How to install NAM Training GUI locally on your PC 

Thanks! I didnt really notice that.

@@jasonzdora Yea, when I ran the setup for Anaconda it recommended that I install for current user for some reason, figured mentioning it might help someone if they got confused. Good tutorial on how the Neural modeler works though. I also use Tonex, having a good free option is pretty cool, keeps paid people on their toes! Lol

This is exactly my intention. Thank you!

17 minutes is pretty good for the highest quality version of a Tonex! Ive heard people say it takes an hour or so. Tonex is great, for sure. I just like the DIY vibe of this system. Good luck with setting up NAM Training GUI on your machine! Join the FB group and share your work, if youre interested! Lots of awesome stuff on there aleady!

Thanks for the info on which cards worked and which didnt! Thats super useful for folks reading along! Glad you got it working :) 6 hours is one of those "set it up before you go to bed" kinda things, but at least it works and doesnt time-out on your like Collabs can!

Awesome! Glad it helped! I get about 3 to 4 seconds per Epoch on my system so a 1K Epoch session is about an hour.

Glad you were able to get through it with the differences! Good luck with your training!

Awesome! Thank you!! I barely know this stuff. Learned it all last week, so I probably said a bunch of stuff wrong. I genuinely appreciate you giving the video some context. Cheers!

Thank you for your original "how to" video! That helped me get this system running on my PC. I dont mean to overshadow your efforts with my own video. I just wanted to clean up some of the erroneous info given by Helga (the list you used) and I also wanted to create an entire "series" of NAM videos to cover all aspects :) Thank you for everything youve done for the group so far, brother! Its great to see how our contributions and our attitudes and excitement (along with Steve Atkinsons genius and super kind nature, of course) have made this thing explode in popularity :)

@@jasonzdora You worry too much. The more people that get their hands on NAM, the better. Thanks for putting the time in to get this video out \m/

Thanks, man! The camera is a Razer Kiyo Pro Ultra. Not bad for ~300 dollars. I also have a Sony Alpha ZV-E10 and the Kiyo genuinely competes with it. Just have to set everything to manual because the Razer software is buggy and the auto-settings go wild sometimes. Much easier setup than the Sony (just one cable, do your settings, done).

This is the first time Im hearing about this. I googled "AIDA-X plugin" and I found their GitHub. Looks very similar to what NAM is doing but using a different system (nam uses WaveNet and AIDA-X looks to be using RTNeural). Pretty cool! I wonder how they compare in the two main places that matter: sound and CPU. Their plugin GUI looks very nice! Thanks for sharing :)

Never mind! I figgured it out 👍 Thanks!

Honestly, I dont know. I would try logging in and doing it again to see if that works. If not, your best bet is to mention your issue on the Facebook group or the Discord.

@@jasonzdora ok thanks... I might try Discord

I believe Tonex uses a different kind of "machine learning" for their system, actually. Theres like 3 popular ones (and I dont remember the names of them!). This stuff is well known in geek circles and has been for years. Its just getting to the point now where folks like the developer of NAM has used this kind of tech in his day job for a while and he understood it enough to build his own system from scratch in his spare time. As far as Epochs, I think youre spot-on with your assessment. The thing is, once you have control over that parameter, you start to see inconsistencies. With NAM you can monitor every single Epoch and see the "score" of them as they progress (the ESR score). You sometimes see training go super low ESR very quickly and then other times it takes way longer. Its nice to be able to monitor that and make your own decisions about when the ESR is low enough.

What action are you taking that causes that error to appear? Does your system have an Nvidia graphics card?

@@jasonzdora First sorry on bad english.. :) I have gtx650. Did everything you said in video. Got NAM to work, loaded input and output audio files but when i press Train button i get that error in anaconda. It was some other error with CUDA at first so I installed CUDA Toolkit ... now it's this one

@@hellios44 No worries at all! I understand you perfectly, brother! ok so the 650 might not be able to work for this. I would suggest going to the Facebook group and asking folks there. I am also not very good at this stuff. Im only good at following directions and then repeating them in a video :) Im no expert, unfortunately. The facebook group will likely have a clear answer for you, my friend.

@@jasonzdora I dont use facebook unfortunately, this is first time i regret it :) It looks like you are right about my graphics card. I have managed to make it work with cpu. I've deleted everything and used conda env create -f environment_cpu.yml command. It's slow but it works :) Thank you for your time and videos

There are two main factors to getting a good ESR score: signal/noise ratio and the overall amount of gain/distortion. I have noticed that the more BTFO a distortion is, the higher the ESR will be. If youre trying to capture a fuzz pedal while its absolutely exploding (gain is cranked into insane sputtery square wave territory) then its going to be hard to model. I know thats the kinda tone you want to model but its tough for machine learning to figure out at this point. I think NAM does a better job than most with this kinda thing but its still not "smart" enough to get a totally blown-out fuzz tone well enough to replciate the intricacies of that kinda tone. Fuzz has so much to do with interaction between the guitar pickups and the opamp in the pedal, but even a regular/conventional distortion pedal will be difficult to model if you crank the gain past the "fully gained up" point. Anything further on the gain knob will just give you the same overall model but with a worse ESR score. The other factor is the signal to noise ratio in your signal chain. If you hear buzzing or humming in the signal chain while no instrument is being played, youre going to get a bad ESR score. That can be fixed with isolation transformers, ground lifts, removing other hardware that might be causing interference, etc. There is no one-size-fits-all answer for that but those are the basics.

@@jasonzdora Thanks for the reply! Leo Gibson made a video diving deeper into some of the settings. After some trial and error I managed to get the ESR down to 0.0076 with 100 Epochs. But the settings I'm using are definitely more demanding on my GPU which is an rtx 4090. I was getting better results than Leo with different settings than he was using. So I think different tones benefit from different settings, and you just have to experiment. Here's his video: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-fxtiItCZKfs.html

@@Geeztown Leo does an amazing job of explaining. He is great! You can tell he has a programming background while I have zero programming skills but I like to get my hands dirty when I know all the steps. I believe I saw you post on Facebook yesterday. I asked if you could share your v111 and output wav files. I wanted to see just how distorted your tone was and just now noisy your silence was (to see if my guesses that I made above were correct).

@@jasonzdora oh, I wasn't sure how to upload an audio file on fb, so I didn't bother. For this particular model, I actually have the fuzz pedal in its lowest setting where it's just starting to break up. I haven't tried a higher setting yet. So I'm really not sure why the ESR was so bad on stock settings. I'm like you in that I have no programming background, but I like to tinker. I've built a few hackintosh's. It's amazing what you can learn just from reading forums and from people such as you and Leo. I've been using your advice to create my own test tone file. I actually recorded the test tones that ToneX uses for guitar and bass, but I also want to use drums and vocals and maybe a few other instruments. I'm still working on that though, so we'll see how it turns out. I noticed that with ToneX the result is very different if you pick the bass test tones. It would be nice to be able to create 1 model that works for everything, even if it takes a little longer.

I'm using 2 little labs red eyes to reamp and di, a uad apollo x4, and mogami cables. So my signal path is pretty clean. I'm not hearing and hum or noise that shouldn't be there.

Whats the context? In general, Ctrl+C means "copy" but Im not sure what part of the video or the workflow you are referring to.

I dont think so but you could ask the Facebook group and you will probably get a definitive answer. I only use PC so Im not sure. I believe there is a way to train with only your CPU but its very, very slow. Since your Macbook doesnt have an Nvidia graphics card, you cannot use the GPU/CUDA training model that Im showing in this video.

@@jasonzdora I finally got that one I hope it`s a good choice: MSI Gaming GeForce GTX 1650 D6 4GB GDRR6 128-Bit HDMI/DP Graphics Card (GTX 1650 D6 Ventus XS OCV1)

@@fretBeastForever That should work! check if that specific card can install the nVidia "CUDA" toolbox.

Thanks! You are better off downloading it from the link in the Github. I believe he has updated the file to v2 a couple months ago.

Just updated my google doc with the new link. Thank you, again!

@@jasonzdora I did wind up getting it from the GitHub. But your video was really helpful in me getting started!! keep up the cool content

Coping?

@@jasonzdora Sorry, copied

@@FFJoe I gotcha. I assume youre talking about getting an error message after trying to run this command: conda env update -f environment_gpu.yml Is that right? If so, what is the exact error? What happens?

@@jasonzdora Thank you for everything, I just solved the problem. I got stucked on the installation of the folder "neural-amp-modeler". I tried all the time to copy it from you document, then I went on the hompepage and got the adress from github and it worked. Thank you.

@@FFJoe Ah, ok! Sorry for the trouble. Glad you got it figured out! Happy modeling!!

Probably this weekend :)

@@jasonzdora oh yes !

Follow the links in the description! Its pretty awesome :)