NAM: Neural Amp Modeler

If your main focus is on how to monetizing it, especially open source with zero infrastructure, you ought to be disappointed.
I don't seek to get anything from it, my effort is rewarded by playing my amp without the extra heat, noise, and electricity bill. :p
I can only really speak for myself here (not actually selling any NAM models yet but I get people asking and thats why I'm considering it - the ToneX stuff I sell has been popular though). There are definitely guys selling models and presumably getting enough interest to deem it worthwhile. For my own use, I totally agree with you - its most helpful as a snapshot of a particular rig in a particular state which makes it easy to recall later.

I think ultimately, its driven by the demand of what people want. It may not be an ideal or preferable way to work for everyone but there are clearly people out there willing to pay money for NAM models. I didn't see the appeal of the Bogren one knob stuff, but they've proven to be very popular too. Making profiles/models is something I do on the side rather than my main job. I suspect the likes of ToneJunkie et al may have more incentive to put a lot of time and effort into their packs, and also more incentive to advertise their products more aggressively.

I don't personally see any issue with monetising their work - it takes time, effort and resources to do, and just because some might be willing to do it for other incentives besides money it doesn't mean that's the right way for everyone. its really up to the individual to decide (and users to decide if they want to buy or not, there is more than enough free stuff out there too).
 
Oli Larkin is on fire on github. Looks like the next version is going to have a feature that will make multi capture packs much more convenient to use.


Now if only people start making captures for 12dBu, Gain 1-10, and not crank the Master volume, that will be just great.

Yes - this is rad. People constantly ask for features and never actually look at github haha.

I don't seek to get anything from it, my effort is rewarded by playing my amp without the extra heat, noise, and electricity bill. :p

To be fair, I'm pretty sure your average desktop is using as much or more power than your amp does (mine is, anyway). But I'm with you - plus I've actually started to really hate the noise tubes make that you can actually hear with a loadbox because it's louder than my actual monitor speakers
 
As far as selling content, I don't care if people do or don't (nor does Steve). In fact, I've already bought a few from Amalgam I think. A lot of people expect free content because the platform is free, but don't really seem to understand that NAM isn't a company like NDSP or IK. Lots of professional folks have dropped free content they would otherwise charge for on other platforms already purely out of interest and maybe to get a foot in the door, but I don't think the alternative to them charging is just giving it all away. I think the alternative is that they wouldn't make anything more.

Some people seem to want to champion nam as a company killer / great equalizer, but it isn't that, nor is that a vision for it. If people don't want paid captures, they can simply not pay for them.
 
I've actually started to really hate the noise tubes make that you can actually hear with a loadbox because it's louder than my actual monitor speakers
Lol yeah, that would be the output transformer. :LOL:
The copper windings rattle against the core at the same frequencies as your playing.
 
Usually the copper windings are lacquered and even potted to prevent rattle, but that does not make them dead silent, the magnetic field is very strong, remember you are putting 100W at random high frequencies (playing) through it, that's not a constant 50/60Hz mains sine like the power transformer sees.
 
Usually the copper windings are lacquered and even potted to prevent rattle, but that does not make them dead silent, the magnetic field is very strong, remember you are putting 100W at random high frequencies (playing) through it, that's not a constant 50/60Hz mains sine like the power transformer sees.

I guess I need vacuum sealed insulation and liquid cooling for my transformers
 
has anyone thought much about doing preamp and power amp models seperately? I'm pretty sure this is how Neural do it, I think it makes a hell of a lot of sense, as it saves a lot of permutations when capturing bigger packs.

NAM could probably benefit from some tweaks to support this - I'm guessing a training signal that is optimized for a fully wide open poweramp (the current signal has been designed around going into a preamp and out of a poweramp at typical useage settings. Most master volume controls are before the FX loop, so poweramp are running fully open when the FX return comes back in - the current NAM training signal is likely to be insanely loud and maybe focused on a less important part of the range than we want to capture).

Likewise, as well as the plugin supporting 2 slots for preamp and power amp, the input gain control would probably be better replaced with a more traditional volume pot (possibly even with the option for level matching the output).

I tested some poweramp only models with algorithm based plugins and it can certainly bring some of them to life. I did notice many power amp models are with the power amp's own MV down (rather than fully open like a typical guitar amp head), or they were captured with a weird load that didnt have the resonant bass peak you'd expect.

I think its a topic worth exploring because it would save a TON of redundant models, and also anyone with crappy reactive loads can just focus on capturing their preamp (and using preexisting power amp models).

This would also be an easier area to implement controls for interpolating between captures, because power amps typically only have a few controls to deal with (that are fairly common between amps). It would also make it much easier to switch impedance curves, because you don't need to capture all the preamps again, just need poweramp models with different loads attached).
 
Yeah I've been thinking about capturing the poweramp, but not yet figured a good way to match levels.
Also, I am not a fan of poweramp distortion of any kind.
 
Yeah I've been thinking about capturing the poweramp, but not yet figured a good way to match levels.
Also, I am not a fan of poweramp distortion of any kind.
poweramp distortion=bad
poweramp->cab interaction=all of the magic

I did some tests with @northern_fox ’s poweramp models where you have this kind of curve:

IMG_2618.jpeg


It sounds wrong, but I tried it on bx_megadual - first with the built in poweramp disabled, but it sounded like ass. I left the in built poweramp on, and added this over the top. INSANTLY makes it sound a million times more like a rectifier should.

Can only assume it was modelled with a crappy sounding load, and adding this kind of curve on top brute forces it into something more accurate.

I think it’s really about capturing the dynamic interaction between power amp and load, along with (largely unwanted but still very much important) distortion characteristics.
 
The main point though, is that it SHOULD result in fewer redundant tone models and perhaps in time a better user experience if poweramp controls can be handled well inside NAM.
 
poweramp->cab interaction=all of the magic

I think so too, that's why I prefer captures with a real cab as load with a low master volume setting.
Reactive load works alright but it's missing that loose resonance and maybe a touch of compression even at low bedroom volumes.

I think I will try capturing mine sometime tomorrow, with Presence at 5, Resonance at 0.
I got a Zero Loss FX Loop from Metropoulos into a standard log-tail PI topology with 2x 6L6GC and adjustable NFB pot.
 
It would be interesting too, to validate preamp+poweramp as separate models vs a combined model (I guess with a patch cable in the fx loop so the circuit is even).

Also curious how to establish a reasonable reamp level for FX returns
 
It would be interesting too, to validate preamp+poweramp as separate models vs a combined model (I guess with a patch cable in the fx loop so the circuit is even).

Also curious how to establish a reasonable reamp level for FX returns

Actually I did try this - there are pre, pre+power captured simultaneously with pre via DI box between, and power captured separately in this folder:

It's not exact, but it's pretty close when combining pre + power to the pre + power capture. I mentioned it on FB, but my process for level setting was to establish what a few of my rack preamps put out voltage wise at a normal use case (0.7-1.3V or higher if its really hot) and used a playbackback of those preamp outs as recorded into my DAW to set the reamp voltage to match. I also included the clips of the preamp audio in the training set for NAM. I guess I'd measure the FX send level for an amp head in that case and do similarly
 
It's not exact, but it's pretty close when combining pre + power to the pre + power capture. I mentioned it on FB, but my process for level setting was to establish what a few of my rack preamps put out voltage wise at a normal use case (0.7-1.3V or higher if its really hot) and used a playbackback of those preamp outs as recorded into my DAW to set the reamp voltage to match. I also included the clips of the preamp audio in the training set for NAM. I guess I'd measure the FX send level for an amp head in that case and do similarly
Getting around 1.3V out of a BE100 preamp here when the MV (i.e. FX Send) is set to 10. At typical settings, its a hell of a lot quieter.

I'm not sure how to land on the optimal levels but I suspect its to capture the poweramp fully wide open with at least 1.3V on the test signal. And then to significantly lower the input volume when using the model.... Will see how I get on, have made a few files for testing.
 
Getting around 1.3V out of a BE100 preamp here when the MV (i.e. FX Send) is set to 10. At typical settings, its a hell of a lot quieter.

I'm not sure how to land on the optimal levels but I suspect its to capture the poweramp fully wide open with at least 1.3V on the test signal. And then to significantly lower the input volume when using the model.... Will see how I get on, have made a few files for testing.

1.3V was for the lee jackson (that was halfway up), which is kind of insanely loud. My Rocktron piranha was more like 500mv to 1. However, at such high levels the amp would have been driven too hard had I set the level to max. This was sort of just how I usually set the amps up, which is why I chose those values
 
has anyone thought much about doing preamp and power amp models seperately? I'm pretty sure this is how Neural do it, I think it makes a hell of a lot of sense, as it saves a lot of permutations when capturing bigger packs.

NAM could probably benefit from some tweaks to support this - I'm guessing a training signal that is optimized for a fully wide open poweramp (the current signal has been designed around going into a preamp and out of a poweramp at typical useage settings. Most master volume controls are before the FX loop, so poweramp are running fully open when the FX return comes back in - the current NAM training signal is likely to be insanely loud and maybe focused on a less important part of the range than we want to capture).

Likewise, as well as the plugin supporting 2 slots for preamp and power amp, the input gain control would probably be better replaced with a more traditional volume pot (possibly even with the option for level matching the output).

I tested some poweramp only models with algorithm based plugins and it can certainly bring some of them to life. I did notice many power amp models are with the power amp's own MV down (rather than fully open like a typical guitar amp head), or they were captured with a weird load that didnt have the resonant bass peak you'd expect.

I think its a topic worth exploring because it would save a TON of redundant models, and also anyone with crappy reactive loads can just focus on capturing their preamp (and using preexisting power amp models).

This would also be an easier area to implement controls for interpolating between captures, because power amps typically only have a few controls to deal with (that are fairly common between amps). It would also make it much easier to switch impedance curves, because you don't need to capture all the preamps again, just need poweramp models with different loads attached).
Wouldn’t be as accurate. NAM’s whole point is accuracy, so I don’t see the point.
 
Back
Top