I asked ChatGPT to build me a $1,000 effects pedal. This is what he did

effects pedal (noun) A little box of knobs, switches and dark magic that musicians stomp Continue to summon everything from angelic brilliance to absolute filth.

If you ever plug in an instrument or use a microphone (or would like to), this might be of interest to you. I’m in the process of purchasing a new bass, which inevitably led me to the dangerous task of looking for things to plug it into.

I’ve played with all the toys, so I know I want a decent multi-effects pedal. A box that gives me any sound I need without carpeting the floor with lots of little pedals.

Of course, the one he really wanted was going to cost more than what would pass the wife test.

After the recent V2 update, I stumbled upon TONE3000. TONE3000 is a huge online library of Neural amplifier modeler captures: digital models of real amps, cabinets, pedals and even complete signal chains. Think Fender, Trace Elliot, Marshall, Boss…

I wasn’t in the effects market at the time, so I thought, yeah, modeling, cool, and I went ahead. Great timing, thanks Alex from the Bitflip podcast for getting me excited about this and getting me on PiPedal. That was the magic device that put the TONE3000 back in my sights. The new Atlas looked at the same technology in Blackstar’s Beam Mini, incorporating thousands of tones captured from amps and pedals into a commercial product.

Enter PiPedal. A free and open source effects platform that turns a Raspberry Pi 4 or 5 into a multi-effects unit. PiPedal 2.0 now communicates directly with TONE3000, allowing users to search and download Neural Amp Modeler (NAM) captures of real amps, preamps, and pedals without leaving the software. In addition to these, there are conventional effects plugins covering reverb, delay, chorus, compression, EQ, distortion/overdrive and pitch shifting.

This was starting to look very, very interesting.

Turns out I had an old Raspberry Pi 4. And luckily, a spare USB audio interface. How fortuitous. What I didn’t have was much of an idea of ​​how to turn those things into the sophisticated effects equipment I now saw was possible.

All I said was:

And this is where the story stops being about a Raspberry Pi.

I didn’t spend the next few days reading Linux audio documentation, GitHub discussions, and MIDI manuals, like in the old days. I only described what I wanted. ChatGPT downloaded, installed and configured the software, diagnosed what wasn’t working, tweaked the settings, and gradually turned a spare hardware stack into an effects processor.

Have it boot directly into the pedal settings. Made.

Give me a bass sound for Sledgehammer. Made.

You could say “here’s my 40-track playlist, make me a preset for each one.”

And so it was.

In just a few minutes.

Amazing!

Within a couple of hours, I had a multi-effects pedal, individual song presets, amp and cabinet combinations, effects chains, and a touch interface on my iPad. Add a six-button MIDI controller pedal and, from a player’s point of view, it begins to behave like a conventional pedal.

Grief, stomp. New York minute, stomp. song of the earth, stomp.

The sounds are actually very good. The new Atlas linked neural network amplifier modeling in 2020, and now it’s so good that Alex and Stephen at Bitflip are discussing why, in blind comparisons, it can be extremely difficult to distinguish from the genuine article.

Generally speaking, I accidentally built an open source cousin of something like Neural DSP’s Quad Cortex, only mine was spread across a Pi, interface, tablet, and controller instead of wrapped in one gorgeous, expensive device.

Modeling is not new. Yamaha was doing it in 1993 with the VL1, mathematically recreating the airflow, reeds and resonances of saxophones and other acoustic instruments, while Line 6, among others, later applied the same broad idea to guitar amplifiers. What has changed is the sophistication of the models and, fundamentally, how cheaply we can now run them.

NAM works by training a neural network with recordings from real audio equipment, learning how that equipment responds to an input signal rather than simply approximating its sound with conventional effects. Its new Architecture 2 can substantially reduce the processing power required, between 30% and 40% less CPU than its predecessors.

A small Pi can now run technology that not long ago belonged firmly in the expensive category.

But let’s not get carried away. I haven’t built an HX Stomp for twenty bucks.

A commercial multi-effects pedal is a beautifully integrated device. It comes in a sturdy metal case, with proper footswitches, instrument-friendly connectors, and a manufacturer responsible for making everything work well. My version consists of a Raspberry Pi and a handful of spare bits.

I also found that the choice of audio interface is important. A perfectly small interface from a renowned guitar maker proved so problematic that I abandoned it in favor of a larger Arturia unit that looked much more like something that would survive being thrown in a case.

And as much as I try to find reviews (undo would be nice), it’s all there, it all makes sense and does what it says on the tin.

So, would you trade this collection of snare drums for a commercial pedal if you went on stage in front of 20,000 people?

Those should be two questions.

Does it sound like a business? Yes. Yes, it does.

Would I trust you with my reputation in a live situation? Not yet, and only because I haven’t been through anything like that yet.

But would I take it to the studio? You bet I would. Would the techs want to know what the hell it was and I would act cool while showing off my custom, self-made device that made me sound like Pino Palladino or Guy Pratt in the blink of an eye? Yes, humbly, I believe so.

Would you show up with a second PiPedal ready, just in case? That would be relevant.

As impressed as I was with my creation, what was truly extraordinary was that I was able to build it, that I didn’t have to acquire all the experience beforehand.

Building this is VERY achievable. This is one of those projects that is now almost too easy not to try. These were the steps to follow:

1. Install the Pi operating system. Pi Startup. (Important: enable SSH).
2. Tell ChatGPT “Use Desktop Commander and turn the RasPi on my network into a PiPedal.”
3. Lean. Put your feet up. Instruct the audience to prepare for the reenactment of that scene from Back to the Future.

From time to time it may ask you for a little help, but nothing you can’t handle (like pressing ENTER). When he’s done, give him your playlist and say, “Make me a patch for each track.”

And it will be.

Some will be nailed. Some will be a good starting point and others you will simply wonder why. But you don’t even have to make the changes yourself.

“Put a little more chorus on it, add some mids, and trade that Boss CS-3 for a Cali76.”

And that’s it.

Projects like this used to have an invisible entry price: knowledge. The hardware may only cost a couple hundred dollars, but it also required knowledge of Linux, audio and networking, and the patience to put it all together. For many, that cost of knowledge was greater than the cost of purchasing the finished product. And fair enough.

AI changes that equation.

I provided the ambition and judgment. I knew what I wanted the finished product to do and I knew if the bass coming out of the speakers sounded good. The machine gave me the experience I needed.

This seems considerably more revolutionary to me than another smart Raspberry Pi project.

However, it is evident that it is free and open source software.

It doesn’t look like it was put together by Pininfarina. PiPedal’s web interface has a faint whiff of Geocities, and no one pretends that this is what you hand to a professional and say, “That’ll be cool, do it.”

But of course, what do you want? It is free and open source.

More importantly, it has the parameters, routing, models, and flexibility that matter. It may not look like a thousand dollar pedal, but in terms of what it can actually do, it comes so close that I personally think it makes the thousand dollar alternative start to look a little silly.

I went looking for an expensive commercial effects pedal. Instead, I spent about $100 on a footswitch to complete a rig with hardware I already had, ended up with more sounds than I’d probably use, and found that changing it now requires little more than barking at my AI.

The most interesting component of my new pedal is not the Raspberry Pi. It’s the phrase I use to tell him what he will become next.

I wonder if Chat can operate Audioprism…