Hearing & sounds#

An APS guard hears exactly what your game says is audible: every sound is something you emit, with a type that says how loud it is and how far it carries.

1. Make a sound type#

Content Browser → right-click → Miscellaneous → Data Asset → SoundTypeDefinition. One per kind of sound.

2. Emit it#

Both nodes work from any Blueprint, with no component to find (category Perception › Sound). They only feed perception: play your actual audio as you always do.

3. How a sound reaches an AI#

For every AI in range, cheapest checks first, so a footstep across the map costs almost nothing:

1. RangeBeyond the sound's Max Range: skipped.
2. Level of detailThe AI's tier above Max LODTier: skipped.
3. CategoryNot accepted by the AI's filter: skipped.
4. Alert levelBelow the filter's minimum: skipped.
5. TeamSame team, and the filter ignores its team: skipped.
6. WallsOnly now the expensive trace.

Then it is weighed: distance 1 − (distance / Max Range)², and every wall in between multiplies it by exp(−Wall Absorption Coeff), so two walls muffle more than one. The walls are counted along a line from the AI's ear (64 cm above its location) to the sound; each solid actor counts once. Pawns aren't walls, nor is anything that only overlaps (a trigger volume), nor the thing making the sound.

heard = Base Loudness × filter's Loudness Multiplier × (1 − (distance / Max Range)²)
      × exp(−Wall Absorption Coeff × walls) × weather

Max Range here is the sound's, times the Hearing Range Scale of the environment the listener stands in (a crowded market below 1: see World).

Loudness above 1 mostly makes a sound carry further, not louder up close: confidence tops out at 1, so a gunshot at 3 stays at full confidence out to a much greater distance.

4. Noises add up#

One distant footstep is a blip; a series in the same place builds toward Suspected and then Detected. Each sound adds as much as it was heard: a step behind two walls adds less than one in the open. The AI uses the higher of the single sound and the build-up, so building up can only help.

How fast: each sound adds what it was heard at × Sound Accumulation Rate × 0.1. At the default 0.5, a footstep every half second takes about 10 seconds to add up past a single step, then keeps climbing for as long as it goes on. For an AI that adds noises up quickly, raise the rate: at 2 it is four times as fast.

The build-up is the same at every level of detail: a far, cheaper AI isn't quicker or slower to add up the same footsteps.

The design lever: a player who moves in short bursts, pausing longer than the hold time, keeps the build-up draining and stays under Suspected. One who runs without stopping builds past it. That is the core stealth loop: tune it with these numbers and the profile's thresholds.

5. Sounds nobody made#

Emit Sound at Location has no maker, so it says nothing about any particular person. The AI forms a belief about the place instead, with its own states, tagged with the sound's Sound Name:

• Its confidence is how loud the sound arrived: weighed by distance and each wall, like any other sound. A can at the guard's feet is surer than one at the edge of its range or behind a wall.
• Its precision is the sound's Location Accuracy Override, or 0.4.
• Repeats in the same spot merge into one belief.
• Once the noise stops it fades like any belief (Lost, Remembered, Expired), so an old noise doesn't hold the AI forever.

Read it with On Location Belief Changed (location, tag, state, confidence) or Get Strongest Location Belief. On AI Hear does not fire for these: there is no one to fire it about.

Projectile hits something
  ├─► Spawn Sound at Location      (your audio)
  └─► Emit Sound at Location       (Hit Location, DA_Sound_Thrown_Rock)

On Location Belief Changed (Location, Tag, State, Confidence)
  └─► Branch: State is Suspected
        True → set "InvestigateLocation" = Location

A classic stealth distraction, with nothing more to build.

6. Filters: hear only what matters#

Give an AI a Sound Filter Profile (Data Asset → SoundFilterProfile), set in its profile's Detection › Sound Filter:

7. Reading what it heard#

On AI Hear (Location, Loudness, Sound Type Name)
  └─► Switch on Name (Sound Type Name)
        ├─ Gunshot  → Broadcast Alert to Squad, sprint to Location
        ├─ Footstep → walk to Location, look around
        ├─ None     → (build-up only, no new sound) keep doing what you were
        └─ Default  → turn to face Location

On AI Hear fires on every hearing update that hears someone, not once per sound: when a sound arrives, and again on each update after it while the build-up lasts (see Noises add up). It has no Target pin.

• Location is the position of whoever made the sound, as it is at that update: on a build-up update, where they are now. How precisely the AI places it shows in its uncertainty radius (Get Uncertainty Radius), from Location Accuracy Override or the distance.
• Sound Type Name is the name of the first sound the AI heard that update, and None on an update with no new sound (build-up only). With several sounds arriving at once, branch on what matters most rather than trusting one name.
• For one event per sound, with who made it, use On Sound Heard instead (below).
• On Sound Heard (on APS Core) gives every sound the AI hears, with how loud it arrived, how many walls it came through and whether the path was clear. Get Recent Sounds returns those heard in the last however many seconds you ask for, up to the last 6 seconds (32 sounds at most): it needs Record Heard Sounds, which is on by itself while a Visualizer is shown or On Sound Heard is bound.

When something's wrong#