The jump sound is the most exposed sound in a platformer. It fires more often than any other effect, it is tied to the player’s primary input, and it is heard in the same acoustic context for the entire session. That combination makes it the first sound to become tiring and the hardest one to fix with a single change.
The problem has a name. Berklee’s game audio course material describes it as repetition fatigue: if a player hears a distinct, identifiable sonic event repeated at a fixed interval, the brain maps the pattern, the immersion breaks, and the sound stops functioning as feedback[reference:0]. A jump that sounds the same every time does not become “a jump sound.” It becomes a pattern the player can predict, and a predictable sound is one the player starts to tune out.
The mistake is to treat variation as one setting — a random pitch value, or a pool of three files — and expect it to solve the problem. Variation is a layered system, and each layer has a different threshold, a different cost, and a different failure mode. The sections below work through the layers in the order they should be built: the sample pool, the pitch and volume range, the selection rule, and the state response.
Layer 1: A Sample Pool of Three Is the Minimum, and It Is Not Enough on Its Own
The first layer is the cheapest and the most basic: more than one take. A documented production pipeline for a game with over 500 creatures provides at least three sound variants per creature per state, specifically to eliminate the player’s sense of auditory repetition[reference:1]. That number is not arbitrary. Three takes give the selection rule enough material to avoid immediate repetition, and they give the ear enough difference to hear the sound as a family rather than a single file.
But a pool of three with a naive random selection still repeats within a short window. With three samples and a uniform random pick, the probability of the same sample twice in a row is one in three. Over a hundred jumps, that happens dozens of times, and each repetition is a moment where the pattern is exposed. The pool is necessary, but it is not sufficient.
The sound variation system guide covers the broader architecture of a variation pool; the jump-specific point is that the pool size is the floor, not the strategy.
Layer 2: The Selection Rule Matters More Than the Pool Size
The second layer is how a sample is chosen from the pool. A uniform random pick is the default in most engines, and it is the weakest rule for a high-frequency sound, because it allows consecutive repeats. The documented alternatives are round-robin and shuffle.
Round-robin cycles through the pool in order and does not repeat until the pool is exhausted. It is simple, predictable in a good way, and it guarantees that the same sample cannot play twice in a row unless the pool has only one entry. The cost is that the order is fixed, so a player who jumps at a consistent rhythm may start to hear the cycle. That is a rarer failure than consecutive repetition, and it is a reasonable trade.
Shuffle is the stronger version: it removes each sample from the pool after it plays, and only resets the pool when all samples have been used. Wwise documents this as the Shuffle mode, where “objects are removed from the pool after they have been played,” avoiding repetition until every object has been played[reference:2]. A separate Wwise guide adds a second control: Avoid repeating last x played, which excludes the most recent entries from the selection even after the pool resets[reference:3].
For a jump, the working configuration is shuffle with an exclusion of the last 1–2 samples. That guarantees that the same jump sound cannot play twice in a row and cannot play third in a row either, which is the repetition window the ear is most likely to notice. The randomized blip variation guide covers the same selection logic for a different sound type; the jump case is a high-frequency application where the consecutive-repeat window matters more than the long-term distribution.
Layer 3: Pitch Variation Has an Identity Threshold, and It Is Narrower Than Most Advice Suggests
The third layer is pitch variation. The common advice is to
“randomize pitch slightly,” and the specific values vary
widely. A game jam post describes randomizing pitch between
3 and 6 values to reduce repetition.
A Unity tutorial shows Random.Range(0.8f, 1.5f),
which is roughly −4 to +7 semitones — a range
that would make a jump sound like a different move on some
triggers. A Steinberg reference describes a random
pitch offset of ±6 semitones at 100%
intensity[reference:6].
Those ranges are too wide for a jump. The identity threshold for a short, contour-heavy sound is tighter than for a footstep or a weapon. A production guide for electronic music puts the acceptable shift at “a semitone or tone… but not a third” [reference:7], which is roughly ±1 to ±2 semitones. That range is enough to break exact repetition without changing the sound’s identity.
There is a second reason to keep the range narrow. The just-noticeable difference for complex tones is around 9 cents[reference:8], which is less than a tenth of a semitone. A variation of ±1 semitone is already many times the JND, so the ear will hear it as a variation. Pushing to ±3 semitones does not add more “variation”; it starts to add “a different jump.”
The practical range for a jump is ±1 to ±2 semitones, applied as a random offset per trigger. If the engine exposes pitch as a multiplier, the conversion is 2^(N/12): one semitone is a multiplier of about 1.06, and two semitones is about 1.12[reference:9]. A multiplier range of 0.94–1.06 covers ±1 semitone, and 0.89–1.12 covers ±2. The pitch variation threshold guide covers the JND and identity-threshold distinction in more detail; the jump-specific conclusion is that the upper bound is lower than the generic advice suggests.
Layer 4: Volume Variation Is a Separate Axis, and It Should Be Smaller Than Pitch
Volume variation is often bundled with pitch variation, but it has a different perceptual role. Pitch variation changes the sound’s identity; volume variation changes its emphasis. A jump that is slightly louder reads as a stronger push, not a different jump.
Because volume is less identity-bearing, it can be varied without the same tight threshold. A range of ±1 to ±2 dB is enough to add micro-variation without making the jump feel inconsistent. A larger range starts to interact with the mix: a jump that is 4 dB quieter on one trigger may disappear under the music, and a jump that is 4 dB louder may stick out.
The phone speaker guide covers the level relationship between jump and music; the volume-variation point is that the range has to stay inside the window where the jump is always audible and never dominant.
Layer 5: The State Response Is the Layer Most Games Omit
The first four layers vary the sound without changing its meaning. The fifth layer changes the sound in response to game state, and it is the layer that keeps the jump from becoming a fixed pattern even when the player is doing the same thing repeatedly.
The clearest example is the jump’s pitch relationship to vertical speed. A platformer developer describes an inverted mapping: the pitch is lower at higher vertical speed, with the highest pitch at the jump apex. The result is a jump that sounds like it is laboring on the way up and relieved at the top[reference:10]. That is not a random variation. It is a state-driven variation, and it is far more effective at breaking the pattern than any amount of pitch randomization, because the sound is responding to what the player is actually doing.
A second state response is surface. A landing already inherits the surface material from the footstep system; a jump can inherit a smaller version of the same cue, so a jump from gravel and a jump from wood are not identical. The footstep surface guide covers the surface vocabulary; the jump can borrow a thin layer of it without becoming a footstep.
A third is jump type. A short hop, a full jump, and a double jump are not the same event, and they should not share the same variation pool. The short hop versus full jump guide covers the structural differences; the variation strategy is that each type gets its own pool, its own selection rule, and its own pitch range.
A Working Configuration, Layer by Layer
The table below gives a starting configuration that combines the five layers. The values are starting points, not a specification, and the order of the rows is the order in which the layers should be built.
| Layer | Setting | What it prevents |
|---|---|---|
| Sample pool | 3–5 takes per jump type | The same waveform on every trigger |
| Selection rule | Shuffle, exclude last 1–2 | Consecutive and near-consecutive repeats |
| Pitch range | ±1 to ±2 semitones (multiplier 0.94–1.06 to 0.89–1.12) | Exact repetition without changing the jump’s identity |
| Volume range | ±1 to ±2 dB | Micro-variation without disappearing under or dominating the music |
| State response | Pitch mapped inversely to vertical speed; surface layer inherited from footstep material | The jump becoming a fixed pattern even when the player is doing different things |
The configuration is not expensive. A pool of three to five takes is small; the selection rule is a property of the container in most engines; the pitch and volume ranges are runtime parameters; and the state response is a single mapped curve. The placeholder sound guide covers the prototype stage, where a single placeholder is acceptable; the variation strategy is what replaces the placeholder once the jump is final.
The layer that most projects skip is the last one. A jump that is randomized but not state-responsive is still a pattern with noise. A jump that responds to speed and surface is a sound that is doing something, and a sound that is doing something is much harder for the ear to dismiss as repetition.