NeuralNetwork.loadNAMModel() functional?

[Orvillain]

Neural Amp Modeler (NAM) in HISE

@Lurch I mean, it works? The performance is seemingly twice as bad as the NAM plugin, but only because the neural node in HISE is processing two channels, an...

Forum (forum.hise.audio)

See this post.

[iamlamprey]

@Orvillain Yeah that thread got me to where I am my use-case can't store the JSON data inside the plugin since I want the end-user to be able to load their own NAM models, I need dynamic save/load (working-ish) and persistence (apparently not working)

[iamlamprey]

Alright I'm officially stumped:

For the record, this behavior happens for any container, not just the soft_bypass.

[resonant]

@iamlamprey what about loading the nam models as json objects?

[iamlamprey]

@resonant said in NeuralNetwork.loadNAMModel() functional?:

@iamlamprey what about loading the nam models as json objects?

Yeah that's how I'm loading them. It was working fine at first but it seems like the more I tinker with it the worse it gets, now the original code isn't working at all

[iamlamprey]

Here's a minimal snippet:

HiseSnippet 1503.3oc2W80aZbDDeOvWRflH0zlGpTeYkUkJoxhXHD6zVUUiwPqSMXjOWq7VzxcKvFu2tmtaObnQ489Z+1zOF8iR+FzNycGvQBwwgzF0TPBtclcl42N67uqen1kGEoCIVkOcZ.mXcSamoJy3ViYBE4vC.518Z10viL0I6OMfEEw8HVVE+AjuUoMHIe9yueeljob4KHQHmoEt7iD9ByBp826mDRYGlG+Tgetc2XuCc0pVZoNFvRQ6sIAL2yYi38X31JXS9QVzXh0WY65Mn988Z3VuA+quOeXiGtas56NrV8Grai5rcZz.3w2cGVch00Z6IL5PGCCfOwZi80dScFquPkZfyDQhARNtnFwArbJ4NZoGdDQpjViERu9ybRQDhkc+Etrhotr6X2U3IlSegq6iSXPWHQdGnUgkgWwkfWs7va6bvaEPxJGj1HER211wMTDXVvAwyGYenxvCGxf6o7PIcujB5B1szvNTlp9ry4cBgEyknxNau8VT3m69skKC2UQFpRQ+NZa0HghW0MjCmid73PlrG2bgN77Jap3Wzq2lf.o6enPxAI5.+4LMxv8qNLT62bPjVFa38YlwU1bJD.fOcO7gtZOtr5SizJPIz6cOZbDmxnfojz.XSkKCzFIlvolw.iAZ3oMQqrIcBKTv.+W9sSMZXIDNS8QESq.vZBOzv8PNOx43d2EUHnKEMV4p88AWQhpU7mYnR3b9MnISOMHrfSCZtpRMyqYzwCdJ20TAcPHVyoAQD8BTq.T7lhLECoBC0SyiTeoghdqsndgrQIVqKf0potRpBPJUGan5gI7.aaf3Kd3Vo3b.jlPEJzhJUBNfyWheqBBPDKCAjXD.X0pdZC+XUk6V94kKU9EkouLqgCWIOLpHTKk7vUxFSkCuLAqnh8GfPdBSFymuQHbc4b.6WeNP9TT2zfzbaTqNTILGGvUutDWRVjM7zOe3ALCCSbxnA6K.BCDHDrNfOAJaklFUx9.dz4Fc.T35UxwfrasWrjYVNkGKLlw.7AKkmg4RpHgYZ9Bm+iUG3pBwaa2WXbGuZLVXEXD7T+afwrpm2xt8vgPZyB.tgcmG+9qTY4rRkfMQieyrkzNOt1pJRZ8q1KBvCB4ArP9o59R1zJQL+.I+D.iaQGH0tm6H9E9qlUDjBg8wcTwcLSo3xn0I44ZWYuRsK+R4Dn9hPMBJ6DJdFjE1K12Ap95xakgNflUALGJc813ZLtvgq7RV7WvmLl0v0VYLqMiYtzurtCIWGYOSrtA56uFNpQcxY7vHLrz551aWE9RZJk5KZo8CDYArv0QBs9Z4zfwZkvEIktiYftouNFrWFxWDsu+ralYGBXvhSYBIF76DGAkP7NV4.5IYZFH3ypGTME.WGlK3smhMmvjn4Ehq5NOxM6.rHrAc3oSvjwB00robRMeqzlDIy3XUD4m5YRMJp0qamZg2..VkUmI55X1aXqRZAsrcuksO1cJi0pr4b4dCFsPtxtoYl3JntSgzCcRSLxYXPOtLYVBn.BKDrBLSRR0jkVbYp6l1GFcFx0kMWmPfw6rBdW6k7oyJB1L1SnORiBmDJcmTJTjDEKuvCqcICf+GW0AvCtxCfCuGfa2jvgEMqR5bbZHLyATuHuoPX1VgCb4km7IZsAGrH+K.3jTlDpZEZ.0uzt4vzXQKqAbXwz3pO2948O43G0t0oOoywGcP6SdwSgSgW0KXSH9nuv152KQH9LrDl0Q+VIhD.UhclwSlfRuY7sJdYugPs07MD13+bugvG.CurTm4Ryvni.CUZCCoKgDiDL9IPZzPVrzLi5xYEc0J8rNB4CrfNaiFwCyi8UdfZZLPv8BJ2YuS3RNKJWD7Wr2QPUWV3Kma7V3Kp8VOkxJuu9L6T3RwTY5GtCcV7+0Cc9dY7p2G1vm4Fpeha5ToXb5MRn.maURA5R1cw0zZjIo8GWbW5CU4dhq6xp5UDr95J38WWAartB9f0UvcVWA2ccE7guYAwlPMiMZ+zTSXb49sSlqwxZdScqhj+FPQfL4d

Unless I'm doing something horribly wrong (or my HISE installation is broken), I don't think this is in a usable state unfortunately, which is a shame because the NAM stuff provides a lot of agency for end-users that don't have the money for expensive amps

[Dan Korneff]

@iamlamprey Does this work on your system?

https://hub.korneffaudio.com/index.php/s/J2XXWaB4HRKJRx7
I can't post this snippet directly cause I get an error

Please enter a shorter post. Posts can't be longer than 64000 character(s).

It's a pretty large cause it has the profile embedded instead of using an external file.

[Orvillain]

@Dan-Korneff Works here Dan. BTW, you can reduce your CPU usage by doing this:

In case you weren't aware. Trick from @aaronventure from another thread.

[Dan Korneff]

@Orvillain nice! I was just trying to keep the original example intact.

[iamlamprey]

@Dan-Korneff said in NeuralNetwork.loadNAMModel() functional?:

@iamlamprey Does this work on your system?

Yeah that one works fine, but if I try and save the object as a .nam again and load it:

Content.makeFrontInterface(600, 600);

const nn = Engine.createNeuralNetwork("newNN");
const af = FileSystem.getFolder(FileSystem.AudioFiles); // thought a wildcard might play nicer

// Any of these:
const file = FileSystem.fromAbsolutePath("myCoolProjectFolder/Scripts/model.nam");
//const file = FileSystem.fromAbsolutePath("myCoolProjectFolder/AudioFiles/model.nam");
//const file = af.getChildFile("model.nam");

// Then convert to json
// Also tried loading the raw .json without the loadAsObject method
const json = file.loadAsObject();

nn.loadNAMModel(json);

Same issue, have to move it to/from a container before it works.

[Dan Korneff]

@iamlamprey so you're having a file extension issue with .nam?
Rename it to .js and include it like any other script and see if that works.

[iamlamprey]

@Dan-Korneff said in NeuralNetwork.loadNAMModel() functional?:

@iamlamprey so you're having a file extension issue with .nam?

I'm not sure it's the .nam being a problem, I can convert it to .json and import it just fine:

This object is what I'm passing to the NN, it's just not updating when I call loadNAMModel() or clearModel() until I move it out of the container. At this point I'd say it's my HISE installation, I'll try debugging it today.

[resonant]

@Christoph-Hart Just a friendly bump. I hope you resolve the issues previously reported here regarding the Neural node (multiple neural model loading, CPU usage, crashes...etc.) as soon as possible. There are many people here waiting to use this node.