithmic is better, sampling does have some merit too. You aren’t approximating the circuit, it IS the circuit.

I'd say component level modelling is the circuit, and profiles/captures are just emulating, or playing back, the output of that circuit for a given input.

 

[SeeD]

Guy said the Uber and Recto models sounded like the real amp, that is all that counts. NDSP said they don`t need to know everything that happends inside an amp to make a good model of it. I would say that is kind of cool
I love to be able to capture stuff, and i`m not alone thinking just that.
I hope Fractal gets into the capture stuff in the near future, as i`m sure that Cliff will do a good job = best of both worlds

It`s just fun to be able to capture/profile stuff

 

[Mongillo19]

Guy said the Uber and Recto models sounded like the real amp, that is all that counts.

I've never had an Uber but the QC Recto does not sound like a MW Recto at the same settings. I believe the guy you are referring to stated the Uber was 1:1, but went on to say that some amps are tricky to dial in to not sound bad (like the recto). Would be curious for clarification but no, to me the recto does not sound like a MW Recto on the QC having compared both.

I think this video highlights it well. Captures are where it's at imo

 

[SeeD]

I've never had an Uber but the QC Recto does not sound like a MW Recto at the same settings. I believe the guy you are referring to stated the Uber was 1:1, but went on to say that some amps are tricky to dial in to not sound bad (like the recto). Would be curious for clarification but no, to me the recto does not sound like a MW Recto on the QC having compared both.

Did NDSP model the amp you have ?
I have now heard a lot of different recto`s, some sound good ..fluffs sounds not so good
I did make mirrorProfiles recto to sound 95% the same with making a capture of a Nam capture he made of his amp, that is kind of cool and it made me laugh out loud, I love the QC

 

[Mike Oliphant]

There are a variety of reasons I'm opposed to all this profiling/ML/AI/etc. stuff:

To be fair, you have a pretty strong personal investment in the alternative.

I love the fact that any random person (without an electrical engineering degree or any computer programming skill) can now create realistic captures of their amplifier.

Am I "opposed" to modeling? Of course not. But ML capture technology is giving us an alternative way to accurately duplicate the amplifier sounds we love without needing access to complicated, painstakingly crafted bits of software that you and other companies develop behind closed doors and keep to yourselves.

 

[Achilles]

can now create realistic captures of THEIR amplifier

This is a really big point. Had a toaster for about a year before I bought a couple of tube heads. Did DI profiles of both and they came out sounding ridiculously close to what I was already very familiar with.

By comparison, look at the chatter about a certain Rectifier model today. One person is saying it sounds dead on while another is saying no way.

 

[Mongillo19]

One person is saying a certain model sounds dead on while another is saying no way.

Such is the way of life

Where are my graphs at

 

[Mongillo19]

Did NDSP model the amp you have ?

No idea what model they used!

 

[MirrorProfiles]

One person is saying a certain model sounds dead on while another is saying no way.

user error/variables are a massive factor. to get the NDSP model to match my amp (different revision/circuit), I had to lower the MV and bass a LOT and also find the most similar sounding load. Using a different version of the amp, at similar settings and a random load being used, its unlikely they'll sound similar. I think its the reason why most people think ANY modeller or plugin sounds different to their amp (along with input levels). Change the load being used and a rectifier will sound totally different.

I've been very guilty of this in the past and its why I try to check all of these factors before assuming there's a mistake or problem.

 

[Achilles]

user error/variables are a massive factor. to get the NDSP model to match my amp (different revision/circuit)

Kicking myself that I never tried doing it with the QC I had for almost a year.
I remember the DI's being pretty easy with Kemper.