Repository navigation
Releases
Empyrean 1.0.0
The first public release: an additive synthesizer that does on a graphics card what no processor could -- and then
keeps going.
- 320,000 partials in real time. Sixteen voices of twenty thousand sine waves each, every partial with its own
frequency and glide, three envelopes (loudness, pitch, place), the filter's exact response at its frequency and its
distance -- in one CUDA graph per block: on an RTX 5090, twelve voices sounding, 0.10 ms of GPU time per block of
2.7 ms, 0.17 ms with every effect on. The sum in a fixed order, so a render is bit-identical every time. - A reverb made of the instrument's own stuff: a modal hall of 65,536 decaying resonances on the card, tuned to the
scale, that remembers what it heard over minutes to hours and rings longer there; a spectral echo of 500 bands. - Physics as sound: a lattice-Boltzmann fluid the notes blow ink into, a room simulated as air (the wave equation
on 250,000 cells, with a time-reversal mirror), Lenia and Turing fields, Kuramoto sync, chimeras and swarmalators,
a slime mould between the harmonics, a gong's energy cascade, an Aeolian harp whose vortices lock onto a harmonic,
thunder from a lightning channel laid out anew at every strike, a frozen lake cracking in its critical state, a
night chorus of two thousand insects and frogs. - The sky and the deep: a star's oscillations (asteroseismology: a sun, a red giant, a white dwarf), a black hole
that redshifts the partials falling in and lenses the rest into a ring, black holes merging, hydrogen's spectral
lines with their Zeeman triplets, twinkling through Fresnel turbulence, the ocean's deep sound channel traced ray by
ray, whistlers in the magnetosphere, giant vocal tracts singing kargyraa. - The ear: binaural per partial (head shadow, pinna, the time round the head, each of 20,000 partials at its own
point), a melody heard only between the ears (dichotic pitch), streams that part and fuse by common fate, Risset's
rhythm that slows for ever. - Tuning that listens: a conductor in just intonation that changes one voice at a time; Self-Tune pulls nearly
meeting harmonics together; Sethares' scales from the cloud's own spectrum; the root walking as a quantum wave on the
Tonnetz. - A mind of its own: a reservoir of 2048 neurons on the card as a modulation source, the instrument dreaming of
what it played (a modern Hopfield network over its own spectra), macros taught by example, presets found by a
recording, a preset map that grows itself (MAP-Elites). - 16384 presets in 256 groups, each levelled by its loudness as it is played (ITU-R BS.1770) and placed on a map
by how it sounds; glides in which the partials themselves travel and filters move into each other by transport; 531
journeys, nineteen of them whole nights. - Real time on Windows: a render-ahead ring (two host blocks and ten milliseconds), MMCSS, and Windows' REALTIME
GPU class, so another program's graphics cannot hold the card. - New since 0.26.0: the render app writes its score cues and chapters beside a journey (
--cues,--chapters)
for KaleidoscopeEnhanced and Heliacal; the signal flow (docs/flow.png), the demos (Tools/demo/make_demos.py), the
social preview, and the publish script.
Downloads
- Empyrean-1.0.0-Setup.exe -- installs the standalone, the VST3, the renderer, Empyrean Journeys and the manual; the MSVC and the CUDA runtime are linked in.
- Empyrean-1.0.0-portable.zip -- the same in one archive, nothing to install.
- EmpyreanQuest-1.0.0.apk -- the Meta Quest app: the head and the hands in the binaural cloud, the sound on the computer (
adb install -r EmpyreanQuest-1.0.0.apk, developer mode). - Empyrean-Manual.pdf -- every page of the panel as a picture, every parameter.
- SHA256SUMS.txt -- the checksums of all of them.
Requirements: Windows 10 or 11 (x64), an NVIDIA GeForce RTX card (20 series or newer) with a current driver, a VST3
host for the plugin, Meta Quest 2 or later for the app. Developed and tested on an RTX 5090: on a smaller card lower
Partials (the Cloud page) and leave the largest simulations off. No macOS (no CUDA there) and no Linux build yet.
The installer is not code-signed: Windows' SmartScreen may warn once ("More info", "Run anyway"). Licence: AGPL-3.0.
Empyrean demos
Demos of Empyrean, rendered by its own empy_render from the presets below, 3 minutes each, the conductor playing -- nothing played in, nothing edited.
| demo | what it shows | file | renderer arguments |
|---|---|---|---|
| Lunar Hall | a sleep concert in just intonation | lunar_hall.mp3 |
--seconds 180 --preset "Lunar Hall" --seed 7 |
| Teeming Garden | a Lenia field the partials follow | teeming_garden.mp3 |
--seconds 180 --preset "Teeming Garden" --seed 11 |
| Thunderous Rainpool | rain, thunder and a night chorus | thunderous_rainpool.mp3 |
--seconds 180 --preset "Thunderous Rainpool" --seed 23 |
| Convective Betelgeuse | a red giant's oscillations | convective_betelgeuse.mp3 |
--seconds 180 --preset "Convective Betelgeuse" --seed 42 |
| Sliding Whistler | whistlers in the magnetosphere | sliding_whistler.mp3 |
--seconds 180 --preset "Sliding Whistler" --seed 5 |
| Rumbling-Voice Steppe Night | giant throats singing kargyraa | steppe_night.mp3 |
--seconds 180 --preset "Rumbling-Voice Steppe Night" --seed 3 |
The video (convective_betelgeuse.mp4) is the Convective Betelgeuse demo with pictures by KaleidoscopeEnhanced (configuration SpaceAmbient), drawn offline from the sound. Made by Tools/demo/make_demos.py.