About Fenophone
A generative, deterministic music instrument you play in your browser.
Pick a sound — eleven packs ship free, from ambient, beat-driven, and cinematic to techno, neo-soul, lo-fi, synthwave, and industrial — and shape it live. The big dials, intensity and spread, lead the playing, with the tempo and key you set; underneath sit deeper controls — change rate, layers, flow, level — and a routing matrix you can edit. Capture the parts you like and export them as MIDI or audio, in your own key and tempo. It works in your browser, with no account.
Fenophone doesn't use AI. It's a deterministic instrument — the same setup always plays the same way, so what you make is reproducible and exportable. The software doesn't make your music. You do, with your hands on the controls.
Nothing you make leaves your device unless you turn on cloud sync or share a link.
Alive, not looped
Most generated music sounds generated. The notes land in the right places — chords resolve, rhythm holds — and something is still missing. The usual fix is a 'humanize' control: a little random jitter, a looping groove template, or a fixed swing offset. None of those is how people actually play. Human timing drifts and clusters with long memory — its loudness and timing fluctuate together across every time scale at once — and when that structure is absent you hear the grid.
Fenophone is built from that structure. The swell of intensity and the microtiming of every note carry the multi-scale statistical signature measured in human performance — not a humanize knob sprinkled on top, but the core of the instrument. None of it is taken on faith: the timing measures near 1/f on every build — α ≈ 0.90, collapsing to ≈ 0.49 once its order is shuffled away, the same long-range correlation researchers measure in human performance — and that check runs in continuous integration before anything ships. Its packs supply the musical floor; the engine supplies the life. And the feel is yours alone: a per-instrument character gives each player a subtly different feel while every note stays bit-identical, so the same seed always replays the exact same performance.
The mathematics is public: a Markov-switching multifractal — born in econometrics — played as an instrument. The measurements
Born in econometrics
Fenophone began as a question in econometrics. Markets have a strange statistical texture: volatility arrives in clusters, calm stretches into calm, and the same roughness repeats from minutes to months. The cleanest mathematics of that texture is the Markov-switching multifractal (MSM) of Calvet and Fisher, in the line of Mandelbrot's multifractal cascades.
Played music has the same texture. The loudness and timing of a human performance fluctuate with long memory and clustered bursts — a large part of why a performance sounds alive while a quantized grid sounds robotic. Fenophone plays that mathematics directly: an MSM drives the instrument's intensity, and everything you hear — harmonic tension, rhythmic density, dynamics, the microtiming of each note — moves with it. The sound packs supply taste and musical guarantees; the mathematics supplies the life.
None of this is taken on faith. The instrument is deterministic and bit-reproducible, so its statistical claims are measured and published — long-range correlation and human-like microtiming enforced in continuous integration, multifractality verified against surrogate nulls in the published datasets — measured, never assumed. Fenophone is built by Evan Tabak Atlas, who came to instrument-making from econometrics research; the measurements, papers, and datasets are public.
The instrument's design is documented in a public research program — measured claims, papers, and datasets anyone can reproduce. Read the measurements and papers.