Back in June I sat down to build a track from scratch. No tutorial, no template… starting with nothing and seeing where things led.
The reference point was Underworld’s Rez. Not because it’s one of my favourite tracks, it isn’t, but because it’s close enough to the sort of music I want to make, while pulling me towards something more industrial and relentless. It also does a huge amount with not very much, which was part of what I wanted to understand… given my small setup.
So I built it bit by bit. Drums first. Completely on the grid, clean machine-like sounds, rigidly timed, then adding the grit myself rather than starting with some sample pack where half the character is already there. Then the arp. For that I started from an ‘init’ synth patch, which is basically a sine wave and nothing else… as close to a blank slate as you can get. I wanted to build the sound up and understand what was actually happening rather than just flicking through presets until something sounded about right. Sawtooth oscillator, resonant filter, envelope shaping the pluck… building it up one bit at a time.
There were a few wrong turns. At one point it sounded like a police siren. At another, weirdly, like a very short cat meow. Then it started sounding like an old video game, which made me wonder whether I’d got the whole thing wrong. The thing that makes Rez work isn’t really the notes though. It’s the movement of the filter… that sweep opening and closing across the sound. Once I got that into my head and stopped messing around with the notes, the arp started to work.
The hardest bit was getting the filter movement recorded properly. Getting the synth and DAW talking to each other, mapping the right controls over MIDI, routing that MIDI to the filter, recording the sweep and then going back in and editing it section by section. It took a bit of figuring out, but it taught me something useful. Push something too far first, just to prove it works. Then pull it back. I ran the filter sweep way beyond anything I’d actually use in the track, because I wanted to know the routing was working. Once I knew it was, I could bring it back to where I needed it. And usually the music needed far less than I initially thought it did when first playing and experimenting. That’s probably part of the craft… knowing what you can do, then choosing not to do all of it… or so I’m told.
I don’t think any of this would have gone the same way a year ago. I’ve known for years, roughly, what I want music to sound like. That’s never really been the problem. The problem has been knowing how to get from the sound in my head to something coming out of the speakers. I know what I like. I just haven’t always known how to make it, and that lack of knowledge kept me not trying. But now I’m starting to use it, experiment and play around.
I’ve been running the music project alongside Arbor, my own AI system, basically using it as a tutor sitting next to me. When I get stuck on something… routing an effect so it hits the arp but leaves the hats alone, working out why a cloned track has inherited settings I don’t want, whatever it happens to be… I can ask. Importantly, the answers are relevant and specific to the thing I’m actually building. Not a generic YouTube tutorial using completely different gear to make a completely different track.
I don’t really think of that as AI making music… it’s AI teaching me and pointing me in vaguely the right direction. The taste is mine, the decisions are mine. My ears and my say on what stays, what goes, what sounds good and what sounds terrible. What AI is doing is helping me close that gap between knowing what I want and knowing how to build it. That’s the bit I’ve found interesting. It’s not replacing the craft. It’s helping me get to the craft.
Once the loop worked, the rest of the track came together fairly quickly. Eight sections, mostly made by taking the full loop and stripping things away rather than building every section separately. The intro slowly comes in from the outside… almost like hearing music through a door before walking into the room. Then everything builds. There’s a breakdown where almost all of it disappears, and an outro that gradually peels everything away again until you’re left with the same bare pulse it started with. And suddenly I had a whole track. Start to finish. Built from nothing.
Which brings me to the less impressive bit. At the start of the year I had this idea of making six tracks about Pembrokeshire. Six places, six tracks… roughly one every couple of months. It’s July and I’ve finished one, with this second one basically there. So, clearly, that schedule went well.
But looking back, I think those first six months were mostly about learning how to do this properly. Learning the gear, signal paths, synthesis… even learning what an init patch was and why I’d want to start with one. You can’t make six tracks quickly until you know how to make one at all. And I think I do now.
The next one is already knocking around in my head. Barafundle. All Korg. Nothing else. Slow, evolving, lots of space… something that can breathe and change without constantly needing another part added to it. After that maybe Valero for something harder and more industrial. Tenby feels warmer somehow… maybe actual crowds and voices recorded down on the coast. There are still two I haven’t worked out. That’s fine.
The gap between what I can hear in my head and what I can actually make is smaller than it was in January. It hasn’t disappeared. I don’t think it ever will. But it’s smaller. And, more importantly, I know what to do with that gap now rather than just waiting for it to disappear.
Two down. Four to go.
