The player is at 18 HP, takes the heal, hears the bloom open, and steps back into the boss. The bloom is still ringing when the slam telegraph starts, and the two sounds arrive as one event. That is not a volume problem. The tail covered the tell.
A heal cue has two jobs that sit in different parts of the envelope. The attack says something happened. The tail says you are recovering, and roughly for how long. Most heal sounds are designed for the first job and inherit the second by accident, from whatever reverb preset was already on the bus.
The Attack Announces, the Tail Confirms
Inside the first 50 ms, a heal cue is doing recognition work: the ear decides that an event occurred and roughly what kind. Everything after that is state. The tail is the only part of the sound that stretches past the frame in which the player is already deciding what to do next.
A 0.4-second room and a 2-second plate can share the same attack and the same pitch gesture and still say different things. One says top-up. The other says the recovery is still running while you keep moving. Players learn the difference fast when the game is consistent about it.
Design the dry cue and the wet return as separate layers. The dry layer carries the event. The wet layer carries duration, and duration is how the player judges how much health came back.
Decay Time Should Track Heal Size
If every heal uses the same reverb time, players cannot tell a sip from a full restore without looking at the bar. The tail is the cheapest place to encode magnitude: no extra layers, almost no extra headroom.
Let the decay scale with the fraction of the health pool that moved. A chip heal gets a tail just long enough to confirm contact. A full restore gets a tail still audible after the player has resumed firing.
A staged tail beats one long reverb
One long reverb smears the energy into a single slab whose first 300 ms is no clearer than its last second. Two layers do better: a short room around 0.4 s doing the confirmation work, plus a long plate 16 to 20 dB below the dry signal. The player hears confirmation first, recovery second.
Keep the long layer low enough to compete with nothing, but present enough that muting it changes the feel. If muting it changes nothing, it is not doing work.
| Heal type | Health restored | Dry cue | Reverb decay | Delay repeats |
|---|---|---|---|---|
| Regen tick | 2 to 5 percent of max | 80 to 150 ms | 0.3 to 0.5 s | none |
| Pickup heal | 15 to 25 percent | 200 to 350 ms | 0.6 to 0.9 s | none |
| Combat heal | 35 to 50 percent | 300 to 450 ms | 1.0 to 1.4 s | 1 to 2 |
| Full restore | 100 percent | 500 to 700 ms | 1.6 to 2.2 s | 2 to 3 |
| Revive | 100 percent plus invulnerability | 700 to 900 ms | 2.0 to 2.5 s | 3 to 4 |
These are starting points, not a spec. A 2.2-second tail in a game where damage lands every 800 ms will bury the next hit. Shorten until the damage cues still read, and the heal will feel faster without losing its warmth.
Pre-Delay Is What Keeps the Attack Clean
Pre-delay is the gap between the dry transient and the first audible reflections. It is the cheapest control on this list and the one most often left at zero.
At 0 to 10 ms the reverb fuses with the attack. That can work for a soft ambient heal, but the bloom starts inside the transient that has to cut through gunfire. At 15 to 30 ms the attack gets a clean window: strike first, room second. Past about 40 ms the gap reads as its own event, and a second event after a heal can read as lag.
Engine defaults vary, so set it explicitly. Godot AudioEffectReverb exposes predelay_msec. Wwise RoomVerb has PreDelay, FMOD Studio SFX Reverb has Pre-Delay, Unreal submix reverb has PreDelay. Unity AudioReverbZone has no pre-delay, so insert a plain 15 to 20 ms delay before the send instead.
The transient needs the same care. An 8 ms noise burst is doing the recognition work described in the first fifty milliseconds, and a reverb with no pre-delay fills it in before it finishes.
Delay Repeats Turn a Heal Into Ongoing Care
For an instant heal, repeats are usually a mistake: a repeat 300 ms later reads as echo, and echo reads as distance or a dropped frame. For a heal over time the same repeats become the mechanic made audible. The health ticks, so the sound ticks.
Space repeats to the tick, not to the tempo
If the heal applies every 800 ms, repeats at 800 ms land on the actual health change. If it applies every 250 ms, spaced repeats blur into a shimmer: usable as a bed, useless as a count.
Start with 250 to 400 ms, feedback 15 to 25 percent, each return 6 to 9 dB below the previous one. High-pass the return around 300 to 500 Hz and walk the low-pass down toward 3.5 kHz as the repeats decay. Damping matters more than delay time.
An outside source is the exception: a shrine, a support character, a field effect. There a wider ping-pong can suggest something else is holding the heal up, but check the cue in mono, where hard panned repeats lose level.
Spacing is the whole design. The patterns in the sound language of heal over time assume the ear counts repeats it can predict, and the count is what reads as duration.
Wet Level Changes Distance, Not Size
Raising the wet return does not make a heal bigger. It makes it further away. That matters because the reflex when a heal sounds weak is more reverb, and the result is a cue that seems to happen in the next room while the player is hit in this one.
For a combat heal, keep the dry signal dominant, roughly 75 to 85 percent, with the wet return 10 to 14 dB down. The heal happens to the player and needs to read close. For a shrine or a revive from an outside source, half wet can work, because the event happens around the player.
Low-pass the wet return around 4 to 6 kHz to push the tail behind the dry cue and add distance without level. Out of combat the ceiling rises, which is why heals in the quiet between fights can afford a much wetter treatment.
When the Tail Eats the Next Tell
The failure is specific. A heal with a 1.4-second tail lands 600 ms before a telegraph sweeping through 2 to 4 kHz. The tail is still in that band, the telegraph is quieter than the heal was, and the warning never arrives. Nothing was wrong with the sound. The player simply did not hear the thing that mattered.
Test it with the screen off: heal, wait the intended gap, damage cue, five times. If you cannot count the damage cues, the tail is too long, too loud, or both. Change one variable at a time.
Budget tails by encounter, not per sound
Tails sum. Six heals in ten seconds with a 1.2-second tail each is overlapping tails for most of the fight, and the reverb bus stops being events and becomes a bed. Shorten combat tails or cap how many can be alive.
Five changes cover most of it:
- Cap reverb time by heal tier. Combat heals get the short room, not the plate.
- Duck the tail under threat cues. A 6 to 9 dB sidechain on the heal return, keyed from telegraphs, keeps the tail audible in calm moments and out of the way when it matters. Same mechanics as ducking under a cue.
- Shorten the wet layer, not the dry cue. The dry cue can stay under 400 ms while the reverb is halved.
- Cut the previous tail on retrigger. A new heal fades the old one by 30 percent over 50 ms.
- Carve 1.5 to 4 kHz. A gentle dynamic dip where telegraphs live costs almost nothing in perceived size.
The tail tells the player how much came back and whether recovery is still running. It is also the part most likely to hide the next thing that will hurt them. Design it in milliseconds, and test it against the damage cue that follows rather than the silence of the sound lab.