-
Lol. You know you’re not missing something. 95% of tone is your playing, and another 90% of tone the actual speaker driver. That’s 185% of tone covered.
Originally Posted by Skip Ellis
From a tech point of view I can say with authority that your amp must be able to go not just to 11 but to 18.5.
Sent from my iPhone using Tapatalk
-
06-19-2026 05:06 AM
-
I thought that Cliff was referring to the NAM file data that is compressed to be able to function in the very small Valeton GP-5 type pedals. These newer small pedals are not processing pure NAM files. Originally Posted by nevershouldhavesoldit
It’s not data compression. Most machine learning, AI, and modeling programs do not use their training data once they’re in unrestricted use. But no data are compressed in the way that 7-Zip, mp3, etc compress it. There is no compressed dataset that can be re-expanded into its original form. AI and modeling programs store summarized and stratified data (which are just complex number sets generated from real world input like sound waves, rates of occurrence, indexed information, etc) in hierarchical lists. But these are not compressed files of the original data. If you’re interested, here’s what I hope is a simple explanation that will make sense. Ignore the rest if you have no interest or bore easily
_____________________________________
Machine learning and AI use data schemes in which all of the data on which they base their “intelligence” are sorted, stratified, and prioritized by the contributions of each bit of data to the accuracy of the program’s output (the “weight” of each individual parameter in the data set). An initial set of algorithms is defined based on deep analysis of those data using forms of regression, recursive analysis etc to organize, arborize, and make sense of the data. During training, the computer constantly compares its output to the training data and adjusts its algorithms on the fly. Reassessing its accuracy after each run helps it to prioritize (ie “weight”) the strongest contributing data most highly and reduce the weight of those that correlated less strongly with accuracy.
Once most machine learning models are trained, they are put into production and their training data are no longer accessed. So technically, they’re no longer learning once they become the kind of AI we use in smart phones, digital assistants, etc. They store algorithms and the volume of highest weighted parameters needed to achieve their target accuracy rates (as determined by their designers and engineers). Some models (eg KNN) do retain their training data and search it directly to answer queries. Most such programs do not.
Picture an AI program to create a fake book on the fly that’s to be used by X% (target TBD by the designers, based on their business model and practical reality) of gigging guitarists. Jazz groups, wedding bands, country players, Latin bands, studio pros, etc. Subscribers could have it create custom books to be displayed on tablets for any given gig in any location, eg a trio playing a second wedding at a Napa winery, a 9 piece commercial band playing a white shoe law firm’s Christmas party in Manhattan, or a country band playing a retirement party for the owner of the oldest grocery store in a small town in Oklahoma, etc. Accuracy would be measured by the percentage of tunes used by subscribers on actual gigs, along with the number of tunes omitted but requested by subscribers. Since the output would be a digital fakebook accessed over the internet, actual usage and all kinds of feedback would be incorporated into the evolution of the model.
Training data might include set lists for wedding bands, night club shows, etc plus DJ request lists, lists of gigs by band / type / region, online reviews of bands across the country (with detailed likes & dislikes) etc. The computer might first sort the data into what songs were most played, then most popular, best liked, most disliked, most requested by brides-to-be, etc and then place them in different gig settings in multiple geographic areas.
It would then identify discrepancies such as a song that’s played often but not requested often or a song that’s almost always played at small high end weddings but almost never at fire hall wedding receptions. The creators, designers and software engineers would have to decide what’s important and in what order these all appear in the algorithms. Will they consider ethnicity, total cost of the affair, etc? How finely will they stratify for geography - by region? state? city? neighborhood?
The data tables are then arranged by the program to reflect the initial priorities (“weights”) of all of these bits of information. A decision has to be made - how big will the book be: 100 tunes, 500 tunes, or a floating parameter based on variables like top X% of tunes for a given gig, location, etc? The designers, engineers etc then have to decide how deeply the program will go into each category, eg do they use the top 10% or 50% or all the data? Then the model trains by spitting out its first set of fakebooks and comparing them to all the categories for accuracy and consistency.
If that first book contains songs played by fewer than 15% of bands but often requested, and the bands that play it have 50+% more gigs in the same area and kind of gig than those that don’t, the machine may increase the weight of band popularity and decrease the weight of how often that song is played. This kind of analysis is applied to each and every parameter in the database until the output is consistent with the designers’ goals for it. That consistency includes the rate of accuracy they want.
There will still be errors, which in this example means tunes that bomb for some bands, omitted tunes that should have been included, etc. So the model will need real time feedback and periodic retraining both to improve accuracy and to incorporate changes like new hits, fads, resurgence of interest etc. Users could request a missing song from the stand. Models like KNN that retain the full training set can retrain at will, but the data may have to be augmented and readjusted offline for those changes to maximize utility.
The same kind of process is followed to create models of sound etc. The data set includes far more parameters that I can detail here, but they’re all collected and archived from digital representations of the entity to be modeled. IR yields data. The audio output signal from a preamp yields data. Then there’s the issue of variable parameters like EQ, which is often digitized by capturing the output of the preamp at all combinations are permutations of the EQ settings. There are even robotic devices that turn the EQ knobs on the actual source device infinitesimally to permit capture of the full EQ spectrum with hundreds or even thousands of takes of the same material
Once the first training set is collected, those data are categorized, collated, and prioritized by letting the program develop algorithms from which to build the modeled output. Outputs are compared to the original captured sound and everything is tweaked to improve accuracy. In this case, the goal is fidelity to the original sound. If a complete amp model is being created, the controls have to work just as they do on the real thing. Etc etc etc.
-
No, I was referring to the act of training a neural network to be a form of compression, but obviously haven't explained myself very well. Once you have your neural net, you can further compress the data using a process called distillation, where a smaller network is built from the larger one, and the least significant nodes are discarded. Also, as Never pointed out, you can further compress the weights from full floating point values to equivalents with lower precision, using less memory and enabling more of them to be processed in parallel. Originally Posted by GuyBoden
-
The advance that makes it possible to run NAM on small devices with limited processing power like the Cortex CPUs in Raspberry Pis is A2 Lite. The full A2 NAM system uses 8 processing channels and the Lite version uses 3. This is not compression - it’s simplification. There’s a slight loss of fidelity compared to the full A2 pro version. Originally Posted by GuyBoden
They’ve managed to put both in the same program, so there’s no difference in file size. The device “activates” the correct version for its CPU, since processing power is the limiting factor in use of the full 8 channel process.
-
The thing is, I can run the same NAM file on a computer DAW and a Valeton GP-5 pedal. There isn't very much overall difference in sound (to my old ears), the main distinction is that the Valeton GP-5 is quieter, so you need to turn the volume up. Originally Posted by nevershouldhavesoldit
I've used NAM files on a DAW for a few years and recently on inexpensive pedals.
-
The reason you can run the same NAM file on both is as I explained in post # 79. There's only one file, not two. The full A2 version contains the A2 Lite version, and the device determines which to use. The full version won't run on less powerful CPUs like the ARM Cortex family. The LIte version only uses 3 CPU processing channels and the full version uses 8. There's not a huge difference in sound quality, but it's definitely audible when you compare the two side by side. I've only heard a NAM pedal a few times. But the differences among the best ones (and between the full and lite processes) seem to me to be no greater than they are among many amps that are the exact same year and model. No two Dumbles sounded exactly the same, so I suspect that this isn't a deal breaker for most players. Originally Posted by GuyBoden
I've never used a NAM pedal on a gig, so I can't confirm this. But everything I read and hear says that the lite modeler diverges further from the real thing in fidelity as the number and complexity of effects increases. Distortion based effects are apparently among the hardest to model perfectly. The clean amp models are easier to create and more faithful.
-
Ok, NAM 2 is very different. Originally Posted by nevershouldhavesoldit
The Valeton GP-5 isn't NAM 2 compatible. It converts the original NAM file format using it's SnapTone software app to a format that can be used on the Valeton GP-5 pedal.
I use NAM files every day. I think that they are very good.
I'm a Software Engineer, I'm a member of the Open Source community.
-
I thought I read on the Valeton website that the latest firmware for the GP-5 enables native A2 file use. I apologize if I'm wrong and encourage you to download the latest firmware if I'm not. I know that Chaos, Darkglass, and PiPedal all run A2 NAM files as A2 Lite. Originally Posted by GuyBoden
Ok, NAM 2 is very different.
The Valeton GP-5 isn't NAM 2 compatible. It converts the original NAM file format using it's SnapTone software app to a format that can be used on the Valeton GP-5 pedal.
I use NAM files every day. I think that they are very good.
I'm a Software Engineer, I'm a member of the Open Source community.
I also have a bit of experience with software design and creation. I started with Fortran II on an IBM 1620 in 1965 and have remained current on modern platforms ever since. I was certified in Cold Fusion by Macromedia University about 25 years ago, and I've been an open source creator and user since converting an Apple Mac 2e to Linux in 1984. Along the way, I learned Ruby, Rails, the Cs, SQL, Python, Java, Javascript, etc, My specialty is healthcare software. Several major medical centers in the US are currently running systems I created for them. Since A2 Lite runs on Raspberry Pis and I have several, I'm working on turning a Pi 4b into a modeler using PiPedal.
I have a few gigs coming up over the next several weeks that include some pop tunes among the jazz. I've posted my Quilter Overdrive 202 for sale, since I no longer play big weddings, blues festivals etc. So I'll use myToneBlock on the few such gigs I get, I'm ordering a Valeton GP-50 pedal to see if it makes a better Dumble than my Zensation pedal. There are also a few effects I wouldn't mind having for select tunes.
-
Excellent. Originally Posted by nevershouldhavesoldit
-
Me too to see how its dumble compares with my Amplified Nation SSS and ODR Originally Posted by nevershouldhavesoldit
also cause it’s such a swiss army knife including 20 second looping and drum machine. And yes I’ll take a quick look at NAM downloads but won’t live there. I’m a happy luddite.
-
A Luddite with a modeler is kinda like a blacksmith with a laser welder Originally Posted by Woodstove
Me too to see how its dumble compares with my Amplified Nation SSS and ODR also cause it’s such a swiss army knife including 20 second looping and drum machine. And yes I’ll take a quick look at NAM downloads but won’t live there. I’m a happy luddite.
-
This place is horrible for my GAS. I've enjoyed playing with NAMs on my Beam Mini and just ordered a GP 50 because of this thread Originally Posted by nevershouldhavesoldit
-
That’s what you get for hanging out with a bunch of degenerate miscreants! Originally Posted by Jthompson48
This place is horrible for my GAS. I've enjoyed playing with NAMs on my Beam Mini and just ordered a GP 50 because of this thread
-
I'm going to drag my feet on buying hardware. While I like the idea of having some standardization I still have no experience with profilers. I'm still bugged by the idea of a static image. Even if there are a zillion static images to choose from. Sometimes you have to learn new things and that's the plan. Surprised to see there are already choices in the $100 to $150 range.
I also look forward to getting reviews of specific NAM captures in this forum where we'll have players looking for various jazz tones. I would like to try a bit of Carr Rambler, Victoria 4X10 tweed, Matchless, and a long list of Fenders. Even a few solid states. There will likely be just a few good captures, some bad ones, and many in between to report on. It's going to be entertaining.
-
Link for the Fender Deluxe Reverb profile 'NAM 2' file by Steve Atkinson, creator of NAM Originally Posted by Spook410
Fender Deluxe Reverb (A2) NAM Profiles by @sdatkinson
(2025) Interview with Steve Atkinson, creator of NAM.
AI and the Future of Guitar Tone: Neural Amp Modeler Interview
-
"Q: But you're just capturing just that one tone, right? Doesn't that create some limitation with what you can model with NAM? Originally Posted by GuyBoden
Atkinson: In a way, yes. But limitations can spark creativity. I like that. There's artistry in saying, "Here's the version of this amp I want to preserve." NAM really offers you a snapshot. It doesn't give you every knob and option, but it gives you the exact tone you dialed in. It's your best foot forward. Of course, parametric NAMs are possible too, where you model controls like gain or EQ, but I think the snapshot idea is really powerful for musicians."
So it appears I was right, the approach takes a snapshot with a single amp setting. He also mentions the capture is like impulse response capture of speakers, but for the entire amp. The problem with IR is it assumes the system being captured is linear, which is not the case for tube amps.
FWIW, I'm not trying to dismiss the tech, it obviously has its uses. Just trying to understand how it works and how it compares to alternatives. (I'm a programmer too, btw.)
-
Cliff it's easy, download a NAM file into your DAW and see if you like the sound. Originally Posted by CliffR
-
I think this is really where the value of modeling lies for most of us. I’ve spent decades finding and buying or borrowing amps and speakers to learn how they sound and if they’d meet my needs and fulfill my desires. Having a box that will let me try Carrs, Victorias, Dumbles, Champs, Two Rocks, Little Walters etc with a twist of the knob or a click of the switch is a dream come true. Originally Posted by Spook410
Once that becomes reality (and it appears to be mighty close), all I / we will need is a modeling pedal or preamp head and FRFR amps & speakers (or powered cabinets) with size and watttage for our playing environments.
Of course, all the debates and comparisons will then focus on the characteristics, neutrality, and accuracy of those FRFR systems and modelers. NAM is only the beginning. And instead of 37 amplifiers in a manufacturer’s lineup, there will be 37 powered cabs in multiple sizes with varying amp and speaker technology. As long as there’s hardware to be had, we’ll continue to debate, chase, accumulate, and trade it.
-
OK.. I give. Amazon Prime Day sale has the Valeton GP-5 at $64.
-
as you know, i embrace digital tech but none of those clips convinced me this sounds as good as my kemper Originally Posted by Navdeep_Singh
-
I think Ill return the one that arrived yesterday! But I may not get another even at this price. My brief experience with it yesterday was not impressive. Not close to the realism of UA Dream and Enigmatic that I have. But the real reason I got it was to test out the experience of downloading NAM And maybe as a looper. Originally Posted by Spook410
-
May I ask where you saw this price on prime day? I don’t see any reference to it. Originally Posted by Spook410
-
Originally Posted by Woodstove
I was planning to pick one up to find out if I like NAM modeling. But given your experience with it, I'm not going to bother. The GP-50 apparently shares the "sound engine" with the 5, so there's no point to spending even more for the same wine in a fancier bottle. But I'm now seeing this as an opportunity to try and learn about NAM and modeling in general, by buying a decent FRFR speaker like an Alto TS410 and a basic NAM modeling pedal.
I'm facing a dilemma, and I was hoping a GP-50 might solve it. I need a PA system with more power and sound quality than my cheapo 8" powered speaker can produce. But after my ruptured disc last year, I need one that I can safely carry. I'm back to working with my trio again, and we picked up a gig at a great new bar / restaurant starting in August. But they don't have a sound system. When I saw a Bud 10 for sale on the forum Sunday, my first thought was to sell my Blu 6 and buy the Bud. I can use my Toob Metro FRFR as the second speaker and run both the mic and my guitar through it. If I move the Quilter Overdrive 202 or the Blu 6 soon enough, I'll buy the Bud anyway (if it's still available) - the one in one out rule applies.
After your ringing non-endorsement of the Valeton, I'm favoring building a PiPedal to experience NAM, and picking up an Alto TS410 as both my PA and a FRFR guitar cab. I've been playing with single board computers since the Beaglebone came out, and I have a few Pis on the shelf. So I'll set one up later today and get started on the PiPedal. Stay tuned!
-
I don't expect the Valeton GP-5 to become my primary. Mainly I want to screw around with captures. Being able to do so with such a small investment is nice. If I like what I'm hearing from the tech I'll upgrade to something else as the market coalesces.
-
Technically a NAM A2 (or A1) capture should be more accurate than a Kemper. If that's true, seems we should be able to find NAM captures that sound as good as Kemper (or ToneX or Quad Cortex). Of course we have to sort through a rather large stack to find the specific ones we like. Originally Posted by jzucker
Reply With Quote
Guitar Amps built into case?
Yesterday, 05:03 PM in Guitar, Amps & Gizmos