“At ten, on the hill behind my school, I played the taepyeongso — a Korean reed wind instrument — for the first time, and felt the whole mountain and every living thing in it tremble with me. That was when I understood: nature and the universe are all connected, through the vibration of sound.” — seungsoon park (Radiophonics)
16 years of AI soundscape research,
woven into a single cosmic audiovisual universe.
COSMOSCAPE.AI is the total project of electronic musician and media artist Radiophonics (seungsoon park). Since the 2010 album 〈Cosmos〉, he has built the Cosmos Trilogy as one continuous body of work — story, music, sound and moving image — and from 2016 wove in AI soundscape, the first practice of its kind in Korea, turning it into a single holistic AI audiovisual project. Each viewer generates their own universe through their own contemplation — an audiovisual landscape where no two people ever see the same cosmos.
In 2026, Neutune — the AI music-tech company he co-founded — takes part in global AI-music standards including DDEX and WIPO, contributing new creative methods, standard models and datasets to the emerging field of AI sound art.
Cosmos Trilogy IP
An interactive music IP (music · visuals · text) — a cosmic trilogy optimized for AI modeling and the flagship work that probes the engine's limits.
cosm.os engine
A 3D AI audio-visual soundscape engine originating from NEUROSCAPE, running on spatial computing — the reusable core technology.
Performance / Exhibition
Cosmic AI performance and exhibition combining engine and IP — proof it works outside the lab.
Open Model & Web
A personalizable model release and web platform — contributing to and expanding the AI sound-art ecosystem.
From IP to engine to stage to an open ecosystem — each stage feeds the next, and the open model loops back as the seed of new work. The cosmos expands.
COSMOS TRILOGY
A three-part immersive audiovisual performance exploring the birth, expansion, collapse and rebirth of the universe through sound, image and AI narrative. All 15 tracks in Dolby Atmos 7.1.4ch · with a musicDNA (BlockMusic) structure, cosm.os performs endless AI remix live.
PART ICOSMOS
PART IIENDLESS UNIVERSE
PART IIIWHITE HOLE
cosm.os — LLM × LWM Agentic Creative OS
An agentic AI operating system that, from input (text · image · seed audio) to output (audio · visual · lighting), understands the artist's narrative intent and reads the state of the performance space to make real-time directorial decisions. It unifies an LLM for narrative and emotion with an LWM for space and world-state.
Neuroscape V2 Agent
Converts text, image and audio into four paired-axis coordinates (point–line–plane · particle–wave · annihilation–genesis · dimension) plus auxiliary context (density · tempo · tone · energy).
Parametric Sound Design
Maps axis coordinates to granular, wavetable, FM and spectral synthesis parameters — outputting creator-tweakable synthesis parameters (the core in-house model).
Spatial Sound Arrangement
Auto-places the position, trajectory and reverb of sound objects in an HOA / Dolby Atmos 7.1.4 environment (8ch and up).
cosm.os Orchestration
An agentic OS that unifies audio, visual and lighting control over OSC/MIDI/DMX, including a Visual Mapping Layer.
The Foundation Engine, live now
Live demos of the NEUROSCAPE foundation engine that cosm.os unifies — type a word or drop an image and a soundscape, spatial audio and a 3D world are generated in real time, stage by stage: 8-channel → quad → Atmos → 3D AI audio → world model. (unified host · single port localhost:8788)
8-Channel Soundscape
The earliest neuroscape model — image/word → label → embedding → an AudioSet 8-channel synthetic soundscape. A faithful rebuild of the original 2018 paper's system.
Quad Mixedscape Editor
FOREST CONTRAST — an editor for mixing sound-source contrast, width and reverb directly into quad (4ch) sound art.
Atmos Spatial Objects
Promotes the 8 channels to spatial objects. Speaker-bed + 3D-room dual view, HRTF binaural, per-channel source & mute/solo.
3D AI Audio
AI-generated soundscape sources spatialized in real 3D — binaural HRTF, sources orbiting the listener with live azimuth / elevation. Put on headphones.
World Model
Word → a navigable World Labs Marble 3D world. Single-channel · 3-face cube · festival projection + sound particles.
- Neuroscape.AI Framework자체 엔진 · in-house
- Stable Audio SFX API
- Marble AI APIWorld Model
- NIA 가상공간 환경음 매칭 데이터셋 (2022) — 데이터셋 구축 총괄 담당 (directly led the dataset construction)
- FSD — Freesound Dataset
- SA3-sfx — Stable Audio SFX Dataset
Partner with
Validating cosm.os and the Cosmos Trilogy on the international stage, atop a global collaboration network across technology, industry, science and art.
Contact
Tell us about your stage, exhibition or collaboration — we'll get back to you.