Way back in the mists of time we used to call the late-1970s, when I was a student at Sheffield Hallam University (or Sheffield Polytechnic as it was known back then), I was a member of the Joe Cool Road Show. In addition to providing the sounds and lights for on-site discos and visiting bands, we also provided support for local bands in various venues around the city.

One such band was the English synth-pop band called The Human League. When the band first formed in 1977, they had their own electronic musical synthesizers, but they couldn’t afford the numerous industrial strength amplifiers, speakers, lighting towers, and such like required for live performances, so that’s what we brought to the party. I still have a photo somewhere showing The Human League on stage with a speaker stack in the foreground and “Joe Cool” stenciled on all the speakers. This also means that—in addition to being one of the leading engineers in my generation (according to my dear old mum)—I can also lay claim to having once been a roadie for The Human League.

One of the things I remember about those days was the glorious sound we used to obtain from the vacuum tube (valve)-based amplifiers. I’ve heard it said that tube-based amplifiers introduce even harmonics while semiconductor-based amplifiers present odd harmonics. I have no idea if this is true or if it’s an old audiophile’s tale. What I do remember is the richness and rotundness of the sound those amplifiers gave us, 

Why am I waffling on about this here? Well, even though we currently tend to bounce around telling each other how we now live in the digital age, the wibbly wobbly world of analog still has the annoying ability to make its presence felt. It turns out that many musicians—especially guitar and bass guitar players—still drool with desire over tube-based amplifiers. But wait, there’s more, because even in studio applications like equalizers and compressors, analog electronics (both tube-based and solid state) are still the gold standard when it comes to sound quality.

And, when you come to think about it, it’s not just quality we’re talking about, it’s also what we’re used to. If you are a guitar player and you’re super influenced by Led Zeppelin, for example, then the sounds you have in your head are sounds that were created by instruments, amplifiers, and speakers that were engineered decades ago.

Furthermore, these sounds were often created by not using the amplifier in the way it was originally intended. The original purpose of a guitar amplifier was to make one’s electric guitar sound louder, but then people started saturating (overdriving) the amplifier to create clipping and distortion. One way to think about this is that the engineers of yesteryear spent a lot of time and effort trying to keep things as linear as possible, while the users spent a lot of time and effort taking things outside of the linear realm to create interesting acoustical effects.

In a crunchy nutshell, all this boils down to the fact that a lot of people want access to classic (or even modern) tube-based amplifiers. One problem is that many musicians can’t afford these tube-based tempters, which can cost thousands and thousands of dollars. Another issue is that, even if you are fortunate enough to call one of these bodacious beauties your own, they weigh “tons of kilos” (I hope I’m not being too technical), and you typically cannot carry one around with you everywhere you go. The solution is to create a model of the amplifier in question and then run this model on a computer, using it to simulate the effects of the real-world amplifier.

Well, that should be easy enough, shouldn’t it? Don’t be silly. Creating an accurate model of a tube-based amplifier is an incredibly complex and time-consuming task. “So,” you might ask, “is anyone actually doing this sort of thing?” Well, although this is admittedly a niche market, it turns out that there are more players than you might expect, but one of them stands proud in the crowd, head-and-shoulders above the throng, with amplifier models that sound indistinguishable from the real thing.

But, before we go there, let’s first consider the way most amplifier modeling companies tackle this problem. How would you set about it yourself? When I was asked this question, my knee-jerk reaction was to reverse-engineer the original design from schematics and (ideally) a real-world unit, creating analog SPICE models of all the components and then running an analog simulation of the entire amplifier.

Well, blow me down, this is just what most of these companies do. The problem here is that many of the brightest minds in computer science are not particularly musical. Contrariwise, many of the people who are particularly passionate about music tend to be a bit “ho hum” when it comes to the computer side of things. This is a conundrum indeed.

Using analog simulation to model the generation of audio signals is mind-bogglingly complex. Things in the real-world interfere with each other in weird and wonderful ways (e.g., transformers coupling with other components based on their physical layout in the system). Meanwhile, the trained ear can detect subtle nuances that are almost impossible to describe, which makes them well-nigh impossible to model.

The result is that even when you’ve managed to build a crack team of the brightest and best musical computer gurus around (good luck with that), it’s still going to take an inordinate amount of time and effort to create even a reasonably accurate model of just one tube-based amplifier. This is an issue because there are hundreds of little rascals crying out to be modeled.

All of which leads us to the fact that, earlier today as I pen these words, I found myself chatting with Douglas Castro, who is the Founder and CEO at Neural DSP Technologies. This is the company I mentioned earlier when talking about companies that create tube-based amplifier models when I said: “But one of them stands proud in the crowd, head-and-shoulders above the throng, with amplifier models that sound indistinguishable from the real thing.”

The Neural DSP models themselves are called Plugins. These come in two flavors, the first of which embraces specific amplifiers like the Parallax X, the Morgan Amps Suite, the Fortin Nameless Suit X, and so on. The second flavor—indicated by the “Archetype” prefix—embodies the sound profile presented by specific musicians and producers, like Archetype: Petrucci, Archetype: Cory Wong, Archetype: Rabea, and so forth.

These plugins can embrace the entire audio chain. Take the Archetype: Nolly X, for example. In addition to the amplifier itself (there are four to choose from), this plugin includes pre-effects (compressor, overdrive-1, delay-1, overdrive-2), a 9-band graphic equalizer, post effects (delay-2, reverb, transpose, and doubler), and more, all for (what I consider to be an unbelievably low price of) only €135.

The plugins can run directly on your own computer.  Alternatively, you can run them on Neural DSP’s Quad Cortex, which is billed as “The most powerful floorboard amp modeler on the planet. There’s also a software application called Cortex Control that runs on your host computer (Windows or Mac) and provides a graphical user interface to complement the Quad Cortex’s physical controls.

So, how are these awesome plugin models created and how can they be so amazingly accurate? Well, it turns out the folks at Neural DSP decided that creating SPICE models of tube-based amplifiers required too much time and effort for too little reward. Their solution was to create a suite of artificial intelligence (AI) machine learning (ML) models, each representing a different tube-based amplifier architecture (numbers and types of controls—stuff like that).

Training these models requires data—lots and lots of data—and that’s where TINA (Telemetric Inductive Nodal Actuator) comes into the picture. Speaking of which, here’s a picture of TINA:

Meet TINA the data collection robot (Source: Neural DSP)

TINA can robotically access the entire spectrum of every control’s range by physically connecting with those controls via actuator arms. The folks at Neural DSP have curated a suite of sound sources, including different types of guitars played in different ways. These sounds are fed into a real-world amplifier while TINA systematically investigates every combination and permutation of controls with the amplifier’s outputs recorded and annotated with the control positions.

Once sufficient samples have been gathered, they are used to train a neural network model, which can subsequently replicate the behavior of the device with hitherto unobtainable precision, accuracy, and fidelity.

As the folks at Neural DSP say: “The collected data is always a complete representation of the device and its history; every tube, every transformer, every pot, every ding, and every scratch; anything you can hear and feel will be a part of the data the models are trained on.”

I was just bouncing around Neural DSP’s YouTube Channel when I ran across two videos pertaining to Archetype: Nolly and Nolly’s Sonic Journey (Behind the scenes making of Archetype: Nolly).