What Is a Neural Network and will it steal my job?

A neural network is a mathematical system of interconnected nodes that process information together. It's not a chatbot, and it probably won't steal your job, maybe… Loosely inspired by how neurons work in the brain, these systems take in information, process it, and learn from it and reproduce it in real time.

What matters for us: neural networks can learn irregularities and non-linear behavior — the stuff that gives analog gear its personality.

Alright... enough of the nerd talk.

Why does any of this matter for audio?

Traditional DSP plugins are, for lack of a better term, approximations of analog hardware. Through measurement, testing, and clever coding, we can reproduce a lot of the mechanical, tonal, and harmonic behavior of analog gear. And there are incredibly talented developers making amazing-sounding tools this way.

But the hardest part of emulating analog equipment has always been its non-linear behavior. Saturation, program-dependent compression, harmonics that shift with level, the way components interact and push back on each other — behavior that changes constantly depending on what you feed it.

Here's the honest distinction: traditional DSP is hand-coded from idealized equations of how a circuit should behave. A neural network is trained on measurements of how a specific unit actually behaves — component tolerances, drift, quirks and all. It's the difference between modeling the schematic and modeling the hardware sitting in front of you.

The Veritas Neural Engine

Veritas is a hybrid DSP/neural architecture. DSP handles everything it does best — filters, gain structure, timing, the linear backbone of the circuit — because when DSP can do the job, it does it exactly and at near-zero compute. The neural side handles what DSP can't: the non-linear behavior of individual components, learned directly from measurements of the real hardware.

This is component-level modeling, not a one-network-fits-all approach. We don't feed a whole waveform through one giant model. Instead, small networks act as specific components in the circuit — a FET here, a transformer there — each replicating the measured behavior of that part in that position. Just like real hardware, components vary and their behavior depends on where they sit in the circuit, so every plugin gets its own circuit-specific design inside Veritas, fed by extensive measurement data captured throughout the circuit.

The payoff is efficiency. Large end-to-end models waste enormous compute redoing work DSP could do for free — and nobody's buying a plugin that eats 50% of your CPU for one stereo instance. I wouldn't.

At the end of the day, these are tools built by someone who genuinely loves audio, engineering, and the process of making records. As a studio partner, audio engineer, and professional musician, I'm as picky and critical as the rest of you.

I sincerely hope you enjoy using these plugins as much as I've enjoyed creating them. If not, the lost sleep and caffeine intake were all for naught.

Listen To Your Ears.

— Brandon Skeens
Founder, AEFX Plugins