NAM: Neural Amp Modeler

Anyway, maybe it's better if you ask for help on the facebook group, there's a handful of very helpful people in there, including the creator of NAM
 
I installed Google Chrome and everything works there, it's a problem with the browser.
I HATE Chrome.

Anyway, a 'Feather' capture is 87kb and takes HALF the time to train with 100 epochs as 'Standard'.
A 48kHz 24bit 2048 samples IR is 6kb.

I will try 'feather' with 500 epochs.
 
Last edited:
From what I can hear and see, most of the difference is above 12kHz, lower ESR results in more accurate high frequencies.
This will be less noticeable with IR.

Standard: Epochs=500, ESR=0.026, 58 minutes, 397kb
Standard: Epochs=100, ESR=0.047, 12 minutes, 398kb.
Feather: Epochs=500, ESR=0.070, 30 minutes, 87kb

NAM Epochs.png
 
Last edited:
James Freeman said:
I installed Google Chrome and everything works there, it's a problem with the browser.
I HATE Chrome.

Anyway, a 'Feather' capture is 87kb and takes HALF the time to train with 100 epochs as 'Standard'.
A 48kHz 24bit 2048 samples IR is 6kb.

I will try 'feather' with 500 epochs.
Click to expand...
Does it also use less CPU when running it in the plugin?
 
DLC86 said:
Does it also use less CPU when running it in the plugin?
Click to expand...
I don't think I can check how much CPU individual plugins take in Cakewalk, but from what I see in windows task manager, Standard jumps round 8% and Feather around 5%, that's with all other processes Cakewalk runs.
About half I'd say.

EDIT:
That's on a i7 at 4.2GHz, I have no idea how that translates to DSP cycles.
 
Last edited:
It’s fantastic, even a transitional moment, and there’s no reason that the continued development of these technical possibilities should fall solely on IK media. As far as tone goes, their product is great today. That said, this thread has very little to do with making music in the present. In the context of a mix, it’s been possible to cut excellent tracks using digital since early Pod bean days or before, and also possible to cut guitar tracks that don’t sound good since the entire history of the advent of great and iconic analog tools, like Fender Valco Marshall etc. The greatest products that I think of have always deferred to human subjectivity and in many instances individual taste. The ubiquitous studio tools we see everywhere benefited from consultation from people like Al Schmitt and George Martin, etc etc. It was never only science that pushed these iconic devices forward.

I’m not sure why people are arguing. I doubt in the context of making music these minuscule differences are always perceptible, but the trend of improvement should have positive implications for all of us going forward, so I’m rooting for everyone going after it in their own way.

Either way, I don’t care how cool independent science is, getting this stuff into a small cheap box is still a gargantuan step forward for actual guitarists.
 
I'm sure a dumb question but I was able to download the vst file and get it in my DAW. Is there a folder area on their website with amp models from various users or no?
 
Mongillo19 said:
I'm sure a dumb question but I was able to download the vst file and get it in my DAW. Is there a folder area on their website with amp models from various users or no?
Click to expand...
They have a Facebook group where people share their captures and also a GitHub that someone is compiling all of them
 
Got my second DI Box, captured my Peavey Franken-Valveking.
Signal is tapped in parallel with speaker output, no IR.
Gain was calibrated for Focusrite Saffire series interfaces (12.5dBu input at minimum gain).

Download Link:
Code: 
https://drive.google.com/file/d/1PcYTsMHDx10LqErTiya7MwkizQLxb-pC




20230314_165148.jpg



Don't mind the buttons, there isn't a single button on the panel that does what it says. :satan

20230314_204631.jpg


ESR.jpg
 
Someone on the Fractal forum in a thread discussing Fractal vs Capture posted a link to ‘A Review of Neural Network-Based Emulation of Guitar Amplifiers’ which is interesting reading if solely for its description of the techniques used for amp emulation and the trade offs.

Something that stood out in their conclusions was the following:

we include the evaluation methods used and the limitations of each method, exploring avenues for further work. Notably the real-time capabilities of such approaches warrant further investigation as the current methods, although some of which are capable of real-time use, are limited to a sampling rate of 44.1 kHz which is the lower bound for acceptable audio quality in music and is insufficient for anti-aliasing techniques based on upsampling the signals. Moreover, the presence of aliasing produced by the networks requires further study.
 
signalpath said:
aliasing
Click to expand...

I just tested, yep, it's terrible.
I wonder if that is an inherent problem with capture itself at a limited sampling rate or the player?
For example, the capture itself is 48kHz but the player has internal x4 upsampling, will that fix aliasing?

Anyone tested ToneX for aliasing?
@MirrorProfiles
Can you run a sine tone at above 11kHz into ToneX and check if it sounds like a broken ring modulator?
 
James Freeman said:
Can you run a sine tone at above 11kHz into ToneX and check if it sounds like a broken ring modulator?
Click to expand...
😂 will do.

I noticed before when doing captures at 48kHz (instead of the recommended 44.1) that it was aliasing like a motherfucker. I think that was more down to IK not implementing oversampling or working at unexpected sample rates very well though.

There were loads of people having issues with captures who were using Fractal units as their interface, which are locked at 48kHz.

The aliasing is probably a bigger concern if you’re stacking up pedals+amp into several blocks. I’m not aware of anyone hearing audible problems from aliasing from even the most high gain captures. I wonder if the slightly softer top end of ToneX is down to oversampling filters.

I know NAM isn’t really supported at multiple sample rates and, should probably look into good sounding anti aliasing filters when support comes.
 
signalpath said:
Someone on the Fractal forum in a thread discussing Fractal vs Capture posted a link to ‘A Review of Neural Network-Based Emulation of Guitar Amplifiers’ which is interesting reading if solely for its description of the techniques used for amp emulation and the trade offs.

Something that stood out in their conclusions was the following:

we include the evaluation methods used and the limitations of each method, exploring avenues for further work. Notably the real-time capabilities of such approaches warrant further investigation as the current methods, although some of which are capable of real-time use, are limited to a sampling rate of 44.1 kHz which is the lower bound for acceptable audio quality in music and is insufficient for anti-aliasing techniques based on upsampling the signals. Moreover, the presence of aliasing produced by the networks requires further study.
Click to expand...

James Freeman said:
I just tested, yep, it's terrible.
I wonder if that is an inherent problem with capture itself at a limited sampling rate or the player?
For example, the capture itself is 48kHz but the player has internal x4 upsampling, will that fix aliasing?

Anyone tested ToneX for aliasing?
@MirrorProfiles
Can you run a sine tone at above 11kHz into ToneX and check if it sounds like a broken ring modulator?
Click to expand...

MirrorProfiles said:
😂 will do.

I noticed before when doing captures at 48kHz (instead of the recommended 44.1) that it was aliasing like a m*****f****r. I think that was more down to IK not implementing oversampling or working at unexpected sample rates very well though.

There were loads of people having issues with captures who were using Fractal units as their interface, which are locked at 48kHz.

The aliasing is probably a bigger concern if you’re stacking up pedals+amp into several blocks. I’m not aware of anyone hearing audible problems from aliasing from even the most high gain captures. I wonder if the slightly softer top end of ToneX is down to oversampling filters.

I know NAM isn’t really supported at multiple sample rates and, should probably look into good sounding anti aliasing filters when support comes.
Click to expand...

Hey all !

Just a dumb question as I really don't know what specifically "aliasing" is (?)

In my simple testing with my Helix, GT1000 and Tonex CS [Free] Software all I have done to "test it" is as follows - but I have no idea if I'm right or wrong:-

=> I've recorded several clean / crunch / gain / lead recordings of each
=> playing basic chords etc .... and then let the final loud chord strum ring and die out to nothing.
=> I've then closely listened back to each - and especially the last chord strum ring out all the way to nothing - and I can hear any "clicks", "pops", sizzle", "crackle", "artifacts" etc... or any kind with any of the three I tried (?) ... just pure normal signal / sound

Is the above in any way a "proper method " to listen for and identify "aliasing" (?)

I genuinely don't know and would like to know so I know what I should be listening for :)

All the best and many thanks,
Ben
 
Last edited:
Its basically when you sample a signal, the frequency spectrum gets copied and added back in at frequencies that are multiples of your sampling rate.
 
BenIfin said:
Hey all !

Just a dumb question as I really don't know what specifically "aliasing" is (?)

In my simple testing with my Helix, GT1000 and Tonex CS [Free] Software all I have done to "test it" is as follows - but I have no idea if I'm right or wrong:-

=> I've recorded several clean / crunch / gain / lead recordings of each
=> playing basic chords etc .... and then let the final loud chord strum ring and die out to nothing.
=> I've then closely listened back to each - and especially the last chord strum ring out all the way to nothing - and I can hear any "clicks", "pops", sizzle", "crackle", "artifacts" etc... or any kind with any of the three I tried (?) ... just pure normal signal / sound

Is the above in any way a "proper method " to listen for and identify "aliasing" (?)

I genuinely don't know and would like to know so I know what I should be listening for :)

All the best and many thanks,
Ben
Click to expand...


The solution is to oversample
forum.fractalaudio.com

Axe-Fx III Firmware Release Version 13.00 Public Beta

Okay, I guess I need to revise my understanding. The way I understand it, Nyquist tells us that in order to perfectly reconstruct a signal, we need a sample rate that’s at least twice the highest frequency in the signal. And we need to leave enough room between that Nyquist rate and our sample...
forum.fractalaudio.com forum.fractalaudio.com
 
Piing said:


The solution is to oversample
forum.fractalaudio.com

Axe-Fx III Firmware Release Version 13.00 Public Beta

Okay, I guess I need to revise my understanding. The way I understand it, Nyquist tells us that in order to perfectly reconstruct a signal, we need a sample rate that’s at least twice the highest frequency in the signal. And we need to leave enough room between that Nyquist rate and our sample...
forum.fractalaudio.com forum.fractalaudio.com
Click to expand...


Alex Kenivel said:
Click to expand...


Awesome video's and link ... huge thanks !

So in a very simplistic way ... using Helix or GT1000 or Tonex Software etc... .... is there anything specific I should be trying to "tune" my ears to "hear-out" for (?)

Thanks again,
Ben
 
You must log in or register to reply here.
Back
Top