Research
I started by exploring general music theory to understand what white noise is made of and how to construct it, so I could guide the models more intentionally. I recorded a short snippet of DSB trains on the tracks to use as a sonic reference. With no prior music experience, I also had to learn my way around audio processing software to fine-tune the output and add effects. I tested several models, Suno and Udio both struggled with making longer tracks, losing context after around two minutes and shifting the feel of what should be repetitive minimalistic composition.
Stable Audio didn't have this problem, making it the right fit for the project, though in general, at this point Stable Audio was weaker in terms of broad use for making music, It really nailed the minimalistic sound we needed for this project. After doing a lot of research on prompt adherence, prompt structure and getting more comfortable with Stable Audio, the tracks started to form one by one. The final songs were sent for professional mastering, and my Danish colleagues created the surrounding marketing materials.
