Four players, one boss, one raid-wide damage tick. The group heal lands, the paladin's aura ticks, the shaman's chain heal jumps, and a regeneration totem pulses. Four heal cues fire inside roughly 120 ms. Nobody in voice chat says thanks for the heal. They say they cannot hear the boss.
Each of those cues was probably fine on its own. The failure is the pile, and the pile is an arrangement problem, not a mixing accident. A heal sound is not really one sound. It is a claim about who got help, and four claims arriving together do not sum into a louder claim. They cancel each other out.
The usual answer is a concurrency cap on the heal bus, set through Wwise's Max Instances or FMOD's max instances with a stealing policy. That keeps the mix clean by deleting information. If the second heal in a 120 ms window goes silent, the player who received it gets no confirmation at all. A cap is a blunt tool. A better arrangement makes it unnecessary most of the time.
Two Jobs, Two Different Questions
A party heal and a single-target heal answer different questions, and the answer for each has to live in a different part of the sound.
The party heal answers: did the group get help?
That is a spatial statement about several targets at once. The listener needs spread, coverage, a sense that the effect landed on more than one body. Loudness is the wrong variable. Six decibels is a modest gain, and stacking four cues can push twelve decibels into a bus that already sits near its ceiling. The result is not four times the information. It is one event that clips and one mix that feels tired.
The single heal answers: did this one get help?
That is a private confirmation. One target, one voice, close in. The player should feel the heal land on them rather than wash over the room. Its whole job is to be believable at close range and then get out of the way.
Swapping the two inverts the meaning. A party heal that sounds like a single heal makes three players think they were skipped. A single heal that sounds like a group heal makes everyone think they were topped off when only one player was. Players act on that belief, and the next damage tick punishes them for it. This is the same class of problem as any cue that means two things at once, and it is worth fixing at the source rather than trying to rescue it during the final mix pass.
Target Count Is a Parameter, Not a Gain Stage
The number of targets healed should drive the cue's internal
structure, not its fader. In Wwise that is an RTPC, for example
one called HealTargetCount running from 1 to 4 and mapped to
layer count, stereo width and reverb send. In FMOD it is a
parameter on the event, automated the same way. In Unreal you
can feed the count into a MetaSound input or
SetFloatParameter on the audio component; in Unity,
a Float parameter on the AudioMixer or a value read by a wrapper
around PlayOneShot.
At one target the party cue should sound like a slightly oversized single heal. At four it should sound wider, denser and more layered at the same peak level. That is where spread rather than loud stops being a slogan and becomes a mapping: count 1 gives two layers, count 2 gives three, count 4 gives four layers plus a wider send. The gain stays where it was.
Per-target cues have a narrower use
If the camera shows the whole party and allies stand in different parts of the screen, a short positional tick at each healed ally can work well. In first person it usually does not. The camera does not move, and a hard-panned heal at the edge of the screen says "over there" about something the player cannot see. Off-screen ally indication is a separate problem with its own conventions.
Stagger the Onsets, Then Keep Them Close
A useful working range for a stacked party gesture is 30 to 60 ms between layer onsets. Two onsets inside roughly 30 to 40 ms fuse perceptually, and the earlier one dominates the perceived location, which is why a stagger in this range reads as one gesture with an internal order instead of four separate events. Push the stagger to 80 or 100 ms and it becomes a run. That is a different message, right for a channeled or chained heal and wrong for an instant group heal.
The layers in that gesture do not have to be complete cues. Use fragments: a 20 ms noise tick, a 60 ms tonal pair, a filtered breath, a single bell partial. Give each fragment its own spot in the field, high-pass them so the low end does not pile up, and pan the whole gesture across the field over 150 to 250 ms. What the player hears is coverage of the group.
Do not apply any of this to the single heal. A staggered single-target heal sounds like a slapback echo, and an echo on a confirmation makes the confirmation uncertain. One onset, one body.
A Wide Gesture That Survives Mono
Width comes from decorrelation, not from hard panning. Two layers of the party cue with slightly different detune and different early reflections, panned in opposite directions, create a wide image without a hole in the middle. Keep everything below about 700 to 800 Hz mono. A hard-panned low body sounds broken on headphones and loses most of its energy on a phone speaker.
Then check the sum to mono. If the identity of the party heal lives only in the side information, it disappears on the device most players will actually hear it on. Put the recognisable content in the centre layer and treat width as the extra. This is the same discipline as layering without muddying the mix, applied to a single event instead of a whole scene.
Reverb for a party heal often belongs on a shared global bus rather than a positional one, because the effect is global. Keep it short, 0.4 to 0.6 s, and let it duck the other heal cues for about 200 ms by 6 dB. That is closer to ducking the mix under an effect than to reverb design, and it is the part that keeps a four-heal moment from turning into a wash.
The Single Heal Stays Close and Short
Close voicing means the interval sits inside an octave and the partials are packed near each other, a major third around 660 and 830 Hz rather than a root-octave spread. The result is intimate at low level and does not eat much spectrum, which leaves room for the boss, the weapon and the music.
Keep the whole cue between 120 and 180 ms: a 3 to 8 ms attack, a 40 to 80 ms body, a 40 to 60 ms exponential release. Skip the stagger and skip the wide reverb, or use a 0.2 s room at low send. If the target is an ally, position the sound at the ally with distance attenuation. If the player healed themselves, play it 2D at the listener with no attenuation and no panning. A self-heal that pans is lying about where it came from.
A combat single heal is usually tighter than a pickup cue. the 0.3-second health pickup cue plays while the player keeps moving, but a single-target heal in a boss fight has to survive under telegraph sounds and often has to finish before the next mechanic resolves. Treat 180 ms as the working ceiling and go shorter when the palette allows it.
A Small Decision Table
These are starting points, not a specification. The right values depend on the game's palette and on how many healers can act in the same second.
| Question | Party heal | Single-target heal |
|---|---|---|
| Where does it play? | Global bus, non-positional | At the target; 2D at the listener for a self-heal |
| How many voices? | One summary cue, two to four internal layers | One voice |
| Stereo treatment | Wide above 800 Hz, mono below | Narrow and near centre |
| Onset structure | Layers staggered 30 to 60 ms | Single onset, no stagger |
| Total length | 180 to 260 ms | 120 to 180 ms |
| Tail and reverb | Short shared send, 0.4 to 0.6 s, ducked | Dry or 0.2 s room, mostly dry |
| Target count | Drives layers and width, never gain | Fixed |
Two checks settle most of this faster than any table. Solo the heal bus during a four-healer pull and listen for whether you can still hear the boss. Then collapse the whole mix to one speaker and confirm that the party heal still tells you the group was healed. If it does both, the two cues are doing their own jobs and have stopped fighting over the same 120 ms.