Yes, there have already been such experiments and in fact the refining effect was better using such a signal. Maybe it's a lead.
It is possible that this could be the officially recommended method with the ready file available on the KPA website? But you'd probably have to test it hard on a few cases. It would rule out some arbitrariness and thus some disappointment and uncertain effects of profiling for some users.
capture tuning could be more detailed to faithfully do full gain sweep from 0 to 11, etc.
I guess this concept is used in proteus.
its good enuf for me thanks
That's great! Well for sure you can't capture your pedals and run it in daw with more NAM instances to make whole guitar chain 
The free modelers, NAM and Proteus, suffer from tying a player to a computer, either laptop or desktop.
Yes and no. There are a DIY and comercial projects that implement NAM or AI core to dedicated hardware or open hardware like Rb pi4 . Even on DaisySeed board. Yes - those are not mature solution - yet. But people gigs with laptops. Author Steven Atkinson was gigging with a laptop with NAM.
So what happened to Kempers patent, did it run its course? too expensive to defend, I know nothing about these things, just like women.
I think the patent was very specific about one type of modeling / profiling. But I don't know much about it either. It would also be difficult to fight in the era of AI progress to limit this (guitar amp) area when there are already applications modeling the tone of voice of famous characters. Of course, I'm just speculating.
It would be enough not to specify what non-linear distortions the plugin imitates.
You could call it a plugin for profiling the non-linear black box response.
In fact some guys are profiling studio preamps and tape recorders with NAM.
There are another project focused on profiling audio compressors (time dependent response) :
https://csteinmetz1.github.io/tcn-audio-effects/
Plugin : https://github.com/csteinmetz1/neural-2a
Profiling: https://github.com/vincnt/tcn-audio-fx https://github.com/csteinmetz1/micro-tcn
There is a lot going on in this field and I don't think any patent can stop it. And rightly so - because it would stop the development only for the profit of some company.
This is list of audio ai startups : https://github.com/csteinmetz1/ai-audio-startups
Im hoping Kemper just upgrades the profing algorithm or whatever it needs too. I highly doubt he is going to lay down an be beaten in the field an market he developed. Not going to happen.......
I would like to see that. I'm all for it.
I'm just trying to look at it realistically.
I think DSP56300 (200 MIPs/core vs 2000 mips for rasberry pi 4 for example) is too slow to do the math with NN. Maybe another new product?
Besides, in my case, I prefer to have access to a solution in the form of a plugin because of what I wrote about - working in a DAW. Reamping each track with one device would take a lot of time. And anyway, I prefer to track the sound of all the guitar tracks at once. But as a live solution, absolutely.
So my last word. I don't know what Chris has some ace up his sleeve. From my observations, KPA has already become a mature product and will become a classic in the world of Guitar-Audio like Acces Virus , and the whole team will go to work on another project, not necessarily about guitars. I don't think the KPA design will suddenly change and I don't believe the team will be able to implement neural network playback in assembler. It comes from the fact that Chris just made something beautiful out of what he knows and which was already a bit outdated at the moment. This shows how you can squeeze the last juices out of this technology. But I have no illusions that the world will go in a new direction and it will be neural networks.
The only solution I see is for Christoph's mind to just use the current tools and use it as creatively as in the case of KPA. Automating additional parameters for the amplifier, etc. But this will require giving up independent participation in programming or learning how to program neural networks and working on it. It seems impossible to me at the moment. I'd like to be wrong. P.S. Both the acces and KPA products are based on the same processor architecture.
real time kernels seems to be much better at this, windows is struggling with inconsistent results , given cpu type, buses, soundcard & drivers...
the 1st company putting NAM in a pedal format will win the race ...
Does NAM computing take place in the CPU or in soundcard's DSP ? staking plugins leads to cracks as soon as I hit 5% of my ( old ) cpu.
There are some DIY implementations and comercial like MOD DWARF.
Wheresthedug there must be something wit your setup. I've checked NAM at my old laptop and with buffer 128 have 20 ms latency.
So true. The same with plugins. The latest ssl emulation from dev A is so much better than old Waves version you hear from many wanna be pro on forums. And still that old Waves version is used by some pro mixers and they get that pro sound amateurs dream about. Or they just use stock plugins in their daw. Anyway most pro mixers approach is learn your tools you have now and stick with them and don’t waste time or money on the latest stuff. It won’t make you sound better. It’s not the tools. It’s your knowledge that Matters. Your or your band won’t sound better with the latest and any audience/listener won’t hear any difference nor will care what’s been used. All they care is if they like your music or not.
But for studio use just for reuse the tone :
1. the more faithful the better since you need the same tone as with live tracking with AMP( you need the same tone not almost the same to do some punch ins and so on)
2. the faster and less problematic profile/train the better (refining is very problematic process)
3. You can track and bounce several tracks without the need for expensive equipment and without wasting time
So for live - yes KPA is very well designed ecosystem
For studio use NAM is better in terms of time and reproduction fidelity and simplicity of making "profiles"
For training you have to follow the steps :
1. First instal Anaconda : https://www.anaconda.com/
Then follow the steps from github site
: How to use (Local)
Installation uses Anaconda for package management.
For computers with a CUDA-capable GPU (recommended):
conda env create -f environment_gpu.yml
Otherwise, for a CPU-only install (will train much more slowly):
conda env create -f environment_cpu.yml
Then activate the environment you've created with
conda activate nam
Train models (GUI)
After installing, you can open a GUI trainer by running
nam
from the terminal.Wheresthedug please beare in mind that for now plugin will only work with 48kHz SR session as training process was designed fot that SR . This will change as Steve will implement oversampling in the plugin to support other sample rates.
DamianGreda I downloaded Version 5 from Github. What do I do now? I want to give it a try but it looks like I need an IT degree to even install it. Definitely intimidating to me before I even get started. Tried reading the ReadMe file. Ran the Setup.py file. But still don't know how to use it. What do I need to do?
You want to install vst3 plugin or setup an environment for training your own amps?
You can download plugin here: https://github.com/sdatkinson/…lugin/releases/tag/v0.7.1
There is a one but big advantage with NAM over the KPA .
There is no magical and difficult refining process. It sound's good right away.
And this is the funniest thing - that suddenly people come from nowhere and try to show how stupid you are. That's an ad persona argument.
Not a merit argument.
And another trick here is to allow an argument and the moderator closes the discussion. How convenient it is. I've seen it here hundreds of times.
Are you still getting on with it? haha. I'm having fun with this.
If by words and psyhology(sic) you mean mature and effective communication, then what you call too old I consider to mean too dumb.
Just more poor communication. You wrongly assume that I have not checked out NAM. Yup, too dumb for any effective communication.
Well you can call me dumb , I dont care.
[Blocked Image: https://media.tenor.com/nwoJ4BS0XHYAAAAC/oh-no-anyway.gif]
The same story over and over - i will not to play with you with that game of words and psyhology . Too old for that and seen this here several times. - just check yourself and spare me and some othes time will ya? Who cares if you can just check yourself for free.
Your text only shows who you are because instead of checking out a free plugin you are trying to undermine my credibility.
I don't care about any credibility in your eyes or anyone else's. I care about my credibility in my eyes. And that's what makes us different.
Yes, I know, I'm exaggerating. There will always be some users for whom the given equipment is useful for their needs. And this will prevail over the choice of a given equipment. Sonically, however, NAM's alborhythm is second to none in terms of fidelity. Although it doesn't profile everything well either. It has problems with profiling the amplifiers together with the loudspeakers. Works better with IR. But what annoys me on this forum is howling with closed minds at news. I suspect that the people from Kemper themselves look at this new technology with delight, but that they have to maintain sales, you know what .... they post that a cool artist here uses something or the like. It's part of running a business.
And I don't like to close my eyes because someone tells me that ours is better.
Our ears decide what is good - not the companies.
What's "funny" to me is how people and teams of given companies are fighting for the lead on the market, and here's one guy who, giving his knowledge for free, swept it all into dust.
but it still only runs as a plugin rather than hardware which is a major issue for me.
Yep sure - I understand. Guys are working to port this to some pedals Daisy seed , rasberry pi and so on . Author giging with his laptop. There are a lot of options. Plus in the studio you have a:
1. input wav and output wav to profile - train with futures versions
2. you can track real time
3. and fast bounce (this is con for kemper as you cannot bounce from HW )
Crazy times I tell you !;)
The sound quality may be phenomenal but as a plugin only tool and (if I understand correctly) a cloud based profiling workflow it isn’t any practical use to me. If it makes Kemper improve their profiling accuracy that would be a win for everyone though.
You can TRAIN with anaconda and repository from github locally on your PC. On CPU or better with some power GPU likeRTX 3x or 4X.
The hell send me a out file https://drive.google.com/open?…%40gmail.com&usp=drive_fs and I will TRAIN it on my GPU for free. Just simple as that.
