UMWELT - Generative Ecosystem Synthesizer (VST / AU)

video upload by Laboratory 0

"UMWELT is a synthesizer made for listening. Press the capsule to generate an autonomous ecosystem with its own weather, tuning, four creature groups, and occasionally a language. Drag across the landscape to rotate the view. Active sounds light up and create ripples, revealing which group is active. The capsules along the top recall recent worlds.

Take your time with each world. Populations grow and decline, day turns to night, storms pass through, and the creatures copy each other's songs with small mistakes, so the same world sounds different ten minutes later.

Free Download: https://laboratory0.gumroad.com/"

How it works

A 64-bit seed sets every parameter of a world. The sound is made in four layers.

Substrate is one of four sustained generators:

• up to 40 detuned additive oscillators with Kuramoto phase coupling, drifting between beating clusters and phase-locked tones

• a 128-cell Gray-Scott reaction-diffusion ring, resynthesised by FFT into a band-limited 2048-sample wavetable

• noise through high-Q state-variable bandpass filters

• a modal bank excited by filtered noise

Colony is one of three:

• resonator grains triggered by Kuramoto- or pulse-coupled phase oscillators

• bubbles at Minnaert frequencies with rising pitch

• crackle timed by a self-exciting Hawkes process

Callers use band-limited harmonic sources built on Moorer's discrete summation formula, with period doubling, a second independent source and chaotic jitter. The source then passes through one of:

• tracking bandpass filters

• a four-formant filter bank with noise and click consonants

• a low-pass body with chorus

Artifacts are drawn from:

• modal banks of up to 24 modes with tension modulation, so their frequencies follow stored energy

• dispersive allpass cascades of up to 160 stages with feedback

• FM with the modulation index limited by Carson's rule

• a granular resampler fed by the mix

An eight-neuron Izhikevich spiking network with delayed synapses times the events. Four-species Lotka-Volterra equations set each layer's activity. A chaotic attractor (Lorenz, Rössler, Thomas, Aizawa or Halvorsen) drives slow parameter drift.

Pitch comes from 12-TET modes, just intonation, EDO subsets, Bohlen-Pierce and other non-octave scales, or scales taken from the minima of the timbre's own dissonance curve.

Each point source is placed in 3D with propagation delay (Doppler), interaural time and level differences, air absorption and a ground reflection. It then feeds a 16-line feedback delay network whose Hadamard mixing matrix slowly rotates. The master applies loudness normalisation, gentle compression and a 1.5 ms lookahead limiter.