NAM: Neural Amp Modeler

yeky83 said:
Wouldn’t be as accurate. NAM’s whole point is accuracy, so I don’t see the point.
Click to expand...
What wouldn't be as accurate?

and I think a large purpose of NAM is its ability to evolve and support different approaches (much like is going on now with various aspects of it). I think if seperate preamp and poweramp captures are optimized, then in many situations it has a lot of advantages.
 
MirrorProfiles said:
What wouldn't be as accurate?

and I think a large purpose of NAM is its ability to evolve and support different approaches (much like is going on now with various aspects of it). I think if seperate preamp and poweramp captures are optimized, then in many situations it has a lot of advantages.
Click to expand...
Separate preamp + power amp models, rather than capturing/modeling it as one.
 
yeky83 said:
Separate preamp + power amp models, rather than capturing/modeling it as one.
Click to expand...

Accuracy is only relevant if you have an "ideal" reference. If we're talking about mixing / matching pre and power amps, and the captures of each are accurate, and we're not doing anything crazy with gain staging, then I'd stop worrying about it and just consider what sounds good. You'd never be doing all the mixing/matching IRL in most cases anyway. Same reason you go about deep tweaking the axe FX even though it makes it less "correct".

If the capture combination you've done initially matches the original setup with good accuracy, then it's as accurate as you can hope for.

NAM doesn't necessarily have a whole point of accuracy. It just happens to be extremely accurate when used correctly - but the goal is really just to build/explore/develop.
 
Last edited:
yeky83 said:
Separate preamp + power amp models, rather than capturing/modeling it as one.
Click to expand...
Is it proven that it isn’t accurate? That’s what I’m trying to determine now, and maybe if the results aren’t good it can be worked on. There’s no reason why it can’t be done
 
A quick note, capture the maximum drive from the poweramp you would actually use.
In my experience captures lose accuracy if you drive them above the maximum of the training signal.
 
northern_fox said:
Accuracy is only relevant if you have an "ideal" reference. If we're talking about mixing / matching pre and power amps, and the captures of each are accurate, and we're not doing anything crazy with gain staging, then I'd stop worrying about it and just consider what sounds good. You'd never be doing all the mixing/matching IRL in most cases anyway. Same reason you go about deep tweaking the axe FX even though it makes it less "correct".

If the capture combination you've done initially matches the original setup with good accuracy, then it's as accurate as you can hope for.

NAM doesn't necessarily have a whole point of accuracy. It just happens to be extremely accurate when used correctly - but the goal is really just to build/explore/develop.
Click to expand...
If you just want to build/explore/develop sounds and don’t care about accuracy, there’s better tools for that. Dunno what the point of using NAM would be.
MirrorProfiles said:
Is it proven that it isn’t accurate? That’s what I’m trying to determine now, and maybe if the results aren’t good it can be worked on. There’s no reason why it can’t be done
Click to expand...
Yes. An integrated head/combo amp has pre & power amps sharing the power supply, and there’s subtle effects due to that. There may also be parasitic effects as well as impedance relationships that don’t get captured if you just capture them as separate blocks and put them together afterwards.

Not saying that it can’t be done or the results won’t be good. But it’s not going to be as accurate. And if one doesn’t care about accuracy, why use NAM…
 
yeky83 said:
If you just want to build/explore/develop sounds and don’t care about accuracy, there’s better tools for that. Dunno what the point of using NAM would be.
Click to expand...
What better tools in plugins are there for capturing the behaviours of my amps? Even if there is a trade off of accuracy, the benefit of saving time and having more options available quickly may well outweigh that in a lot of situations. Fractal has nice options but the workflow is awful compared to using multiple plugin instances in a DAW. There also several amps (included modded stuff) that aren’t commonly modelled in plugins.
yeky83 said:
Yes. An integrated head/combo amp has pre & power amps sharing the power supply, and there’s subtle effects due to that. There may also be parasitic effects as well as impedance relationships that don’t get captured if you just capture them as separate blocks and put them together afterwards.

Not saying that it can’t be done or the results won’t be good. But it’s not going to be as accurate. And if one doesn’t care about accuracy, why use NAM…
Click to expand...
It won’t capture the power amp sagging or parasitic interactions but I can’t think of a single instance in my life where I’ve ever missed these behaviours. It varies amp to amp, and it’s importance to a tone is extremely context dependent.

If they’re that essential for a sound, then I can make another model of the whole system, use another modeller, or just use the amp itself. There’s always a trade off and in any situation you pick the ones that are least important. That will vary case to case, so each approach has its merits. Even if NAM isn’t there yet, the main thing it has in its favour is it’s always being developed/explored/updated. If these behaviours aren’t explored, then they won’t get worked on and improved.
 
Given the number of people making models with duff sounding load boxes, or too much poweramp distortion, then I think being able to salvage preamp tones is going to be very useful.

In the same way that direct models are much more useful even though you trade off having the correct cab+impedance behaviour when mixing and matching IR’s and different loads. That’s arguably a bigger trade off IMO, but again, often something that is still worth it.

Either way - every approach has its own pro’s and con’s. There is no plugin solution that doesn’t involve some form of compromise, and different situations dictate which compromises are acceptable. Seems odd to totally discard NAM’s ability to do something that hasn’t been fully tested (or developed yet).
 
Last edited:
I agree with Ola that it's going to put some pressure on established companies.
NAM is only going to get better, more options, more customization, soon enough cheap hardware players will appear and take a big chunk of the market.

This train is rolling, the smart ones will jump on it early.

 
James Freeman said:
I agree with Ola that it's going to put some pressure on established companies.
NAM is only going to get better, more options, more customization, soon enough cheap hardware players will appear and take a big chunk of the market.

This train is rolling, the smart ones will jump on it early.

Click to expand...


this made for a busy day :D
 
NAM trainer v0.6.0 was released 2 days ago.
Online easy_colab was also updated.
A lot of tweaks and fixes, new v2_0_0.wav training file, full rig chain (amp+cab+mic) training is now supported,

github.com

Release Version 0.6.0 · sdatkinson/neural-amp-modeler

What's Changed Only update metadata flag after batch by @vossenv in #205 New reamping file v2_0_0.wav by @sdatkinson in #217 CPU-friendly training parameters by @sdatkinson in #219 Strong and weak...
github.com github.com

Worth redoing some captures?
 
pretty sure v2_0_0.wav just warns you if there is a problem with the reamp file so you don’t waste training on something that will turn out badly.

main improvement is for mic’d tones. not something i’m particularly bothered about, it just gives more models to sift through when trying to find what you need. Much more beneficial to find the right IR than find the right amp settings AND cab+mic in one IMO
 
You must log in or register to reply here.
Back
Top