New neural lv2 plugin from Aida DSP based extensively on existing NeuralPi, reduced to the bone

Less noisy but still noisy.

Noise is there even if input side is not connected (red circle)

MoonJazz.json Vocaroo | Online voice recorder

1 Like

Okay, that’s because now


is running instead of the cmake install command. We need an APPEND, I know how to do that in yocto but not in buildroot any clue?

Try looking at other plugins’ mk on mod github.

Okay, I remember now I fall into the very same issue while playing with gcc optimizations. To build the plugin I use GNU 9.3.0. I see you have GNU 7.5.0. Can you do a test? Tweak the mk file by hand and remove -DRTNEURAL_ENABLE_AARCH64=ON. This should makes the plugin to consume more cpu, but the noise should go away (if I am correct)

    mkdir -p $(TARGET_DIR)/usr/lib/lv2/rt-neural-generic.lv2
    cp -rL $(@D)/rt-neural-generic/* $(TARGET_DIR)/usr/lib/lv2/rt-neural-generic.lv2/

Removing the -DRTNEURAL_ENABLE_AARCH64=ON works and the cpu usage it’s not increased that much. I’m going to test all the models. Later I will try to record some models from Cube Baby and Nux MG-30.

This has huge potential! Thanks a lot @madmaxwell!


Okay perfect! As a reference for cpu consumption MoonCrunch.json is 15.7% on my board, what cpu has Dwarf? I think we could patch the optimization in the CMake file for this specific build (and keep them in a patch in mod-plugin-builder). Can we talk more about models in the other thread @spunktsch @rogeriocouto @redcloud. If I can stop paying 10eur/month for Colab Pro I would be glad!


I’m sorry, the model was MoonJazz.json. CPU consumption with AIDA DSP alone in the board is around 28-30% (empty pedalboard is around 8-10%) at 128 frames. Dwarf uses a

RockChip PX30 CPU:

    CPU family: Cortex-A35
    Platform bit depth: 64 bit
    Cores Clock: 4 x (up to) 1.3 GHz

@madmaxwell it feels like some oversampling is missing.

1 Like

I’m really keen on trying it out on my dwarf. Can you provide just the compiled plugin that I can drop into the dwarf?

If its just 10€ / per month then we just pay it until its not needed anymore. I had not the time to look into running the script with azure but should be doable.

1 Like

Try unpacking it into /root/.lv2
Don’t know it it’s enough.

To change .json model, unless you want to patch GUI files, you have to edit this line on the bottom of rt-neural-generic.ttl file

state:state [
    <> <models/SIMPLE-RNN/48kHz/LSTM-12/Moon/MoonJazz.json>

or make a symlink to models/SIMPLE-RNN/48kHz/LSTM-12/Moon/MoonJazz.json with the model you want to use, like

(preliminary steps)

cd /root/.lv2/rt-neural-generic.lv2/models/SIMPLE-RNN/48kHz/LSTM-12/Moon
mv MoonJazz.json MoonJazz.json.bak

to use a model

ln -sf <MODEL_FILE>.json MoonJazz.json


ln -sf ../Vintage/<MODEL_FILE>.json MoonJazz.json

All currently available models

ln -sf MoonAcoustic.json MoonJazz.json
ln -sf MoonCrunch.json MoonJazz.json
ln -sf MoonJazz.json.bak MoonJazz.json
ln -sf MoonRiff.json MoonJazz.json
ln -sf ../Vintage/VintageEdgeOfBreakup.json MoonJazz.json
ln -sf ../Vintage/VintagePedalPlatform.json MoonJazz.json
ln -sf ../Vintage/VintageRockRythm.json MoonJazz.json
ln -sf ../Vintage/VintageRockSolo.json MoonJazz.json
ln -sf ../Vintage/VintageUnderWater.json MoonJazz.json

Every time you change a model you have to remove and add the pedal from pedalboard.

1 Like

thanks. I installed it with no problem and put it on a pedalboard with a cabsim.
I only get noise and hear my guitar a little bit. Testend with a few models.

EDIT: nevermind. A reboot got rid of the noise but it sound definitely somethings missing.


Thanks for the feedback. Actually we have to investigate what’s the root problem here. Samples from @madmaxwell sounds decent.

1 Like

the Neural DSP plugin that I’ve modeled was not using oversampling in the first place. Was sounding good. Oversampling would increase CPU also.

Can we be more precise about what’s wrong with the sound? I know is not that easy but since I don’t have a Dwarf and apparently I don’t have those problems I don’t know where to look

made a short clip of the problem - nothing is clipping.

Which preset / model are you using? What is the peak input level of the signal reaching the plugin? Thx

I used the MoonCrunch. Levels seem fine.

Hardly, but we never know. Does this happen with any other plugin or if you connect the ins to the outs do you have anything resembling this? Did you change something in the settings while sshing? Maybe the best would be to do a factory reset, although at this point in time it may be a bit hard :confused:


I’ll pay attention if it will happen again and do my private investigations :sweat_smile:


@madmaxwell Just got a chance to catch up on this thread, amazing work! And glad to see the community is getting excited about it. It’s unfortunate what’s happening to Mod-Devices but this gives me some hope that it can continue in some form. Planning on checking out your code and training dataset soon! I’d be curious to get a recording of the Quad Cortex training input, I’ve only heard bits and pieces but it seems to be some kind of generated frequency sweeps as opposed to actual guitar playing.


Are you on the discord server for Mod? Theres a test wav file under the plugin category. It’s similar to what kemper and neural use