After spending a couple of days with NAM and Proteus (and they do sound good), and then running S-Gear alongside them, all without using their own built-in effects or cabs, into Mikko2 free cab vst, into Delay and Reverb busses. I am shocked at just how good S-Gear is. I've been using S-Gear from its beginning, and have owned a Fractal and a Kemper and many, many sims, S-Gear has always remained my go-to home studio setup.
I don't mean to gush, but S-Gear is the real deal for great tones and those who haven't tried it, really should spend a few hours with it before purchasing anything else. Mike knew he had a winner when he released it, and he was right.

Thanks for the kind words!

I think what can be done with training a neural network to emulate an entire guitar amp is amazing, exceeding all my expectations. I've played with both NAM and ToneX, and in fact I've been experimenting with a NAM module inside S-Gear. I have to say also that I'm very happy with how the S-Gear amps stand up alongside. And why not? I'm often told by producers and engineers that they've replaced the real amps in a mix with S-Gear. Last week I posted a short article with Scott Sharrard and co-producer/engineer Charles Martinez, Charlie happily confesses that he'll often use S-Gear from the get-go, not just to fix troublesome recordings.

Here are a few of my thoughts on the new Neural Network based techniques…

Right across the spectrum of applied science and engineering, machine learning is coming of age and is set to have a huge impact. Stuff that for which previously we laboured over really complex mathematical representations, well now they can often be more easily solved by training a neural network on the data. This is a big game changer for many engineering disciplines. I have to admit, I had pretty much ignored this area up until a few months ago. Neural networks was more a data science thing, and it wasn't part of the mix when learning about audio engineering, where the focus was more on understanding the nature and theory of electronic and acoustic systems.

The technology in NAM (and probably most others) originates from WaveNet developed by DeepMind (a google company). WaveNet was developed with the goal of synthesizing realistic human voices using neural networks and machine learning. The model can also be trained to emulate amps (and some other audio systems) by using an input and output waveform and specifying a time period over which to consider historical data. For most guitar amps, probably you need a data window of up to 200msec to capture all the time constants, but perhaps you can capture most of the characteristics with a 50-100 msec window. This is a critical trade-off in terms of the processing power required in the final neural network model, and the amount of training time required.

This technology is largely open to everyone, so I'd expect to see rapid improvements and more optimised implementations come along soon. Right now, NAM has some limitations, a fixed sample rate, limited history window and some questions over aliasing. But again, I think improvements will come quickly. In time we might even see some clever machine learning of amp control knobs too.

I'm also interested in how this technology can be applied to create better component emulations, and possibly better software-based amp designs overall. It might also be used to model the power amp section of an amp (since this usually has only one variable parameter, presence), then drive that with a physically modelled pre-amp. There are lots of possibilities! I've always been interested in being able to design amps and effects, that's still what I hope to do, even if there are technologies to perfectly replicate physical devices.

Of course, I do also expect that the market for amp software will become even more difficult to operate in, especially given all the other economic and political factors we're living through currently. It is a concern, and we'll need to think carefully about where to focus the development effort, and how to best market what we already have.

It's an awesome prospect that there is now so much amazing technology at our fingertips. But the task of curating sounds isn't necessarily going to get easier when options are exploding. In some ways, I feel it is a huge distraction for anyone trying to learn to play guitar these days. Back when I first picked up the guitar, I had nothing but a cheap practice amp (and used the overdriven input of a tape deck to get my distortion). Once I could afford a half-decent amp, I was still stuck with a very narrow sound palette, you just make it work! If you learn a classical instrument like the violin or cello, its just you, the instrument and your practice time 😉

It's important to highlight the fact that you're so open to up-and-coming technologies. I think that attitude is part of the reason that you managed to make S-Gear so great from the beginning. The guitar community is known for being occupied by people who are passionate about very specific things, but perhaps with a closed mind, committed to what they like at that given moment. Many designers/developers would view neural networks as something to be ignored or even shunned -- I'm excited that, instead, you're embracing whatever advantages it might bring to the table.

I don't know what S-Gear might "need" in the future, but it's comforting knowing that you're always thinking about it.

Interesting discussion. I just got a promo mailer from IK Multimedia (Tonex) boasting that they now have over 10,000 tone models on their sharing site ToneNet. Who has time to sift through all those amp models and still play guitar or write songs? We are indeed drowning in possibilities. A similar situation applies to DAW plug-ins. There must be hundreds of compressor plugin choices alone. I like having choices, but analysis paralysis sets in.

I've forced myself to whittle down amp sims. In a given guitar track, I used to toggle between dozens of options. Now I limit myself to S-Gear, Tonex (with amalgam audio captures of a few favorite amps), and NeuralDSP Cory Wong. If I can't find the perfect tone quickly between these choices, my guitar track probably sucks and I need to spend more time practicing...

You are right about having too many choices.
Whatever I try I always coming back to S Gear because I have it for so long that I know how to tweak it very fast to my needs.When I say tweaks I am talking about very minor and normal things you have to tweak on a real amp anyway.
As far as Capture technology I own TONEX and a TONEX pedal and apart a couple of specific Captures I bought I will use it to Capture my own S Gear presets to be able to have them in my Tonex hardware pedal when I do not have a computer to play S Gear.
When it comes to the software version I really prefer S Gear because it reacts like a real amp when your play with the knobs.
I compared my captures to real amps captures on Tone Net and still like my S Gear captures more LOL!

Can you tell me how you capture software like S-Gear presets?

A.F. wrote:

Can you tell me how you capture software like S-Gear presets?

See this thread: www.scuffhamamps.com/forum/5-the-lounge/...ng-s-gear-with-tonex

As I have written before I am so glad that S-Gear does not offer 100 amps and 200 IRs! There is such an inherent risk to get trapped in the "endless-possibilities-hole".
When tube amps were designed the technicians used the parts that were available and tried to create a unique sound for their brand. There are many good sounds delievered by these "originals" of the tube era. However, I do like the sound because of the sound, not because of the amp and its name.
E.g. I love the American Clean tone in S-Gear. And I don't care which physical amp sounds like that (prob. a Fender twin?!).
I think that's somewhat Mike's approach, as far as I have read.

The past months I spent more time in setting up GigPerformer hosting S-Gear and MidiGuitar to get good sounds with various amp-speaker combinations than playing guitar. Now I got a setup where I can chose from around 50 good sounds in a second (by using a midi pad controller), which works with a self built light weight stereo cab powered by a cheap class d amp. People make compliments how good my sound is. My ears tell me the same. So I'll try to settle with these sounds and PLAY instead of porgramming!

Just my 1/50 Euro.

mike wrote:

Thanks for the kind words!

...

It's an awesome prospect that there is now so much amazing technology at our fingertips. But the task of curating sounds isn't necessarily going to get easier when options are exploding. In some ways, I feel it is a huge distraction for anyone trying to learn to play guitar these days. Back when I first picked up the guitar, I had nothing but a cheap practice amp (and used the overdriven input of a tape deck to get my distortion). Once I could afford a half-decent amp, I was still stuck with a very narrow sound palette, you just make it work! If you learn a classical instrument like the violin or cello, its just you, the instrument and your practice time 😉

Hahaha, I did the same to get a distorted sound... Back then I only had one clean amp and no pedals.

We are all waiting to see if you are going to eventually release some hardware versions of S-Gear. It's such a good sounding tool! If that happens, I will be one of the first folks to have it. I am not even looking anymore at all the offers out there... I trust S-Gear and am very comfortable with it, and it continues to surprise me from time to time.

Thank you once more for your hard work and amazing product!
Cheers,

sorry but i disagree with the OP,S-Gear is a cool sounding plugin but the NAM plugin is another great sounding plugin,plugins like Helix Native,Neural DSP,Softube,ML Labs,TH-U,Plugin Alliance,Amplitube,Tonex,Bogren Digital,STL all those plugins in my opinion are great plugins.

i love S -Gear too but im not deaf to admit out there are awesome amp plugins.

i own the Axefx 3,Kemper and QC but i like to play with plugins too

im not a fanboy of any device or plugin,i just like to have fun with my gear and plugins

I have three general issues with ML amp modeling. First and foremost is option paralysis. Look at how much time we’ve wasted auditioning IRs only to circle back through the same ones over and over, never being able to differentiate them enough to choose one. Now amplify that with captures (pun intended). The combination of the two is likely to have a significant negative impact on song quality. The second is more technical. Machine Learning is a way of giving a training set of observed input and output data, and letting a computer figure out a model that produces the output from the input. That’s great for practical engineering. But it doesn’t lend anything to our understanding of the underlying physical, social or economic process that might be involved. So maybe ML may have the unintended consequence of reducing human learning. And third, ML and captures only reproduce things that already exist. They don’t invent anything new. This is what I especially like about S-Gear. It doesn’t try to just emulate existing tube amps. Rather it exploits the digital domain to create guitar amps that might not be possible, practical, or economical as physical amps. So I’m still kind of leaning towards the explicit modeling of things, so we can learn, innovate, and create new things, not just exploit what has already been done.

I very much agree with this approach Masden, in other software you get lost with the multitude of options they have. With the advent of machine learning, it's good if things are done right, but if not, we're still bound to waste time auditioning all the captures made by users. Having said that, I think that if these new technologies are implemented well in a software in which learning is carried out with rigor and reliability, perhaps a good advance. It's said a lot, but it's true, we always come back to s-gear, and although we always ask mike for more options and models, we have everything we need to make music. Because in the event that I were to move what s-gear gives me to the analog world, I could not have 4 different models of amplifiers, nor all the screens that provide me with the IRs that we have available. The vast majority of times, the word more does not mean better, as a saying from my country says, he who has more money is not happier. Health and music

Good points.

I think sometimes we look to the software to create the perfect tone whereas we often just need to spend time working with the guitar or the part we're attempting to play. Or at least tweaking the amp in front of us before rapid-firing through 1,000 presets.

As tends to happen, technology gets overwrought with options, then people get exhausted and it retreats back to its simplistic roots. S-Gear never took the overly-complicated approach, which we seem to appreciate.

jamsden wrote:

This is what I especially like about S-Gear. It doesn’t try to just emulate existing tube amps..

This ^^ +1000

Been giving Tonex a shot for the one thing i can't quite get out of S-Gear which is that Vox thing. Given how well S-Gear has held up across a very wide range of projects is only highlighting how awesome Mike's take on an AC30.might be. It might not seem like it's worth the effort these days given the slew of options out there, but it sure would make perfect even more perfect.

jmho, of course, but have to add that I've done a few projects with an old school LA session pro who, despite his collection of (self) modded amps and 35+Telecasters, has acknowledged that there's "something" to S-Gear. can't get him to even look at anything else.

If you'd like an excellent set of tone models of AC30 for Tonex, try the amalgam stuff at amalgamcaptures.com He used a nice copper panel Vox TB30, and it sounds splendid.

If you'd like an excellent AC30 for Tonex, try the amalgam captures at amalgamcaptures.com He used a nice copper panel Vox TB30, and it sounds splendid.