Participants 2025
About the TEAM
marts~ is an alias of Martin Heinze, sound artist, composer and electronic musician working in the field of experimental electronic dance music. His current practical research revolves around algorithmic composition and generative frameworks for electronic music, in particular Jungle and Drum & Bass. His project "Fibonacci Jungle" received recognition at the Generative Music Prize 2024, hosted by IRCAM. In recent years, he has focused on developing compositional techniques using real-time neural audio models trained on his own musical material. Projects such as "Saatgut Proxy" investigate semi-generative improvisation practices in a setting of shared human/ AI agency.
About the SONG
"Reykjavík Sunburn" is an output from a larger research project dedicated to exploring musical qualities in working with generative neural nets for audio, conceived both as hybrid instruments and as autonomous actors.
Four different neural audio models, trained on my own musical material (and a private voice dataset) are employed in an improvisational real-time setting to create a new piece of electronic music.
About the HUMAN-AI PROCESS
"Reykjavík Sunburn" was created through a live exchange between human agent and four custom neural audio models, trained on my own back-catalogue (plus a private voice dataset). Two models were built with RAVE: "Black Latents" delivers gritty, percussion-focused material while "Nobsparse" supplies clean bass lines, stab sounds and airy chords. The other two models were created using vschaos2: "VSC2_Nobsparse" generates evolving pads and drones while "VSC2_Martha2023" reshapes recorded vocal snippets into rhythmic textures.
All four models run inside Pure Data, a visual audio programming environment. Rather than using them to create samples for later processing in conventional DAW routines, I act in real-time inside the models’ latent space - steering mood, density and tempo by exploring parameter and signal constellations. This practice, dubbed "Latent Jamming", replaces deterministically driven composition with guided exploration: tweak, listen, stabilize, vary.
After several sessions, the submitted take was captured on eight separate tracks, mixed lightly in Audacity and rendered to a final stereo master. The finished piece balances human intention with the models’ spontaneous variations.