• TalkBass has been independent since 1998. Add your voice.
    Create a free account to reply to discussions, view embedded media, and browse with fewer display ads.
    Join freeLog in
    Want zero display ads or expanded classifieds tools? Compare plans.

NAM (Neural Amp Modeler) big thread

Okay, I'm back. According to the trainer, my ESR is MUCH too high and I absolutely can't figure out why. I've tried to reduce gain and noise as much as I can, but if I turn it down any lower then I'm not getting complete waveforms because the output is too quiet.

EDIT: I have now tried 12 different times with no difference in results. The ESR slightly decreases to 0.25 as I turn my gain down while recording, anything lower than that and it tells me my output file is invalid OR the ESR spikes back up to 0.75+.
 
Last edited:
Okay, I'm back. According to the trainer, my ESR is MUCH too high and I absolutely can't figure out why. I've tried to reduce gain and noise as much as I can, but if I turn it down any lower then I'm not getting complete waveforms because the output is too quiet.

EDIT: I have now tried 12 different times with no difference in results. The ESR slightly decreases to 0.25 as I turn my gain down while recording, anything lower than that and it tells me my output file is invalid OR the ESR spikes back up to 0.75+.
How many epochs did you set?
 
How many epochs did you set?
I left it at 100, but have played around with going up as far as 500 and it did not help the problem. It did change the ESR very slightly, though I don't remember by how much.

Here's a more detailed bit of info I posted elsewhere (and forgot to add here after the fact) plus some added info I've learned since:
My setup is as follows:

Output of Audient id4 into Radial Reamp JCR's Input -> Output of Radial Reamp JCR into +0dB Input jack of Peavey Amp -> Line Out from Peavey Amp into Line In of Audient id4.

No matter what I do, the ESR is always much too high - the lowest I've gotten it is 0.257. I suspect the issue has to do with my gain settings, as the Audient interface has both an output and an input gain, and the Radial Reamp also has an output volume/gain knob. The amp ALSO has TWO pre- and post- gain settings, for a total of seven different gain knobs. Right now I'm walking a tightrope - if I turn the gain on any of these devices up too high, the output.wav clips near the very beginning, but if I turn any of them down too low, then the output file is so quiet that the trainer either does not recognize it as valid (presumably whatever it is listening for at the beginning is inaudible) OR the ESR shoots up to 0.75 or higher. I assume there would be a "sweet spot," but if there is, I can't find it, the closest I've gotten is 0.257.

I have confirmed that my file types are correct (24-bit 48Hz mono) and I am keeping all the settings "standard" as far as epochs, architecture, and latency, though changing "epochs" does not seem to make a difference. I am totally stumped. I can't get Tensorboard to display or I would post my graphs.

Any thoughts? Is this indicative of something happening with my gain settings? Do I need to record it as quiet as I can and then normalize the output.wav to unity gain?
 
I left it at 100, but have played around with going up as far as 500 and it did not help the problem. It did change the ESR very slightly, though I don't remember by how much.

Here's a more detailed bit of info I posted elsewhere (and forgot to add here after the fact) plus some added info I've learned since:
FWIW, before doing all my conversions, I've tested different numbers of epochs for the same input/output files, here are the results of this test:

Number of epochsESR
100.001531
1000.00025
5000.00002302
10000.00001886

I record the reamped file at a level that is slightly lower than the "input" file, then I normalize it to unity.


7 gain stages is a lot.

Maybe you've got some noise added somewhere in the chain?

I would first try with a simple bassline (instead of the "input.wav" file) and reamp it just to make sure you don't have any noise added somewhere.
 

Attachments

  • 10 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    10 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    60.1 KB · Views: 25
  • 100 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    100 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    56 KB · Views: 30
  • 500 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    500 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    54.3 KB · Views: 28
  • 1000 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    1000 epochs - Quilter Bass Block v803 Flat Qtube Gain 03.png
    54.2 KB · Views: 32
I think we haven't talked about the SONULAB StompStation pedal, a 279 € NAM loader.
They've just released the beta version of Magic Match, a software that lets you capture the tone of a song, and put it on the pedal.



 
In REAPER the NAM plugin GUI has stopped working, so NAM is dead for me at the moment, at least on Windows 11. The GUI for Amplitude 5 has also stopped working, so it's obviously something more fundamental than NAM.
 
I think we haven't talked about the SONULAB StompStation pedal, a 279 € NAM loader.
They've just released the beta version of Magic Match, a software that lets you capture the tone of a song, and put it on the pedal.



There's talk on another forum about how much better this sounds than even the NAM plugin, which is wild if true.

Strongly considering this. And the Dimehead. But only as a temp option until the Helix Stadium drops.
 
There's talk on another forum about how much better this sounds than even the NAM plugin, which is wild if true.

Can you share a link?

Strongly considering this. And the Dimehead. But only as a temp option until the Helix Stadium drops.
I don't think it's comparable to NAM, in terms of quality and in terms of use. NAM get much closer to the source IMO. And more importantly NAM captures an amp that you have with you, whereas Magic Match capture the tone of an instrument in a song.

I've tested Magic Match and made a demo video in the other thread:

 
Can you share a link?


I don't think it's comparable to NAM, in terms of quality and in terms of use. NAM get much closer to the source IMO. And more importantly NAM captures an amp that you have with you, whereas Magic Match capture the tone of an instrument in a song.

I've tested Magic Match and made a demo video in the other thread:


I am HOPING the Helix capture tech will be at least as good as NAM. Because there's a LOT of cool features it has that I'd use.

Here's a link. It's responding to a video which is also at that link

Post in thread 'Stompstation by Sonulab (new NAM player pedal)'

I'll check your video as well!
 
I am HOPING the Helix capture tech will be at least as good as NAM. Because there's a LOT of cool features it has that I'd use.

Here's a link. It's responding to a video which is also at that link

Post in thread 'Stompstation by Sonulab (new NAM player pedal)'

I'll check your video as well!
I think you're mixing up 2 things, let me try to make it clearer:

1- The link you provide talks about the NAM player feature of the Stomp Station pedal.
2- I was talking about the new Magic Match feature of the pedal. It's different and complementary to NAM captures.

NAM captures a pedal, and amp, or a full rig whereas Magic Match captures the sound of a guitar (or bass) in a song. Big difference.
I'm certain that Magic Match is far less accurate than NAM in all circunstances (but the training process is also much faster, and the goal is different anyway). With NAM, you will probably won't be able to tell the difference between the original amp and the capture. With Magic Match you only get close to the tone in the original song, but you will definitely be able to hear the difference between the original and the copy, like you will see if you watch my demo.

I think The Stomp Station is a good NAM player, but probably not the best one. And it's not comparable with Helix, at all.
Darkglass Anagram plays in the same league as Helix IMHO, not Sonulab Stomp Station. It's not only a matter of price, also and mostly a matter of features and sound quality.
 
I think you're mixing up 2 things, let me try to make it clearer:

1- The link you provide talks about the NAM player feature of the Stomp Station pedal.
2- I was talking about the new Magic Match feature of the pedal. It's different and complementary to NAM captures.

NAM captures a pedal, and amp, or a full rig whereas Magic Match captures the sound of a guitar (or bass) in a song. Big difference.
I'm certain that Magic Match is far less accurate than NAM in all circunstances (but the training process is also much faster, and the goal is different anyway). With NAM, you will probably won't be able to tell the difference between the original amp and the capture. With Magic Match you only get close to the tone in the original song, but you will definitely be able to hear the difference between the original and the copy, like you will see if you watch my demo.

I think The Stomp Station is a good NAM player, but probably not the best one. And it's not comparable with Helix, at all.
Darkglass Anagram plays in the same league as Helix IMHO, not Sonulab Stomp Station. It's not only a matter of price, also and mostly a matter of features and sound quality.
Thanks for that. I did know all that. But I appreciate you taking the time to write it all out.

Did you go to the link I posted? See the aliasing comparison? And the chat about converting NAM profiles?
 
  • Like
Reactions: gtirard
Thanks for that. I did know all that. But I appreciate you taking the time to write it all out.

Did you go to the link I posted? See the aliasing comparison? And the chat about converting NAM profiles?
OK, sorry for misunderstanding your initial comment, I though you were commenting about the Magic Match feature because you posted that comment after my post about Magic Match. Now I understand you were refering to the Gear Page thread where they say that the Stomp Station plays NAM profiles better than the original NAM by converting them into their own format, which is based on classic amp modeling tech rather than AI. Like you, I think that it would be wild if true.

But here's the catch (IMHO):
  • From my experience, NAM captures when played with the DAW plugin or a good pedal are indistinguishable from the original, and I've never heard any artifact, so I don't really care to find "better than NAM" tech even if it exsists.
  • The main limitation in the current NAM version is that it captures a snapshot of an amp with its setting at the time of capture, and you're not able to 'tweak the knobs' of the virtual pedal or amp. So you have to make one capture for each setting you want to use. It's not very practicle, and even if you make a lot of captures, you will never have the infinite number of settings you have with the real amp or pedal.
  • So the next revolution I'm waiting for is not to improve the sound quality of the captures, but to include the possibility to modify the settings of the pedal or amp in the capture. I know the technology exists, it's call "Parametric NAM", but it's only available through a couple of experimental products, like this one, it's not something we can use right now when we capture our amps or pedals. I hope parametric NAM becomes fully available to users, eventually one day.
 
OK, sorry for misunderstanding your initial comment, I though you were commenting about the Magic Match feature because you posted that comment after my post about Magic Match. Now I understand you were refering to the Gear Page thread where they say that the Stomp Station plays NAM profiles better than the original NAM by converting them into their own format, which is based on classic amp modeling tech rather than AI. Like you, I think that it would be wild if true.

But here's the catch (IMHO):
  • From my experience, NAM captures when played with the DAW plugin or a good pedal are indistinguishable from the original, and I've never heard any artifact, so I don't really care to find "better than NAM" tech even if it exsists.
  • The main limitation in the current NAM version is that it captures a snapshot of an amp with its setting at the time of capture, and you're not able to 'tweak the knobs' of the virtual pedal or amp. So you have to make one capture for each setting you want to use. It's not very practicle, and even if you make a lot of captures, you will never have the infinite number of settings you have with the real amp or pedal.
  • So the next revolution I'm waiting for is not to improve the sound quality of the captures, but to include the possibility to modify the settings of the pedal or amp in the capture. I know the technology exists, it's call "Parametric NAM", but it's only available through a couple of experimental products, like this one, it's not something we can use right now when we capture our amps or pedals. I hope parametric NAM becomes fully available to users, eventually one day.
So from my perspective aliasing is a issue. And I say this because I've used the various quality Slammin Mofo captures and can hear a distinct lack of aliasing in the hyper quality versions. And he's shown the aliasing graphs to back that up.

The problem is that no NAM hardware runs it.

The stompstation supposedly converts NAM models to something called Classic Amp Models, which supposedly have almost no aliasing.

So in that context I'm interested.
 
Fractal Audio just made some noise this weekend where they've teased that their next generation product will support NAM, among other features. No timeline or concrete specifics given (not prudent at this time), but this is GREAT news for opening up the Fractal Audio world to more bassists.

Lack of bass models has been a longstanding gripe (though new AxeFxIII firmware adds a bunch of bass goodies).