Modelling can do things I can't

What hardware sampling actually is

Sampling a hardware device means recording reference tones through it, then writing software that reinterprets those recordings to recreate the response of the original. Instead of building a virtual circuit and validating it against the hardware (if you have it, which is not always the case), you record the hardware's response itself.

I use a blend of convolution and neural networks, fed by a very deep capture of each unit and tuned by ear against the hardware and my own taste. It takes a long time. To my ears, the results are worth it.

Limitations of hardware sampling

What you give up is flexibility. A sampled effect inherits the limits along with the sound: if the original stepped in six positions you get six positions, and if one of its bands was always a little off, it's still a little off. You're keeping what makes the hardware sound the way it does.

Sampled plugins also cost more CPU than algorithmic ones. That's true of the Felt plugins, and of Acustica Audio, another company sampling hardware and one I have a lot of respect for.

What justifies the cost is the sound. When you deliver the final mix, nobody knows what your CPU meters showed. They hear how the record sounds. And of those two problems, only one gets solved by computers getting faster every year.

Little things that add up

A frequency response is the easy part of a unit. It's a curve. You can measure it and match it. What takes longer is how the thing behaves in time: the phase relationships between bands, and the first few milliseconds of a transient. That's a huge part of what your ears use to decide whether they're hearing a device or a picture of one.

Ask what makes hardware sound better than plugins and most people say saturation. That's partly true. But a hardware EQ might also compress a little, because of how its tubes, transformers and capacitors respond. It might ring slightly, a couple of milliseconds after a strong transient. Nobody will tell you a unit sounds great because it has a 1 ms decay after a transient, but little things like this add up to the bigger picture.

Sampling captures all of it, and with neural networks the saturation comes too. To my ears it lands at about 98% of the hardware, which is about as close as a plugin gets. Add some noise and you're basically there.

It also applies to instruments

The same split runs through instruments, under the name physical modelling. A modelled piano computes strings, hammers and a soundboard. A sampled one is recordings of a real piano, played by somebody, in a room, through microphones, preamps and cables.

Modelling wins there the way it wins on effects. Every note responds exactly as the model says, you can change the instrument's behaviour from a panel, and nothing is tied to what happened in one room on one day. What it doesn't have is a person in it. A sampled instrument done well is somebody's hands on a real thing, with all the unevenness that brings, and that's the part a model has to invent.

You could argue that you're the human part, playing the modelled instrument. Partly true: the performance is yours, and that's the beautiful bit. What's missing is the sound of fingers touching keys, the microphone, the air around it. The creak of wood, or your breath. No model provides that.

How much do these details matter? That's personal, and everyone's answer is different. For me they matter, and in the era of AI-generated music, more than ever.

So which one do you want?

It comes down to what you're asking the plugin for. Modelling gives you more flexibility, lighter CPU use and more features. Sampling gives you a truer sound.

Getting to where I can say that comfortably took a few years of trial and error. Before the hardware bug got me, I used the plugins everybody recommends. The a-ha moment was discovering Acustica Audio. Their sampling approach set me off down this road, and it took years to arrive at the techniques I use today. Being a hardware guy now, I'm very proud of how the Felt plugins sound.

In the end it's aesthetics and personal preference. If there's one rule worth following in audio: if it sounds good, it is good. Trust your ears.