NAM: Neural Amp Modeler

[northern_fox]

NAM is open source. I'm not sure which license, but I'm pretty sure anyone can fork it and develop their own version as they please.

Its MIT and you are correct - anyone can use it. It's most definitely not about money

 

[2dor]

It's very long and wordy, but I think the tl:dr is "Steve Ack didn't do the very specific thing I wanted, so it's all about the money".

NAM is open source. I'm not sure which license, but I'm pretty sure anyone can fork it and develop their own version as they please.

Wut?!

 

[Audio Transformer]

Its MIT and you are correct - anyone can use it. It's most definitely not about money

Not about money? Open source doesn’t mean you don’t make money Most of the time the creator of the main repo can sell many services, implementing, helping teams to understand and help for project. That’s why NAM is NAM.com . Also it’s about ego, Steve wants NAM to become and industry standard like .wav for IRs…! That’s why he doesn’t want any fast moving innovations and that why Francois Neural and his team aren’t interested in helping for better sound in main project.

 

[MirrorProfiles]

The aliasing talk recently seems very odd to me, especially when accusations are being made about Steve and his motives. Steve has done an unbelievable job of being impartial and open and generous with what he does.

It’s not the first time I’ve seen people talk like they expect Steven to do their work for them for free. If someone has certain criteria they need met, anyone is welcome to take NAM in whatever direction they like, and they’re also free to make their work open source too. Expecting Steven to turn the entire project into their own personal vanity project is purely entitlement.

 

[BenIfin]

3 Days old but still interesting read about how S.A see's things going forward:-

https://www.neuralampmodeler.com/post/stewarding-nam-as-an-industry-standard

It reads like he's specifically "aiming his comments" at someone or a group somewhere that are wanting to move things in a different direction ? ... like he's almost "re-asserting" his "control" over the main project ?

And before anyone accuses me [again] of sh%t-bagging on NAM - I've been using NAM Captures in GigFast Lite in AUM Mixer on my IPad Air M1 for 2 months now :)

 

[DLC86]

Francois Neural

Why are you referring to yourself as if it was a different person?

 

[2dor]

Not about money? Open source doesn’t mean you don’t make money Most of the time the creator of the main repo can sell many services, implementing, helping teams to understand and help for project. That’s why NAM is NAM.com . Also it’s about ego, Steve wants NAM to become and industry standard like .wav for IRs…! That’s why he doesn’t want any fast moving innovations and that why Francois Neural and his team aren’t interested in helping for better sound in main project.

I think you need to take a chill pill buddy. Steve shared his work, free of charge, for anybody out there to use.

Then vendors picked it up and were able to spin it into their solutions (Two Notes Codex, Tonocracy, Dimehead etc).

It's free & available for everybody to do whatever on Earth they want from it.

Just as an FYI, Steve turned down donations from early-adopters (I was there on that FB group day 1 when it got created) so what you're saying there is straight up false & you should take it back; it's rude & out of touch with reality.

If NAM does go on an become an industry standard, and I hope this actually happens, then it's because it brought innovation to this space.
Sure, everything can be improved - look at the numerous updates FAS' Cliff is churning out - and Steve's showed, time & time again, that he cares about improving NAM's:

- training & accuracy (input /output calibration metadata & functionality as well as auto alignment calibration just recently added / tweaked etc)
- ease of use
- codebase for wider adoption & helping devs out

 

[[Nathan]]

3 Days old but still interesting read about how S.A see's things going forward:-

https://www.neuralampmodeler.com/post/stewarding-nam-as-an-industry-standard

It reads like he's specifically "aiming his comments" at someone or a group somewhere that are wanting to move things in a different direction ? ... like he's almost "re-asserting" his "control" over the main project ?

And before anyone accuses me [again] of sh%t-bagging on NAM - I've been using NAM Captures in GigFast Lite in AUM Mixer on my IPad Air M1 for 2 months now :)

I mean it all makes sense to me as a casual observer. NAM can be pushed and pulled in different ways and I guess some people will want ultralight captures to run on $30 hardware, some will want 13minute training files to reduce aliasing and others might want more complex models for the highest accuracy possible meaning It needs more horsepower to run the NAM files on playback.

I just see it as Steve going… all things considered the way it is now is pretty balanced and if you want the wider community to keep building with NAM it’s detrimental to start making sharp turns to accomodate these extreme changes. Anyone can take a copy of NAM and make their own.

 

[laxu]

3 Days old but still interesting read about how S.A see's things going forward:-

https://www.neuralampmodeler.com/post/stewarding-nam-as-an-industry-standard

It reads like he's specifically "aiming his comments" at someone or a group somewhere that are wanting to move things in a different direction ? ... like he's almost "re-asserting" his "control" over the main project ?

I see his comments as very reasonable rather than trying to take control over the project. While things that improve the performance of NAM by whatever metric are welcome, they're not the only consideration.

Perhaps the best route forward would be offering different performance tiers that e.g hardware vendors can target, then make the capture tools always just generate all those tiers. This avoids that Tonehunt for example is not full of only "hyper accuracy max tier" models that then don't work on any current or future NAM player hardware.

As an example you have a standard tier, then a super accurate one that prioritizes better aliasing etc.

 

[laxu]

As for "being just about money"...doesn't Steve work at one of the FAANG companies? I would assume that means he's well paid and doesn't exactly need to turn NAM into any sort of business.

 

[paisleywookiee]

Not about money? Open source doesn’t mean you don’t make money Most of the time the creator of the main repo can sell many services, implementing, helping teams to understand and help for project. That’s why NAM is NAM.com . Also it’s about ego, Steve wants NAM to become and industry standard like .wav for IRs…! That’s why he doesn’t want any fast moving innovations and that why Francois Neural and his team aren’t interested in helping for better sound in main project.

 

[EOengineer]

Ah the good ole days where we could kick our feet up, hit the gear forums, and…argue about the direction of open source software.

 

[jellodog]

Why are you referring to yourself as if it was a different person?

Who is this Francois guy, then?

 

[2dor]

Who is this Francois guy, then?

A member of the NAM community on FB who is very passionate about reducing aliasing.

He goes to great lengths to test and discover new ways of getting this done.

The methods involve augmenting the training signal with additional signal/sweep patterns. These, however, may be risky on some equipment as well as extending training times.

Some challenged this because the training overhead & accuracy gains are somewhat bordering on the point of diminishing returns since most can't distinguish a standard NAM profile from one derived by means of these new anti-aliasing methods.

Apparently some people are very bothered by aliasing; personally I find that most NAM models turn out great with a solid reamp chain.

 

[jellodog]

A member of the NAM community on FB who is very passionate about reducing aliasing.

He goes to great lengths to test and discover new ways of getting this done.

The methods involve augmenting the training signal with additional signal/sweep patterns. These, however, may be risky on some equipment as well as extending training times.

Some challenged this because the training overhead & accuracy gains are somewhat bordering on the point of diminishing returns since most can't distinguish a standard NAM profile from one derived by means of these new anti-aliasing methods.

Apparently some people are very bothered by aliasing; personally I find that most NAM models turn out great with a solid reamp chain.

Thanks! Very useful in understanding the backstory here. There’s no way I’m ever going to use FB, so much appreciated

 

[paisleywookiee]

Ah the good ole days where we could kick our feet up, hit the gear forums, and…argue about the direction of open source software.

I remember that time when Steve was interviewed. Real smug SOB, had grand plans for NAM in everything, even refrigerators!












Oh wait. No. He's just a nerd like us, that loves cool guitar tech.

 

[Humbug]

I remember that time when Steve was interviewed. Real smug SOB, had grand plans for NAM in everything, even refrigerators!

 

[BenIfin]

I mean it all makes sense to me as a casual observer. NAM can be pushed and pulled in different ways and I guess some people will want ultralight captures to run on $30 hardware, some will want 13minute training files to reduce aliasing and others might want more complex models for the highest accuracy possible meaning It needs more horsepower to run the NAM files on playback.

I just see it as Steve going… all things considered the way it is now is pretty balanced and if you want the wider community to keep building with NAM it’s detrimental to start making sharp turns to accomodate these extreme changes. Anyone can take a copy of NAM and make their own.

I think you're bang on the money Nathan.

Its the same with the top tier hardware modelers - just take Fractal and L6.

I've got no doubt they could release, say, a next gen Axe 4 and Helix 2, with 16 or 32 core processors and algorithms that are 10x times more complex and 64x oversampling etc..... and in so doing "eliminate" digital aliasing and artifacts to the -120db range ie: .... utterly inaudible and functionally irrelevant to even measure.

But whose going to pay $5000 or $6000 dollars for one ? ... not too mention how much more involved and complex updates, new additions and bug fixing etc..... would be.

There *is* a point where it becomes pointless in the real world - and the problem only exists in the mathematical measurement world.

If people want to modify NAM and make training 5 x longer and files sizes 10 x bigger to get %2 - %3 better aliasing - good luck to them :)

NAM is already - compared to Tonex - a CPU hog - imagine how few people would even bother trying to run it at those levels ?

And then, when is enough enough - people will then no doubt want to make models that take 10 x longer to train and be 30 x bigger in file size to get another %1 better aliasing ?

 

[northern_fox]

And then, when is enough enough - people will then no doubt want to make models that take 10 x longer to train and be 30 x bigger in file size to get another %1 better aliasing ?

Also the extent to which aliasing is exhaustively discussed would make you think it was far more of an issue than it actually is. You'd think it was the bane of digital modeling when in fact these days its almost never making a significant enough difference for most people to notice or recognize as aliasing at all.

 

[Humbug]

Also the extent to which aliasing is exhaustively discussed would make you think it was far more of an issue than it actually is. You'd think it was the bane of digital modeling when in fact these days its almost never making a significant enough difference for most people to notice or recognize as aliasing at all.

This is true. At least in the test I did, the difference was subtle enough that if I didn't have the actual amp to test with I would probably not notice. And how much of the difference in fidelity was caused by a reduction in aliasing, and how much was caused by a reduction in ESR I don't know yet.

One day I _will_ have a perfect model of the D. B. Cooper amp!