Ten billion profiles/captures = one amp

[Mongillo19]

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

Never too late to snag one!

 

[MirrorProfiles]

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

Since getting an Ampete Amp Switcher its made it dead simple to A/B and change loads on the fly and hear the influence they have. I dont have a QC and have never really spent more than about 15 minutes on one at any point. Been very easy to match the tones of my 2203/Uber/Recto/Friedman though

 

[Achilles]

Never too late to snag one!

Falls under 'been there, done that'. Plus still waiting on that carrier pigeon from the Ukraine.

 

[yeky83]

it seems NDSP split their preamp and poweramp models which must cut a lot of permutations down.

It wouldn't though.

Guessing they have multiple machines working on it and probably several amps on the go at once.

Surely we'd see it plastered all over their social media if this were so... dunno why folks are so intent on making their outlandish claims work somehow. Ignoring the problem of amp-to-amp variance, let's say they have 10(!) six-knob amps of the same model going at it at once, it lowers their learning time to 5 months.

 

[MirrorProfiles]

It wouldn't though.

Yes it would. Doing every permutation of every setting in one model would require more models than doing the power amp seperately to the preamp.

For 6 controls sampled at 10 settings each it’s 10^6. If you did (say) 4 controls and 2 for the power amp it’s 10^4 + 10^2. WAY smaller number

 

[yeky83]

Yes it would. Doing every permutation of every setting in one model would require more models than doing the power amp seperately to the preamp.

For 6 controls sampled at 10 settings each it’s 10^6. If you did (say) 4 controls and 2 for the power amp it’s 10^4 + 10^2. WAY smaller number

Oh you're saying they probe inside to separate out the pre & power amp. Why have a knob robot at all then? Why not just probe and capture pre-tonestack, post-tonestack, and power amp all separately, then just model the tonestack in?

 

[Byrdman]

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.

Although I do agree somewhat, but Capturing is not the be all either IMO. Just taking Tonex and NAM for example, because that is what I have experience with. Decent UI experience is pretty lacking, adjusting parameters of the capture is not really possible. The end of the sound chain, speakers, headphones, amp, etc., can make the same capture golden or total garbage. Point is that there are too many variables that can populate in the equation and as shown in other posts, those variables plus the math required to make up the difference is too great with today's technology.

Personally, and I mentioned at TOP before Tonex was even released, once real AI technology gets involved in modeling and captures, all bets are off. It may take 5-10 years, but eventually, machines will be able to make something so close that human hearing will not be able to tell the difference and the machines will know our physical limitations and laugh at us for those limitations. It's going to be a lot different over the next century or two, IMO.

 

[MirrorProfiles]

Oh you're saying they probe inside to separate out the pre & power amp. Why have a knob robot at all then? Why not just probe and capture pre-tonestack, post-tonestack, and power amp all separately, then just model the tonestack in?

I’m not claiming any of that stuff. Several people have posted from looking in their code that their models are separated into preamp and poweramp. Mercuriall use ML for their power amp models (who also split preamp and poweramp).

I think it’s fairly typical of any company to try and isolate sections of the circuit that don’t interact with other parts, it makes a more efficient model no matter what approach they are using. Dead simple on any amp with an fx loop to split preamp and poweramp. Companies like UAD and Softube often measure circuits as different “blocks” and use various techniques for different parts of the circuit. Whatever amp (or piece of gear) is being modelled, it makes sense to study the circuit and approach it in the most logical way.

 

[yeky83]

I’m not claiming any of that stuff. Several people have posted from looking in their code that their models are separated into preamp and poweramp. Mercuriall use ML for their power amp models (who also split preamp and poweramp).

I think it’s fairly typical of any company to try and isolate sections of the circuit that don’t interact with other parts, it makes a more efficient model no matter what approach they are using. Dead simple on any amp with an fx loop to split preamp and poweramp. Companies like UAD and Softube often measure circuits as different “blocks” and use various techniques for different parts of the circuit. Whatever amp (or piece of gear) is being modelled, it makes sense to study the circuit and approach it in the most logical way.

Yes, all that is fine, but changes nothing about how the whole NDSP capture knob robot story doesn't add up. 10^4 + 10^2 still makes for 14 days for instance.

 

[MirrorProfiles]

Yes, all that is fine, but changes nothing about how the whole NDSP capture knob robot story doesn't add up. 10^4 + 10^2 still makes for 14 days for instance.

Yeah hard to know what they’re doing or whether it’s constant for every model. Presumably it’s about identifying the bits of the circuit that makes the most sense for a particular approach.

The difference between 10^6 and 10^4 + 10^2 is 989,000 (or approx 10^6), so any kind of optimisations like that can save a huge amount of time.

 

[yeky83]

The difference between 10^6 and 10^4 + 10^2 is 989,000 (or approx 10^6), so any kind of optimisations like that can save a huge amount of time.

I find it really neither here nor there since both are nowhere close to their supposed 8hrs or whatever.

 

[northern_fox]

Have not read this thread yet

However, IMO it doesn't matter. When's the last time you used 10 billion variations of a tone? If you have a sound you want, you don't necessarily need all the knobs to work. Also, parametric training isn't actually the same thing as a million static captures that you're flipping between.

If your goal is simply to recreate a sound identically, that's good enough for me. But neither is objectively 'better' in the sense of how this poll is phrased - at least not yet. Both are just tools in the end.

 

[Lysander]

If you have a sound you want, you don't necessarily need all the knobs to work.

Yeah, but you have to have it to begin with. Profiles/captures really suck when it comes to tweaking settings.

Both technologies have their place; i'm legit surprised when i read people asking Line 6 or Fractal to implement profiling as if what they do were somehow inferior.

 

[Mongillo19]

Line 6 or Fractal to implement profiling as if what they do were somehow inferior.

Maybe they are phrasing it as they want the additional option? Or maybe they have a tube amp they really like but it's not in the fractal or doesn't sound the same?

 

[northern_fox]

Yeah, but you have to have it to begin with. Profiles/captures really suck when it comes to tweaking settings.

Both technologies have their place; i'm legit surprised when i read people asking Line 6 or Fractal to implement profiling as if what they do were somehow inferior.

Yes, but if you have gear you love then presumably you do already have it. That's the best use case for me anyway - I tweak the real gear and capture what I like. IMO tweaking is over-rated, but that's a matter of preference. Mostly using post EQ and IR's will get me what I want regardless as long as the starting point is reasonably close.

The reason I'd love to see a loader or something in FAS isn't because amp modeling is inferior, it's just because I'd like to use my captures along with all of the other things the FAS system has to offer. Plus the bonus of using anything you want, vs what's in the box. Clearly I'm biased here.. but add a NAM loader - no need to implement training tech.

 

[Achilles]

i'm legit surprised when i read people asking Line 6 or Fractal to implement profiling as if what they do were somehow inferior.

Maybe they are phrasing it as they want the additional option?

Bingo.
If you have other boxes on the market that are doing both modeling and profiling/capturing then it's natural to wonder why everyone else isn't also doing both. It's all about having more options - especially if like Mongillo says, you've got some priced hardware you'd love to turn digital.

 

[Boudoir Guitar]

If you have an amp with 10 knobs, no other switches, channels etc and you ‘capture’ every permutation of the knob settings you can capture the ‘whole’ amp.
Then you have a control ‘model’ that lets you adjust 10 virtual knobs to dial in the amp sound you like.
and you have some kind of machine learning to interpolate the missing data between the captured states of the amp.

Do you then simply have a top quality amp simulation? Do you have something better?

You have a box you can plug a guitar into, listen, and decide for yourself whether or not you like the sound it makes.

 

[Enchilada Jones]

...
If your goal is simply to recreate a sound identically, that's good enough for me. But neither is objectively 'better' in the sense of how this poll is phrased - ...

Yea it was really me getting to vent a bit of snarky commentary. I didn't set out to create a poll when I started the post thus I made the questions inappropriate for the answers kind of like I find the arguments on both sides of the debate to be. Sometimes my improv skills are mediocre lol

 

[SeeD]

Yeah, but you have to have it to begin with. Profiles/captures really suck when it comes to tweaking settings.

How many `sounds` does a jcm800, uber, fender have? Just thinking about the amp here ( IR`s will do more to the sound )
Sure a Mark V will have loads of things to tweak, but even the modes on a Mark amp `sounds the same` to my ears, but it does something for sure

 

[Mongillo19]

How many `sounds` does a jcm800, uber, fender have? Just thinking about the amp here ( IR`s will do more to the sound )
Sure a Mark V will have loads of things to tweak, but even the modes on a Mark amp `sounds the same` to my ears, but it does something fore sure

Yeah tbf I load up a good JCM 800 capture and do I really need to tweak it more?

I will disagree with the mark series. Some of their switches do drastically change the tone (especially the presence shift) and then Friedman's with BE HBE