Open-Ended Evolutionary Synthesis: Integrating Quality Diversity and Unsupervised Dimensionality Reduction within a Coupled CPPN-DSP Framework for Autonomous Sonic Discovery

  By , , ,
For decades, musicians have used synthesizers to either imitate real instruments or twist knobs until they found something they liked. But what if sounds could grow, evolve, and branch off like living organisms? This is the visionary world of Björn Þór Jónsson, a researcher and designer who has spent years bridging the gap between cutting-edge Artificial Intelligence and the art of sound design. His work has culminated in Synth.is, an experimental platform where “evolving instruments” aren’t just a metaphor—they are a reality.

Beyond the Knobs: How Sounds Evolve

At the heart of Björn’s work is a technology called Kromosynth. Think of every sound as having its own “sonic genome.” Instead of a human manually setting every parameter, an AI system uses “nature-inspired” evolutionary processes to grow complex sounds from scratch. It starts with something called a CPPN (Compositional Pattern-Producing Network), which acts as the DNA of a sound. This digital DNA determines how a sound’s texture, or “timbre,” develops over time. These networks are then paired with a DSP graph—a digital blueprint for a synthesizer—and both are “co-evolved” together. Tust like in nature, the system starts simple and gradually “complexifies,” discovering intricate sonic patterns that no human designer might have thought to program.

A Sound Innovation Engine

The proposed system doesn’t just create random noise; it uses a Sound Innovation Engine to find sounds that are both high-quality and incredibly diverse. Using an algorithm called MAP-Elites, the system “illuminates” a vast landscape of possibilities. In its earlier versions, the AI used a critic (a deep learning model called YAMNet) to categorize sounds into hundreds of different classes—like “Banjo,” “Speech,” or “Whistling”. More recently, the system has moved toward autonomous discovery, where the AI creates its own maps based on the raw characteristics of the audio itself, freeing the search from human-defined labels.

Exploring the “Sound Garden”

The website, Synth.is, turns this complex science into a playground for creators. Here are some of the ways you can interact with these “living” sounds:
  • The Sound Biomes: Imagine an interactive 3D space—a “Sound Garden”—where thousands of sounds are arranged by their similarity. You can wander through these biomes, listening to how one sound family bleeds into another.
  • Trace the Lineage: Every sound has a history. On Synth.is, you can view phylogenetic (evolutionary) trees to see exactly how a sound was “born” and who its ancestors were. You can even “fork” a sound you like, becoming a part of its evolutionary journey by steering it in a new direction.
  • Daily Discoveries: The system is open-ended, meaning it is constantly exploring. Users can listen to brand-new sounds generated daily that have never existed before.
  • Pro Integration: For the musicians and producers out there, Synth.is isn’t just a lab—it’s a tool. You can drop these evolved sounds directly into professional software like Ableton Live, bringing “alien” textures into your own music sets.

Cultivating “Open-Earedness”

The ultimate goal of this project is to foster “open-earedness”—a psychological openness to unfamiliar and unconventional sounds. By exposing us to sounds that sit outside our traditional musical categories, Synth.is challenges our expectations and provides fresh “stepping stones” for creativity. The instrument is no longer just a means to an end, but a starting point for a journey into the unknown. Whether you are a professional sound designer or just a curious listener, Synth.is invites you to step into the garden and see what has grown today.

Source Code

Demo Videos

Related Publications:

Tagged  , , , , , .
Bookmark the permalink.

Comments are closed.