Run two hundred and something. The player is at thirty percent health, sitting on a potion they are saving for the boss, and a heal fires from a support unit they forgot they had recruited. The number in the corner moves. The sound plays. They notice neither, because it is the same 280-millisecond file they have heard every forty seconds for six hours, and their ear filed it under room tone sometime yesterday.
The cue is not bad. It was probably good at run ten. The problem is that a heal cue in this genre is written for the dramatic case and then played overwhelmingly in the undramatic one, and nothing in the design separates those two situations.
Every lever below is a way of spending a budget the player grants once. The budget is attention, and it does not refill.
The Curve Has Two Failure Modes and They Need Different Fixes
A heal cue wears out in one of two directions. Habituation is the quiet one: the sound still plays, the ear still receives it, and the player stops treating it as information because nothing about it ever varies. Irritation is the loud one: the player hears every instance and has started to resent it. A cue that is loud, bright and unchanging tends to reach both, and only the second one gets reported.
Do the arithmetic before you tune anything, because the number is usually worse than the intuition. A thirty-minute run that heals every forty to sixty seconds puts the cue in front of the player thirty to forty-five times; at a hundred and fifty runs that is several thousand plays of one short file, and every play spends whatever decorative tail the cue carries.
The curve is not linear. The tenth play still does its job. The hundredth is neutral. After that the cue either joins the wallpaper or becomes the reason someone closes the game, and which one happens depends on whether it still tells the player something true.
Scale the Cue to the Heal Before You Scale It Down
The most common source of heal fatigue in these games is a four-point heal from a common drop playing the same cue as a full-restore shrine. Players see twenty of the small ones for every large one, so the cue is almost always bigger and more decorated than the event deserves.
Expose the heal amount as a continuous parameter and drive the cue from it. In Wwise that is an RTPC named heal_pct; in FMOD it is a parameter of the same name with a zero-to-one range, normalised against the player max health. Map that single value to gain, low-pass cutoff and layer count together, so the cue shrinks in every dimension instead of only getting quieter.
- 1 to 10 percent of max health. Transient plus a sixty-millisecond body, no tail, around -14 dB and low-passed near 4 kHz. This tier confirms; it does not celebrate.
- 11 to 35 percent. Full body with a 120-millisecond tail, around -8 dB, filter open. This is the sound most players would name if you asked them what healing sounds like.
- 36 percent and up, or a crit heal. Full cue plus the decorative layer, around -5 dB. Rare by definition, which is exactly what lets it be the one that still lands.
Keep the first thirty milliseconds identical across all three tiers. Size should change body and tail, not the transient, or the small heal stops reading as a heal and the player learns the cue twice.
Cap the Voices and Merge Stacked Heals Into One Event
Auto-battlers with pets, regeneration auras and multi-target heals will fire several heals inside the same frame. Five overlapping copies of one file do not read as five heals. They read as one heal with a comb-filtered smear and a level jump of roughly seven decibels, because correlated identical sources sum at about three decibels per doubling.
Set a hard cap of two or three concurrent heal voices. In Wwise that is a limit-to-instances property with an explicit kill rule; in FMOD it is max instances with the stealing mode set to quietest. The fourth heal in a frame should not spawn a voice at all.
Then aggregate deliberately. Open a hundred-to-150-millisecond merge window, count the heals inside it, and play one instance at base level plus ten times the base-ten log of that count. Two stacked heals land around three decibels louder, four around six. The player hears one bigger heal, which is what actually happened to their health bar, instead of four copies fighting over the same spectrum.
The Merge Needs Its Own Envelope
A merged event that is only a gain offset can sound like a glitch when the stack resolves late. Rather than retriggering the cue, let the merged instance stretch its release by thirty to fifty milliseconds per additional stack and stop at about four hundred. Duration carries the sense that more happened, and the cue still closes before the next event arrives.
Rotate Families, Not Sixteen One-Off Samples
Variation is the standard answer to repetition, and here it is half right. Sixteen distinct heal sounds is not variety; it is sixteen cues the player has to learn. What survives hundreds of runs is two or three related families, each holding four to six variations, so the ear can generalise instead of memorising.
A workable structure: three families separated by register and timbre, say a low warm body, a mid bell, and a breathy filtered noise layer, with five variations inside each. Fifteen files, one contour. A random container with avoid-repeating-the-last-two plus pitch randomisation inside plus or minus twenty-five cents breaks the machine-gun repeat without ever making the player ask what they just heard. The reasoning behind that approach is set out in building a variation system for repeated cues , and it matters here more than for most sounds because the play count is higher.
What Has to Stay Identical
Root register, transient shape, and the direction of the pitch gesture. If one variation rises a fifth and another falls a fourth, the player is receiving two different messages from one cue. Change articulation, noise colour and decay length; leave the grammar alone.
Small Heals Belong in the HUD, Decoration Belongs to Rare Heals
Regeneration on half-second ticks is a two-hertz event. Putting a voice on every tick is a decision to run the heal cue as a drone, and players will treat it as one. Below roughly 3 percent of max health, the health number is the better channel: a twelve-frame brightness shift on the numeral, no audio at all.
The cost is real. Silent healing is easy to miss, and a player who does not know they are regenerating will misjudge a fight. Pair the HUD change with a sustained indicator, a soft outline on the bar or a particle that persists while the effect is active, so the information is continuous rather than event-based.
The decorative layer works the opposite way. Choir, shimmer, an added octave, a sparkle around 2.5 kHz: those are what make a large heal feel large, and they only stay legible while they are scarce. Reserve them for crit heals, shrine heals and the full restore. A decoration played three thousand times is not decoration, it is noise with a nicer name.
Why Lowering the Global Heal Bus Costs Readability
The lazy fix is a six-decibel cut on the heal bus. It reduces irritation immediately, which is why it survives playtests, and it takes away the thing the cue was for. Six decibels is close to half the perceived loudness, and the heal now sits underneath footsteps, weapon tails and impact debris across the one-to-four-kilohertz range, which is the exact band that fills up during the fight where the confirmation matters most.
Worse, a bus change flattens the size axis you just built. A four-point heal and a shrine heal arrive at nearly the same level, so the cue distinguishes nothing and becomes a generic blip. Irritation has been traded for invisibility, and invisibility is harder to catch in a playtest because nobody complains about a sound they have stopped hearing.
Cooldown Cues Draw From the Same Budget
These games also fire a steady stream of not-ready sounds: ability on cooldown, not enough mana, no charges left. In many builds those outnumber heals. Every decibel handed to the heal comes out of that account, and the reverse holds too. If healing has to feel reliable, the cooldown cue should get shorter and drier, not louder to compensate. The companion notes on the sound of an ability on cooldown cover the other side of that trade.
What works instead is spending the budget on differentiation: size tiers, a voice cap, a merge rule, rotating families, a HUD channel for the ticks. None of those is a volume cut.
The base cue itself, with its attack, interval, timbre and release, is a separate problem covered in the guide to health pickup cue design . Fatigue management starts after that cue already works. It is the second design problem, and in a roguelike or an auto-battler it is the one the player will actually live with.