NAM A2 and Slimmable

[MikeOliphant]

@fer

When using calibrated NAM, best to decrease the input lvl of the plugin around -10dB, as the plugin expects 12dBu input, but Mod Dwarf input is 2.2dBU (with input setting at 0db )

I could make the input level default a compile-time option - then you could bake it in for the Dwarf.

[AndreaDelSignore]

@fer

If somebody is interested… here the pull request Nam a2 by fragfz · Pull Request #197 · mod-audio/mod-plugin-builder · GitHub

Great! I played with it today and it was wonderful. I hope it gets updated as soon as possible.

@fer

The buffered option is removed, but you always have the portal plugins. As a side note, this could trigger an error with Starless image if open or bootup with a pedalboard using the old NAM plugin. error opening pedalboard when a plugin is updated with less parameters · Issue #32 · sejerpz/alabs-mod-custom-images · GitHub

I’ve just fixed it, and the change will be part of the upcoming aLabs 16 release, once the mod-host transition is complete.

[ignis32]

Are there any chances of A2 capabilities to appear in store, and specifically for ModDuoX?

[gianfranco]

hi gents

@fer

If somebody is interested… here the pull request Nam a2 by fragfz · Pull Request #197 · mod-audio/mod-plugin-builder · GitHub

Will try to push this up today

[MikeOliphant]

@gianfranco If you release a new version, please make sure that you use the current .ttl from my lv2 repo. I just updated it with a link to tone3000.

[gianfranco]

I’m trying a local build with your PR before pushing up.

It works great! Thank you for making the effort. I’ve tried a couple of amps in the Dwarf using the Lite quality and it works supper well. NAM + MOD Cabinet goes at 42.5%, leaving still considerable space for other FX while delivering quite a remarkable quality.

But there are small issues.

• The file browser is listing both the NAM and AIDA-X files. I’ve checked against the online version and the issue is in the PR version.

• And, regarding the quality adjustment, maybe use a drop-down with only the two options (Full / Lite) instead of a continuous knob control? Given the inability of the Dwarf to run full size models, I’d even consider hiding the control for the Dwarf builds and have it internally always set to Lite.

[MikeOliphant]

@gianfranco

The file browser is listing both the NAM and AIDA-X files. I’ve checked against the online version and the issue is in the PR version.

The plugin supports both NAM and AIDA-X files.

[gianfranco]

@MikeOliphant

The plugin supports both NAM and AIDA-X files.

Grazie

[mark_melvin]

Can’t wait to try this…

[fer]

Probably then is better to activate DBUILD_INTERNAL_STATIC_LSTM and DBUILD_STATIC_RTNEURAL in the .mk file for better performance…

[Cipralex]

and how can I get this plugin? is it in the store? how is ist called?

[gianfranco]

it is the normal NAM plugin that we already had, but updated to use the new A2 architecture. I’ll try to push this up asap. Working on some minor adjustments

[MikeOliphant]

@fer

Probably then is better to activate DBUILD_INTERNAL_STATIC_LSTM and DBUILD_STATIC_RTNEURAL in the .mk file for better performance…

Yes for “BUILD_INTERNAL_STATIC_LSTM”. “BUILD_STATIC_RTNEURAL” won’t actually do anything (except slow compile time and increase executable size) as the RTNeural path is only used for unsupported model types (which wouldn’t have a pre-built static path anyway).

[MikeOliphant]

@fer If you get a chance, can you try building for Dwarf with the current head of the “fix_prewarm” branch for my fork of NeuralAmpModelerCore (under deps/NeuralAudio/deps)? It has a change that makes A2 full much faster under some conditions. It is worse for me on RPi4, though, so I’m guessing it will also be slower on Dwarf - but it would be useful to know for sure.

[Cipralex]

cool. but I can not get it to work. i loaded some a2 models from tonezone but nothing happens

[LievenDV]

@gianfranco

I’ve tried a couple of amps in the Dwarf using the Lite quality and it works supper well. NAM + MOD Cabinet goes at 42.5%, leaving still considerable space for other FX while delivering quite a remarkable quality.

This sounds promising!

[fer]

@MikeOliphant I tried the “fix_prewarm” branch, but I couldn’t see any difference. Still many xruns with the A2 full. CPU shown in Mod-ui is jumping a lot. With SSH top, I can see the busiest jack thread is between 94% and 99% of 1 CPU core. On the other hand, A2 lite shows around 22% CPU in a jack thread with top.

Edit: with the “fix_prewarm” branch A2 lite is also worse 46% CPU shown in Mod-UI

[gianfranco]

Hi gentlemen

Just finished publishing the new Neural Amp Modeler LV2 plugin.

I’ve altered the metadata to make the Quality control a list with only “Lite” and “Full” options. The plugin defaults to “Lite”. My impression is that ideally there would be a rule to exclude the Quality control from the Dwarf build. If there is no case where a person would be able to go to Full, so why offer it at all? I’d be glad to get a different POV on this if any.

The A2 improvements are really awesome. Congratulations for the effort @MikeOliphant

With these results I think that integrating Tone3000 directly in mod-ui becomes much more appealing, no?

Have fun everyone

[MikeOliphant]

@gianfranco

I’ve altered the metadata to make the Quality control a list with only “Lite” and “Full” options. The plugin defaults to “Lite”.

While that is ok for now, just be aware that there is nothing preventing future A2 models from having 3 or more quality levels.

[gianfranco]

@MikeOliphant

While that is ok for now, just be aware that there is nothing preventing future A2 models from having 3 or more quality levels.

Understood. I think that as long as we have these scale points it should be ok.