FAudioAttenuationSettings — one shared attenuation description (UE's Attenuation Settings asset), referenced by many sources instead of every source carrying its own curves. Distances are measured from the shape's surfac
Sampler for the author-editable attenuation curves carried on AudioSourceComponent. A curve is a small sorted set of {distance, value} keys; the engine evaluates gain-over-distance and cutoff-over-distance from these eac
One snapshot of everything the mixer is doing (UE's au.Debug / Audio Insights): voice counts by state, per-submix meters, concurrency groups, the memory budget, the parameters and the active spatializer.
FAudioDebug — the debug surface. Capture() takes the snapshot; Format() turns it into the text the au.Debug console command prints and the editor's Audio panel shows; DrawAttenuation() draws every playing spatial voice
Base for every DSP node we put on a bus. Owns one ma_node*-shaped object — derived classes pick which ma_*_node they wrap and how to interpret their config. Lifetime: created via the AudioEngine::CreateXxxEffect factor
One shape for every filter's parameters — the three floats and the order cover cutoff/gain/Q/slope/frequency across all six, so publishing and draining stay in the base class and each subclass only says what its fields m
Delay effect. miniaudio's ma_delay_node implements a feedback delay line (echo / slap-back) — delay length is fixed at init time, decay + dry/wet can be changed live.
Convolution reverb — the impulse response of a real (or designed) space convolved onto the signal via FConvolutionEngine (uniform partitioned FFT convolution, one engine per channel). Fixed latency of one engine block (2
Compressor — feedforward peak compressor with soft knee. Per-channel envelope detection is linked (envelope = max across channels) so stereo signals don't drift across the centre image.
Envelope follower — a passthrough tap that publishes the level of what runs through it: the RMS of the last block and a peak envelope with an attack/release, both read from any thread. What lip-sync and a meter read; the
FAudioFileDecoder — a whole clip to interleaved float PCM at a rate and channel count the caller picks, outside the live mixer. For offline work (the Sequencer's movie render mixes its audio tracks with it): a source fil
FAudioImpulseMemory — the ambience's short memory of loud things: where the recent impulses (a scan, a door slam, a gunshot) happened and how much of each is still "in the air". Two things read it:
One binding: a named parameter read through a curve onto a target. The curve keys are {parameter value, target value} — FAudioCurveKey's distance field carries the parameter value. EAudioCurveInterp::Step turns the cur
FAudioParameters — the RTPC layer: named, game-set values ("Attention", "Heat", "ScanRange") that sounds read instead of being driven one by one. A value set here glides to its target over the parameter's smoothing time,
What a spatializer decides for one voice this tick. Gain multiplies the voice's own volume; Pan is -1 (left) .. +1 (right); the low-pass joins the voice's occlusion and modulation cutoffs, the darkest winning.
The seam a spatialization plugin lives behind (UE's IAudioSpatialization): given where a voice and the listener are, say how loud, where, and how dark it is. A plugin (Steam Audio, an HRTF library) registers a factory un
FAudioSpatialization — the registry and the switch. "Native" (always present) is miniaudio's own spatializer; "Panner", registered by the engine, is an equal-power panner with inverse-distance gain — the reference implem
One node of the submix tree: a mixer group with its own fader, pitch and ordered effect chain, feeding its parent. Sources attach to a submix by name, so the five default buses (Master, Music, SFX, Voice, Ambient) are ju
One authored point of an attenuation curve: the value the curve holds at a given distance from the listener. Keys need not be authored in order — the evaluator sorts by distance before sampling. Reused for both the gain
One authored default for a MetaSound control input: the value the graph's named input holds when the voice starts. Applied once at voice creation, so a designer can tune a synth's cutoff or an ambience bed's density per
A control input that follows an audio parameter (FAudioParameters) every tick: input = offset + scale * parameter. How a game value reaches a graph without code — an infrasound rumble whose amplitude is "Threat", an ambi
MetaSoundComponent — a placeable MetaSound voice on an entity. Mirrors AudioSourceComponent's lifecycle contract: the AudioSystem resolves the .metasound graph on the first tick that sees the path, spins up a MetaSound
An opening between two rooms that sound travels through: a doorway, a hatch, a vent, a door with how far it stands open. The opening lies in the entity's local XY plane (Width along X, Height along Y) and the rooms it jo
A room the sound graph knows about: an oriented box around the entity (rotation and scale follow its transform). Two points in different rooms hear each other only through the portals that join them, or through the walls
Uniform partitioned convolution (overlap-save), the engine behind convolution reverb. The impulse response is split into blockSize partitions, each FFT'd once at Init; per input block one FFT + P complex multiply-accumul
FEchoTail — the echo a space gives back, built from where the scanner's rays actually landed instead of a reverb preset per room. Every shot contributes its hit distances to a decaying histogram (0.5 m bins out to 64 m)
What sound bounces off and is stopped by. The scan hierarchy is the one the LiDAR game wants — the same triangles the scanner paints, castable from any thread, no Jolt query on the game thread — and the physics scene is
How a surface treats the energy that lands on it: the share it absorbs (Sabine coefficient, mid band) and the share it scatters diffusely instead of mirroring.
Object-oriented audio handle exposed to Visual Scripting. Every MFUNCTION below becomes a node in the Visual Script catalog (built from MClass function tables), so a graph can load a clip and drive its playback the same
Reflected façade over the AudioEngine singleton, so the bus mixer and fire-and-forget playback show up as Visual Script nodes without annotating the whole miniaudio-laden engine. Every method forwards to AudioEngine::Get
A DSP synthesis graph: oscillators, envelopes, filters and math nodes wired by typed (audio-rate buffer / control-rate scalar) connections. Prepare() topo- sorts the graph and allocates per-node block buffers; Process(bl
A live instance of a MetaSound DSP graph, wired into the AudioEngine as a custom data source so its synthesized blocks flow through the normal bus / spatialization path. Own a copy of the compiled graph; drive it live wi
A node of the sound-class tree (UE's Sound Class): volume and pitch multiply down the chain, the low-pass takes the darkest. Independent of routing — a footstep class can live on the SFX submix and still be turned down w
FSoundClasses — the sound-class tree, the sound mixes that adjust it, and the concurrency groups. Classes and groups come from the project config (audio.sound_classes = "Name:Parent:Volume:Pitch:LowPass", `audio.concur
One layered voice inside a multi-voice cue (a Mixer branch): its own weighted clip pool plus the gain/pitch/attenuation resolved for that branch. The voices of a cue start together on Play and stop together on StopAll —
FSoundOcclusion — the one occlusion law every sound in the engine obeys: world sources, proximity voice, radio speakers and AI hearing all ask it. Before it there were three, with their own constants, and a comment in th
FSoundPropagation — rooms joined by portals, and the one answer to "how does a sound get from here to there". World sources, proximity voice, radio speakers and AI hearing all ask Evaluate; that the creature and the play
The reader moved down into AssetFormats so the packaging cook can use it without linking the audio engine; this name stays because it is what the runtime, the importer and the tests already spell.