NAM: Neural Amp Modeler

Deadpan said:
Can you post your favorite tone comparison?
Click to expand...
They’re all consistent for me honestly, I never have issues with it not sounding right. Hard to pick a favourite when they all do what they’re supposed to.

Are you doing amp DI’s or mic’d cabs? what’s the chain you are using?

I’m doing new examples now btw.
 
MirrorProfiles said:
They’re all consistent for me honestly, I never have issues with it not sounding right. Hard to pick a favourite when they all do what they’re supposed to.

Are you doing amp DI’s or mic’d cabs? what’s the chain you are using?

I’m doing new examples now btw.
Click to expand...
Referring to full rig captures. I've only done a couple its mostly playing others captures.
 
Deadpan said:
Referring to full rig captures. I've only done a couple its mostly playing others captures.
Click to expand...
I can't really offer an opinion on full rig models as I don't do them or really see the point when amp DI's are so much more flexible (and useful for HW devices). It's also hard to verify what other people are doing without knowing the specifics of how they capture and what the intended result is supposed to be like.

I'd try and make some direct models first and verify they turn out as expected. Then give some full rig models a go.
 
MirrorProfiles said:
anyone available and willing to guide me through what I need to tweak on TZ3000 to get those sounding closer?
Click to expand...
What settings did you use & what was the ESR? Training for 1000 epochs should be pretty good already.
You can use a custom architecture too if you want to push it further.
 
2dor said:
What settings did you use & what was the ESR? Training for 1000 epochs should be pretty good already.
You can use a custom architecture too if you want to push it further.
Click to expand...
Had TZ set to default settings (which is 1000), ESR is just over 0.1. Usually I can get down to 0.06 or 0.07 except when there’s more gain and/or noise.

I’d be curious to try custom architectures if it’s possible to do in the cloud (or if the M4 can do it at a bearable speed, ToneX is super slow to train).
 
MirrorProfiles said:
anyone available and willing to guide me through what I need to tweak on TZ3000 to get those sounding closer?
Click to expand...
That's more gain than any amp I own, so YMMV, but I've found the xSTD architecture and Carmelo DL's TTSV10 input has gotten me the closest to the real amps.
 
Last edited:
we
MirrorProfiles said:
Had TZ set to default settings (which is 1000), ESR is just over 0.1. Usually I can get down to 0.06 or 0.07 except when there’s more gain and/or noise.

I’d be curious to try custom architectures if it’s possible to do in the cloud (or if the M4 can do it at a bearable speed, ToneX is super slow to train).
Click to expand...
0.1? ...that's strange.
did you use the sweep trainer or the wet/dry?

(The reason I ask that is because the wet/dry trainer on TZ doesn't auto-align, but the normal sweep trainer does auto-align).
 
Dom Mcsweeney said:
we

0.1? ...that's strange.
did you use the sweep trainer or the wet/dry?

(The reason I ask that is because the wet/dry trainer on TZ doesn't auto-align, but the normal sweep trainer does auto-align).
Click to expand...
it’s using the latest sweep (I never use wet dry). It’s not uncommon for me to get ESR’s in this range if I use a lot of gain/boosts. But I’d definitely like to tweak things to get more accurate
 
Hmmm. It's probably something related to something noisy in the reamp chain, rather than the trainer itself.
Ground lifts etc usually help, and I guess not slamming levels into the pedals/amp too hard.
 
MirrorProfiles said:
it’s using the latest sweep (I never use wet dry). It’s not uncommon for me to get ESR’s in this range if I use a lot of gain/boosts. But I’d definitely like to tweak things to get more accurate
Click to expand...
0.1 is pretty bad as far as the ESR's concerned (usually aim for 0.01 or below). You can try this architecture:

DLC86

Post in thread 'TONE3000 (previously ToneHunt and TONEZONE3000)'

Starting from today Tonezone3000 offers the ability to use custom architectures! Just select "custom" and paste whatever code you'd like to use.
To use the xSTD just paste this code:

JSON: 
{
  "layers_configs": [
    {
      "input_size": 1,
      "condition_size": 1,
      "channels": 8,
      "head_size": 8,
      "kernel_size": 6,
      "dilations": [
        1,
        3,
        9,
        27,
        81,
        243,
        729
      ],
      "activation": "Tanh",
      "gated": false,
      "head_bias": false
    },
    {
      "condition_size": 1,
      "input_size": 8...
  • Like

Maybe even set one of the "gated" flags to "True" in either one of the layers.
 
Dom Mcsweeney said:
Hmmm. It's probably something related to something noisy in the reamp chain, rather than the trainer itself.
Ground lifts etc usually help, and I guess not slamming levels into the pedals/amp too hard.
Click to expand...
(0.1 was a typo btw, 0.01 is the ESR but I'd still like it to sound more accurate). Yes, I am using ground lifts, and have several decent reamp boxes (I’ve tried quite a lot over the years). Sometimes I’ll even add the Lehle after the reamp box.

The levels hitting the amp are not being slammed - the peaks are the same level as plugging a guitar directly in. No boosts being used here.

Do you have any advice on how to tweak the training so it’s more accurate for these kinds of tones?
 
Last edited:
2dor said:
0.1 is pretty bad as far as the ESR's concerned (usually aim for 0.01 or below). You can try this architecture:

DLC86

Post in thread 'TONE3000 (previously ToneHunt and TONEZONE3000)'

Starting from today Tonezone3000 offers the ability to use custom architectures! Just select "custom" and paste whatever code you'd like to use.
To use the xSTD just paste this code:

JSON: 
{
  "layers_configs": [
    {
      "input_size": 1,
      "condition_size": 1,
      "channels": 8,
      "head_size": 8,
      "kernel_size": 6,
      "dilations": [
        1,
        3,
        9,
        27,
        81,
        243,
        729
      ],
      "activation": "Tanh",
      "gated": false,
      "head_bias": false
    },
    {
      "condition_size": 1,
      "input_size": 8...
  • Like

Maybe even set one of the "gated" flags to "True" in either one of the layers.
Click to expand...
Thanks man! will try
 
MirrorProfiles said:
on.soundcloud.com

UBER A/B

MIRROR PROFILES thanks for listening.
on.soundcloud.com on.soundcloud.com

@Deadpan Just a quick one with the uber (these are maybe a bit more different than average as the gain is up and there's a bit of noise but still good enough for rock n roll).



Click to expand...

Sounds good.

Assuming A is the NAM I feel the difference is similar to what I feel from the studio captures but less so.




Has NAM improved with full rig captures?
 
Dom Mcsweeney said:
0.1?
Click to expand...
2dor said:
0.1 is pretty bad as far as the ESR's concerned (usually aim for 0.01 or below
Click to expand...
Was a typo, ESR was 0.01

Screenshot 2025-04-30 at 19.36.34.png


Training an xSTR model now, will see if it gets any better. I also removed my amp switcher from the reamp chain (one less cable). I don't think there is any difference in noise but I'll make a model with this just to see how much it factors. I can't see a meaningful difference in noise from the waveforms though.
 
MirrorProfiles said:
Was a typo, ESR was 0.01

View attachment 42933

Training an xSTR model now, will see if it gets any better. I also removed my amp switcher from the reamp chain (one less cable). I don't think there is any difference in noise but I'll make a model with this just to see how much it factors. I can't see a meaningful difference in noise from the waveforms though.
Click to expand...
Oh, yeah - definitely easy to get confused by all them decimals; (happened to me as well way too many times to keep track of lol.

The xSTD will probably get you a hair under 0.01 and if you set the "gated" flag to "true" in either one of those layers in the xSTD it should bump it down even lower but come with a slightly increased CPU cost at runtime.
 
You must log in or register to reply here.

Similar threads

B
Tried to match some more Vox tones with NAM
Replies
0
Views
218
bikescene
B
Back
Top