DSP-DEMO — Procedural Room Acoustics & Source Occlusion Engine active
yusufuca vunknown added 2026-08-01 verified 2026-08-01
You want reverb and occlusion parameters derived automatically from runtime room geometry and materials in Unity/FMOD, instead of hand-placing reverb zones or authoring per-room presets.
about
A Unity + FMOD Studio runtime system that calculates reverb and source occlusion directly from a room's detected geometry and surface materials, frame by frame. There are no manually authored reverb presets — a flood-fill algorithm scans outward from the player to determine room bounds, volume, and portal connections, caches the result per grid cell, and derives FMOD snapshot parameters (reverb time, early/late delay, diffusion, density, HF decay, high-cut, low gain) from volume, material hardness, ceiling height, and wall proximity.
Per-source occlusion uses raycasting toward the listener to shift frequency, volume, and pan toward occluded target values, with early/late reflection timing derived from speed of sound. A validation suite tests the model against controlled scenes that isolate single variables (size, material, ceiling height) and a seeded procedural house generator confirms generalization beyond hand-built rooms.
Debug tooling includes a live HUD showing reverb parameters and room stats, plus in-editor RoomVisualizer gizmos for detected cells and portals. Portal-based aux routing between adjacent rooms and outdoor/open-space handling are currently disabled and being reworked; core room detection, reverb model, and occlusion are stable. Built on Unity 2022.3, C#, and FMOD Studio.
related
16bits-audio-mcp
Standalone · Engine-agnostic16bits-audio-mcp is a Zig-based MCP server that exposes game audio generation as tools callable by Claude. It produces 16-bit PCM WAV files using FM synthesis, reverb, ADSR envelopes, and multi-track mixing, all from the Zig standard library with zero external dependencies. It covers three asset types: BGMs in 15 styles (adventure, dungeon, boss, cyber, horror, etc.) with 4-track arrangements; 20 sound effect types (jump, coin, explosion, laser, etc.) with pitch and volume control; and 12 jingle types for common game events like stage clear and game over. Scales, tempo, key, and a deterministic seed parameter give reproducible iteration. A post-processing tool (wav_fx) chains effects such as reverb, delay, bitcrusher, chorus, and distortion, and a mixer tool combines multiple WAV files. Additional tools handle direct note/FM synthesis, WAV info, and playback. The workflow is prompt-driven: describe what you want to Claude, and WAV files are written to disk.
ArcadeAudioKit
Standalone · Engine-agnosticArcadeAudioKit is a Swift package for modeling and rendering short synthesized sound effects on Apple platforms (iOS 18+, macOS 12+, tvOS 18+, watchOS 11+, visionOS 2+). It provides a recipe-based API: you describe a sequence of waveform segments (sine, triangle, or square), each with pitch (scientific notation notes, Hz values, sweeps, or interpolated pitch), timing, amplitude, attack, and decay. The renderer converts that recipe into deterministic mono Float PCM samples at a caller-specified sample rate. The library deliberately stops at PCM output. Audio session management, AVFoundation integration, mixing policy, playback, and accessibility behavior are left entirely to the consuming app. This makes it suitable as a dependency in games or game engine extensions where you want reproducible procedural SFX without taking on a full synthesis framework. v1 scope is narrow: recipe modeling, note-to-Hz conversion via A4=440, repeated motif generation for tails and loops, and the PCM renderer. Requires Swift 6.2+ and has no external dependencies.
Audio Agent Creator
Standalone · Engine-agnosticAudio Agent Creator is a Go-based offline audio synthesis engine designed for programmatic use by game agents or automation scripts. It takes a JSON project description (or a named preset) and deterministically renders stereo PCM, then encodes to 192 kbps MP3 via FFmpeg/libmp3lame or exports lossless PCM16 WAV. The synthesis chain supports oscillators (sine, square, saw, triangle, noise), frequency sweeps, ADSR envelopes, per-track volume/pan/transpose/mute, first-order low-pass/high-pass filters, and basic feedback delay. The tool exposes three interfaces sharing the same project DSL and render contract: a CLI with commands for generation, validation, preset inspection, and schema export; an HTTP API with synchronous and async job endpoints; and a Vite/React web workbench for JSON import, server-side validation, job submission, and MP3/WAV playback. Eight built-in presets (6 SFX, 2 BGM) provide deterministic starting points. This is explicitly an agent-first SFX/BGM MVP, not a DAW replacement. It has no MIDI, piano roll, sampler, plugin host, mixer bus, or generative AI music model. The prompt field currently uses keyword heuristics to select presets rather than text-to-audio synthesis.
Audio Asset Lab
Engine-agnostic · StandaloneAudio Asset Lab is a local, YAML manifest-driven pipeline for producing sound effects and music for games, video, and web projects. You describe generation parameters, seeded variants, post-processing steps, and export formats in a single versionable manifest; the tool produces candidate audio files, selected lossless masters, delivery exports (WAV, OGG, MP3), and provenance records containing prompts, model revisions, processing chains, licenses, and hashes. Two generation paths are available: a deterministic procedural engine (no model weights, included by default) for tonal UI cues, jingles, and synthetic effects; and optional Stable Audio 3 MLX engines (Apple Silicon only) for prompt-driven SFX, ambience, and music. Three model variants are supported: stable-audio-3-small-sfx, stable-audio-3-small-music, and stable-audio-3-medium. Consumer repositories receive only the delivery audio files and have no dependency on the pipeline, Python environment, or any model runtime. Builds are seed-addressed and refuse to overwrite existing outputs without an explicit --force flag. The CLI exposes engine routing and the manifest schema via JSON flags to support agent-assisted manifest authoring. The project is described as usable but early; the manifest and CLI are the stable interface.