Earlier this year, Ben Cantil aka Encanti trained a Dance Diffusion model on my discography, so that a theoretically infinite number of brief segments of audio resembling my music could be generated from white noise. The result felt like flipping through radio stations from parallel universes, hearing fragments of songs I wrote in other timelines. Finding glimmers of meaning in these fragments feels like a musical inkblot test from a science fiction story.
What a time to be alive!
"Bobular Synthesis" was started by looping a strange, short, fuzzy segment of the very first test output I heard from that Dance Diffusion model. In the end, I added some more synths, vocals, and a few drum samples, but the core of the track is still sculpted from just a few seconds of generative audio being manipulated and arranged in various ways.
"Concatenation Situation" explores concatenative synthesis / audio mosaicing using PointZero (encanti.gumroad.com/l/kqtXF
). I like to describe it as machine learning assisted granular synthesis, which uses tiny pieces of some source audio to create something resembling some target audio. Many of the textures and sequences for the track were created by using a simple version of the main beat and melody as targets, and various stems & custom samples of mine as source material.
"Frozen Mango Chunks" was made in a similar way to Bobular Synthesis - it was built around edits of a short loop from the 2nd test output I heard from that original Erothyme Dance Diffusion model. Once again, after creating the structure of the track with a restricted palette, I added material (synths, vox, etc), riffing off of whatever was catching my ear.
The tools available to us now are wild...
Meanwhile, better tools are rapidly approaching...
something something... murmur murmur...
real-time neural synthesis...
what am I but
a confluence of
from my senses
find the essence