NAM: Neural Amp Modeler

[mhelin]

Yes - you can use checkpoint-based training to resume training from a given checkpoint allowing for such tweaks. It's a bit niche but can be done.

Ok, I see Steve implemented NAM import in some of the latest versions so you can now also open a NAM file instead of checkpoint file. So you could download a NAM file from https://www.tone3000.com/ and use it as starting point for your own capture I guess. Wonder how much time that would save and what would the results be when compared to full training using you own data set...

 

[2dor]

Ok, I see Steve implemented NAM import in some of the latest versions so you can now also open a NAM file instead of checkpoint file. So you could download a NAM file from https://www.tone3000.com/ and use it as starting point for your own capture I guess. Wonder how much time that would save and what would the results be when compared to full training using you own data set...

You're still going to need the wav files as well. As for the time things take to converge - can't generalize; it's going to vary from case to case

 

[2dor]

I put together a quick video to help users get the best experience while using calibrated profiles:



EDIT: I must have fat-fingered something & the first vid got rendered at a crap resolution; I've uploaded a new take
EDIT2: audio's fixed now

 

[JiveTurkey]

 

[MirrorProfiles]

Just to save myself needlessly running GPU cycles to test with.....

xSTD with 0.0064 and 0.0023 learning and decay rate seems to get about 1/2 the ESR of default NAM training settings. So it's a good bit more accurate as far as ESR numbers go. Is this down to the tweaks to the code, or by altering the learning and decay rate?

 

[2dor]

Just to save myself needlessly running GPU cycles to test with.....

xSTD with 0.0064 and 0.0023 learning and decay rate seems to get about 1/2 the ESR of default NAM training settings. So it's a good bit more accurate as far as ESR numbers go. Is this down to the tweaks to the code, or by altering the learning and decay rate?

It's actually a combination of both. You can tweak the architecture but that alone may not render the best results by itself. Parameters such as learning rate can have quite a noticeable effect: it tells the neural net what magnitude of adjustments it is allowed to make to its weights to match the output. Too high of a learning rate, the training can vary widely & not be able to converge to a good ESR. The decay cuts that learning "amount" defined by the learning rate with each epoch. Essentially, as your training progresses, the network will end up only making very fine tweaks overall to help it converge.
Usually when you create a new architecture or augment the training signal, the learning rate & decay need to be tweaked at the very least.

There's also the NY parameter which defines how many samples of audio thr trainer looks at. By default it's abou 170ms .. tweaks to it almost always require changes to the architecture/dilations at the very least.

 

[paisleywookiee]

Fellow NAM Nerds (now that I've downed the whole gallon of Kool-Aid) is the Tonex Capture box a good solution for getting head-only caps? I want to go fucking nuts with the amps at my buddy's shop.

 

[Mongillo19]

Fellow NAM Nerds (now that I've downed the whole gallon of Kool-Aid) is the Tonex Capture box a good solution for getting head-only caps? I want to go fucking nuts with the amps at my buddy's shop.

Oh dang. I haven't been as active on here but when did you get into NAM? It's pretty great imo.

Sorry, this doesn't answer your question

 

[KingsXJJ]

It's actually a combination of both. You can tweak the architecture but that alone may not render the best results by itself. Parameters such as learning rate can have quite a noticeable effect: it tells the neural net what magnitude of adjustments it is allowed to make to its weights to match the output. Too high of a learning rate, the training can vary widely & not be able to converge to a good ESR. The decay cuts that learning "amount" defined by the learning rate with each epoch. Essentially, as your training progresses, the network will end up only making very fine tweaks overall to help it converge.
Usually when you create a new architecture or augment the training signal, the learning rate & decay need to be tweaked at the very least.

There's also the NY parameter which defines how many samples of audio thr trainer looks at. By default it's abou 170ms .. tweaks to it almost always require changes to the architecture/dilations at the very least.

Thanks for posting this! It’s very informational to folks getting their feet wet and wanting to come up to speed.

 

[Richard_G]

Fellow NAM Nerds (now that I've downed the whole gallon of Kool-Aid) is the Tonex Capture box a good solution for getting head-only caps? I want to go fucking nuts with the amps at my buddy's shop.

I believe several experienced NAM users - like @2dor and slamminmofo - use a setup like this:

Audio Interface Line Output -> Lehle P-SPLIT III reamp box -> amp head loaded down with high quality reactive load box -> Audio interface Hi-Z input

I’d do that instead of using a Tonex Capture box.

 

[Humbug]

Fellow NAM Nerds (now that I've downed the whole gallon of Kool-Aid) is the Tonex Capture box a good solution for getting head-only caps? I want to go fucking nuts with the amps at my buddy's shop.

It's not entirely clear to me what the Tonex box actually does, but if it allows you to insert it between the amp and the speaker/load to get a line signal it will work great*.

*As long as the circuit isn't stupid.

 

[paisleywookiee]

It's not entirely clear to me what the Tonex box actually does, but if it allows you to insert it between the amp and the speaker/load to get a line signal it will work great*.

*As long as the circuit isn't stupid.

It does.

 

[MirrorProfiles]

I believe several experienced NAM users - like @2dor and slamminmofo - use a setup like this:

Audio Interface Line Output -> Lehle P-SPLIT III reamp box -> amp head loaded down with high quality reactive load box -> Audio interface Hi-Z input

I’d do that instead of using a Tonex Capture box.

+1

A Lehle for the reamp box, and a DI box (or reamp box) of your choosing would be my advice.

If the amp already has a slave output, you can just run that straight to your interface DI. Just make sure there is a cab or load connected.

IMO a lehle p-split and a used Suhr Reactive load would be way better money spent and give you way better results as well as more flexibility.

 

[Humbug]

It does.

Seems like a pretty good all-in-one for a reasonable price to me, then. I have no idea what the build quality is like or if the circuit inside is, in fact, stupid, though. Some of the marketing speak on their website worries me a little

 

[Chocol8]

I have no experience with the Tonex box, but it looks to me like a reasonable attempt to put everything into one device. That doesn't necessarily mean quality or good value, and I think it may have limitations on levels.

If you already have something to use as a DI, like the Suhr RL, or whatever, the P Split makes more sense and the combo will likely be a little better quality wise. Even if you don't, I think a good load like the Suhr or the newer Two Notes Reload would have significant value for other uses.

 

[Chocol8]

reactive load box -> Audio interface Hi-Z input

That should depend on the load box output, but if it is a line out, why go to a hi-Z input?

 

[Humbug]

That should depend on the load box output, but if it is a line out, why go to a hi-Z input?

If the line out is not sufficiently low impedance a high Z input is preferable. I believe the Suhr box is an example of this.

 

[MirrorProfiles]

If the line out is not sufficiently low impedance a high Z input is preferable. I believe the Suhr box is an example of this.

Think it depends on which Suhr. I’ve always run my non-IR one straight to A/D with no issues. Same with React IR and Fryette PS. I know some Suhr units work better going into a DI though

 

[Chocol8]

If the line out is not sufficiently low impedance a high Z input is preferable. I believe the Suhr box is an example of this.

Makes sense for the regular RL possibly, but the RL-IR has a powered low impedance output. I think the regular Suhr is probably the exception, but I am not familiar with all the popular DI boxes out there. But yes, if you don't have a low impedance line out that's an option.

 

[Humbug]

Think it depends on which Suhr. I’ve always run my non-IR one straight to A/D with no issues. Same with React IR and Fryette PS. I know some Suhr units work better going into a DI though

Iirc the output impedance on the Suhr is pretty high for a "line" output. The effect will depend on the interface line level input impedance, though. 10K might be fine, 2.2K maybe less so.