NAM: Neural Amp Modeler

[DLC86]

Thanks DLC86 for all the work !

I'm totally out of my depth here, but is it therefore the case that AI generated Capturing techniques from Tonex or NAM cant be oversampled to mitigate aliasing (?) whereas component modelers like the Axe and Helix can be and are oversampled to mitigate aliasing (?)

Ben

As I said previously it could probably be done by capturing and training at a higher sample rate, but currently they don't allow to do that. We'll see in the future.

 

[northern_fox]

RE this:

Support modeling cabs (#250) · sdatkinson/neural-amp-modeler@29d1fb0

* Update data.py * Sample rate more flexible in wav_to_x * Better error messages in ConcatDataset * Black * Docstrings on Dataset * Amp and cab training in core * Cab modeling optio...

github.com

How does that work?
You have to include the IR with the output.wav for training?
What if I'm reamping a full rig without an IR?

There are two different things going on I think - one is just some tweaking that improves the standard full capture results as they are now. He hasn't made any statements regarding the other items or separate capturing yet, so I'd just watch where it goes for now. You're just seeing the leading revision edge there, not necessarily what's ready for release. And also there's a lot of NAM which is DIY oriented - the way the colab and GUI trainer work is limited to what works "best" most of the time and is easiest to use/understand. You can go a lot deeper config wise, so even if it's exposed in code it won't necessarily be something you use unless you want to (like the other model types and parametric capturing)

As I said previously it could probably be done by capturing and training at a higher sample rate, but currently they don't allow to do that. We'll see in the future.

Yes - it IS allowed, but you'll have to do it yourself open source style I don't think anyone has tried yet

 

[DLC86]

Yes - it IS allowed, but you'll have to do it yourself open source style I don't think anyone has tried yet

And that's the issue for me, I'm not a programmer. When I tried to feed it 192 kHz files I got some errors that pointed to specific lines in some configuration files, so I went to edit those lines and change all "48000 Hz" to "192000 Hz", but at that point I got several other errors. Not knowing what I was doing exactly I gave up and reverted back all the changes.

 

[Mike Oliphant]

And that's the issue for me, I'm not a programmer. When I tried to feed it 192 kHz files I got some errors that pointed to specific lines in some configuration files, so I went to edit those lines and change all "48000 Hz" to "192000 Hz", but at that point I got several other errors. Not knowing what I was doing exactly I gave up and reverted back all the changes.

The training software uses specific patterns in the .wav files to align them. I wouldn't be surprised if this code doesn't generalize to other sample rates.

 

[northern_fox]

The training software uses specific patterns in the .wav files to align them. I wouldn't be surprised if this code doesn't generalize to other sample rates.

Yes, it would be a lot of this sort of thing to get it to work, and you would also have to rework the core player for the plugin I think. It's definitely a non-trivial task and then none of the other models would work either.

 

[Mike Oliphant]

you would also have to rework the core player for the plugin I think

Actually, an advantage of the (otherwise disadvantageous) fact that the plugin doesn't know anything about sample rates is that if a model was trained at a different sample rate and you have your DAW set to that sample rate, it would just work as it should.

In other words, the plugin isn't hardcoded for 48k - it just assumes that the playback sample rate is the same as what the model was trained for (which happens to be 48k for all of the existing models out there).

 

[northern_fox]

Actually, an advantage of the (otherwise disadvantageous) fact that the plugin doesn't know anything about sample rates is that if a model was trained at a different sample rate and you have your DAW set to that sample rate, it would just work as it should.

In other words, the plugin isn't hardcoded for 48k - it just assumes that the playback sample rate is the same as what the model was trained for (which happens to be 48k for all of the existing models out there).

Yess - I wasn't sure if the NAM core in the plugin was also pinned to expect 48khz in any way, but it sounds like it isn't.

In other news, I'm the proud owner of a dual rec for the first time ever, and I'm in love already! Just a mostly random capture I did playing around with it last night with a sound I happened to like. Nothing earth shattering here, but I couldn't help giving it a try :)

Mesa Dual Rectifier 3ch Modern Red | ToneHunt

Mesa Dual Rectifier 3ch Modern Red is a model by northern_fox. Suhr RL EL34 ESR ~ 0.01 Reference level: 12dBU Settings: Channel 3 modern Presence: 7 Master: 4 Gain: 6 Bass: 4 Mid: 8.5 Treble: 4.5 Output: 1 Unboosed and SD1 boosted...

tonehunt.org

 

[James Freeman]

I have a capture request.
Untrained wav exports, also good.

Peavey 5150/6505
Pre Gain: 5 (if you have time 0-10 in 1 increments).
EQ, Pres, Reso: 5
Post Gain: 1 out of 10

Preferably with a real 4x12 cab, that really does make a difference particularly for palm mutes.
12dBu: 400Hz Sine Tone at -10dBFS = 1v RMS at reamping box 1/4" output.
I like boosting with external TS-9 or SD-1 from the Helix or a real one, more control.

My goal is to have a reference 5150 capture set that is not a total random mess, something that I can trust as reference.

Thanks!

 

[James Freeman]

The latest JCM800 2203 v2 captures from the user "Jesco" finally specify dBu value.
His amp is not stock and has 3.4k and a bypass cap on the second gain stage in place of the stock 10k, so it's gainier.
Still no amp settings in the description but including the dBu value is a big step forward towards more accurate gain.
Also, I love the 0.5 gain steps format, it's the most importan control in an amp and makes a very versatile capture set.

Marshall JCM800 2203 | ToneHunt

Marshall JCM800 2203 is a model by Jesco. These captures are made on my 1984 JCM800 2203 The amp has had a couple of mods, including a gain mod adding a little more pre-amp gain than stock versions. I dialled in the amp on the high input and did a set of captures increasing the pre-amp (gain)...

tonehunt.org

EDIT:
Hmm, Gain on 10 has less gain than 7, something is not right.

 

[StratsoundFan]

Probably a typo. Wouldn't be the first time.

 

[James Freeman]

I don't want to open a can of worms but why is asking for a paypal donation on every capture description on ToneHunt necessary?
Not to be too blatant but ToneHunt starts to have a beggars street vibe...
Maybe put the paypal link in your profile page? People who are looking for it will find it.

 

[laxu]

I don't want to open a can of worms but why is asking for a paypal donation on every capture description on ToneHunt necessary?
Not to be too blatant but ToneHunt starts to have a beggars street vibe...
Maybe put the paypal link in your profile page? People who are looking for it will find it.

Sounds like people realized that making several captures can be a lot of work.

 

[James Freeman]

I've seen people asking for donation for a single tubescreamer capture... put that into perspective with the tremendous amount of time and work done by actual professionals on github and without asking for a single cent.
To me it seems opportunistic and pitiful at the same time, but maybe it's just my weird view on this.
Anyway, I'll leave it at that.

 

[Deadpan]

I've seen people asking for donation for a single tubescreamer capture... put that into perspective with the tremendous amount of time and work done by actual professionals on github and without asking for a single cent.
To me it seems opportunistic and pitiful at the same time, but maybe it's just my weird view on this.
Anyway, I'll leave it at that.

I don't disagree. Annoying for sure.

 

[Elf]

I don't want to open a can of worms but why is asking for a paypal donation on every capture description on ToneHunt necessary?
Not to be too blatant but ToneHunt starts to have a beggars street vibe...
Maybe put the paypal link in your profile page? People who are looking for it will find it.

It doesn't bother me at all, If I find something I like i might send some cash why not?

 

[northern_fox]

so I did a little experiment:

However, as it turns out this hotone jogg has at least 6 dB higher input noise floor than my clarett, and even more so against the tonex as an interface. Once that noise propagates through all the plugins, it kind of renders this solution less than ideal. If this is really gonna work well, I need an interface that's still tiny but better quality. Suggestions?

So probably just sticking with my tonex rigs live for now because it's substantially quieter overall. This isn't a NAM thing, it's just this interface mainly. Also the jogg outputs are split stereo, which means they are half as loud as they could be when I'm running just one into a power amp, and even at max output it's still a struggle to get enough room volume. I find this to be the case with a lot of digital stuff - just overall tends to be a lot quieter than normal pedals. Definitely a pet peeve of mine. The QC was nearly impossible to get enough output from on a bass at band levels.

 

[James Freeman]

Best combination I found to use with plugins is my Helix Floor with a Focusrite Interface via SPDIF In/Out.
I get the benefit of a super low noise guitar input (-127dBFS) with a big MIDI controller with Snapshots from the Helix, and stability + low latency from the Focusrite.
FM3 also has a super quiet guitar input and SPDIF Out.

Not very portable though.

The market needs Helix/Fractal's super low noise guitar input technology in a small and cheap box with SPDIF+Optical outputs, this will make a lot of plugin players very happy.

 

[BenIfin]

Best combination I found to use with plugins is my Helix Floor with a Focusrite Interface via SPDIF In/Out.
I get the benefit of a super low noise guitar input (-127dBFS) with a big MIDI controller with Snapshots from the Helix, and stability + low latency from the Focusrite.
FM3 also has a super quiet guitar input and SPDIF Out.

Not very portable though.

The market needs Helix/Fractal's super low noise guitar input technology in a small and cheap box with SPDIF+Optical outputs, this will make a lot of plugin players very happy.

I've often thought there was/is/may-be a market for a "dumb" VST Loader Floor Box with reasonable I/O and Midi I/O.

Kind of like Tonex size and I/O with IR Loading but able to load and run any VST 2 or VST 3.

Downside I suppose is that it would have to be windows based (?) ... cpu issues / cooling and Windows O/S etc....

I guess though its been thought of many times but probably the market just isn't big enough -or- there's jut too many software and hardware variables to warrant a sellable price-point.

Ben

 

[the swede]

so I did a little experiment:

However, as it turns out this hotone jogg has at least 6 dB higher input noise floor than my clarett, and even more so against the tonex as an interface. Once that noise propagates through all the plugins, it kind of renders this solution less than ideal. If this is really gonna work well, I need an interface that's still tiny but better quality. Suggestions?

So probably just sticking with my tonex rigs live for now because it's substantially quieter overall. This isn't a NAM thing, it's just this interface mainly. Also the jogg outputs are split stereo, which means they are half as loud as they could be when I'm running just one into a power amp, and even at max output it's still a struggle to get enough room volume. I find this to be the case with a lot of digital stuff - just overall tends to be a lot quieter than normal pedals. Definitely a pet peeve of mine. The QC was nearly impossible to get enough output from on a bass at band levels.

Xtone Pro?

 

[northern_fox]

I've often thought there was/is/may-be a market for a "dumb" VST Loader Floor Box with reasonable I/O and Midi I/O.

Kind of like Tonex size and I/O with IR Loading but able to load and run any VST 2 or VST 3.

Downside I suppose is that it would have to be windows based (?) ... cpu issues / cooling and Windows O/S etc....

I guess though its been thought of many times but probably the market just isn't big enough -or- there's jut too many software and hardware variables to warrant a sellable price-point.

Ben

Yeah - mainly because it sounds simple until you realize just how much ground is covered by VST format and how hard it would be to maintain cross compatibility for a pedal like that. People struggle already using that stuff on ipads and phones that have a much more complex processing setup. I think there are a few companies that basically do it, but you have to use *their* plugins because they can't both supply hardware and provide support for other brands' development essentially. There's no money in it, basically, and I think it's actually a pretty hard task to accomplish because of what's actually required to play a VST. Small M1/M2 boxes and some the DIY chips seem to be the closest we really have currently. I think even NDSP was a little too overconfident in making those promises, and they've got the money and platform to actually do it.

Xtone Pro?

I'll check it out next! Also Audient maybe

Best combination I found to use with plugins is my Helix Floor with a Focusrite Interface via SPDIF In/Out.
I get the benefit of a super low noise guitar input (-127dBFS) with a big MIDI controller with Snapshots from the Helix, and stability + low latency from the Focusrite.
FM3 also has a super quiet guitar input and SPDIF Out.

Not very portable though.

The market needs Helix/Fractal's super low noise guitar input technology in a small and cheap box with SPDIF+Optical outputs, this will make a lot of plugin players very happy.

I could use the HX stomp I guess, or I can actually use the tonex as the interface too, but the sizes start to get big and if I reach a point of bringing a full size focusrite and having to rack it all up to organize it starts to feel a bit less worth it just to use NAM and some other things. I dunno, I'm not quitting yet - just need to try some less noisy interfaces. Shame about the jogg because it was almost too perfect.

Otherwise I've just got a tonex standing on my axe 3 or sharing a tiny board with HX stomp. Both good options, but neither as cool as running NAM

The tonex input and Axe3 inputs had better noise floors than all of the other interfaces I tried, and I think the tonex was even better than axe fx in my test, but I didn't try adjusting the input sensitivity on fx3