Generative Audio, Web Audio API, 2026

This project is an uncompromising rejection of modern, mediated audio tools. There are no pre-recorded samples, no external libraries, and no hidden abstraction layers. Every sound wave is synthesized from absolute nothingness - raw mathematics executing sample-by-sample within a custom isolated audio thread.

The engine reconstructs musical structure not as a sequence of notes, but as a rigid acoustic physics simulation. Amplitude envelopes abandon standard linear programming; instead, they calculate the equations of motion for an overdamped harmonic oscillator. High-frequency partials dissipate exponentially faster than lower ones, mimicking the thermodynamic death of physical sound.

To spatialize the audio, the system bypasses standard convolution reverb. A continuous numerical integrator solves Lorenz differential equations in real-time, mapping chaotic attractors to the interaural time difference (ITD). The signal organically drifts through the stereo field like a derelict transmission in the vacuum of space, while a 4x4 Householder Feedback Delay Network perfectly diffuses the echoes across a geometric hyperplane.

The visual interface abandons traditional UI for a dual-channel Lissajous phase-scope, plotting the Left and Right audio buffers against each other on a Cartesian plane. As the math unfolds, viewers watch the phase differences physically bloom into complex, glowing knots. It is an architecture of numbers - a machine learning to sing by calculating its own physical decay.