How our switch sounds are made

Every one of the 126 switch sounds on this site is synthesized, and this post explains exactly how our switch sounds are made, because "trust me" is not good enough for audio you might base a purchase on. The short version: a keystroke is not one sound but up to four separate acoustic events, our generator models each event from the switch's published specifications and its documented sound character, and we label the result "simulated" everywhere it plays because that is what it is.

The four acoustic events of a keystroke

Listen to a slowed-down keystroke and you can pick apart four distinct things happening, and a credible simulation has to model them separately rather than playing back one blob of noise.

1. Downstroke noise. As the stem slides down its rails you get friction sound, the scratch and hiss enthusiasts complain about on unlubed switches, plus a tiny tick as the metal leaf makes contact. It is the quietest of the four events, and its character is dominated by lube state, which is why a factory-lubed switch gets no scratch layer from us at all, only the contact transient.

2. Bottom-out. The stem pole and slider base strike the bottom housing. This is usually the loudest event of the whole press, and it is where "thock" versus "clack" lives. Housing plastics and long stem poles account for much of that on a real switch, and neither is a field in our switch database, so the generator does not pretend to model them. It builds the bottom-out from the fields it does hold: a heavier switch lands on a lower, fuller body, and a short-travel or low-profile switch gets a shorter ring-out afterwards.

3. The upstroke clack. Release the key and the stem slams into the top housing on its way back. It is often nearly as loud as bottom-out, which is why real typing sounds like a double tap per keystroke, and it is precisely what silent switches dampen alongside the bottom-out.

4. Parasitics. Spring ping (a high metallic ring from coil oscillation), leaf crunch on tactiles, stem wobble rattle, and on a real board, stabilizer rattle, where the spacebar alone can dominate how a keyboard actually sounds in a room. We render the two the specs can predict: spring ping on switches that are neither factory-lubed nor silenced, and a delayed stabilizer tick on every key wide enough to be stabilized.

Clicks are five different mechanisms, not one

"Clicky" hides five genuinely different sound machines, and the generator carries a separate click model for each, because pitching one click sample up and down could never make a Blue sound like a Jade. A Cherry MX Blue uses a click jacket: a loose two-part stem where the jacket is released to fly down and strike the housing, giving a higher-pitched, slightly rattly click. A Kailh BOX Jade uses a click bar, a small metal torsion bar that snaps on both press and release, so you get a crisper, more mechanical double click, and thicker bars mean louder and crunchier. An IBM buckling spring is its own thing entirely: the spring physically buckles and its hammer strike is simultaneously the tactile event, the click, and the actuation.

The other two are the ones buying guides forget. Vintage Alps parts like the SKCM Blue, and the Matias Click that descends from them, click off the leaf itself, so the click lands with the tactile event instead of trailing it: tighter and fuller than a jacket, with less ring than a bar. The NovelKeys Cream Clickie pops a bowed metal disk, the one enthusiasts call the "snapple cap", which behaves more like a bottle cap than like anything else on this list. A switch with no click mechanism gets no click layer at all, which is why a Topre 45g sounds like nothing else here. In place of a click it runs the rubber-dome recipe: a snap at actuation, louder and higher than a leaf bump because it has to be heard over what follows, then the soft swell of the dome collapsing into the same bottom-out body every other switch lands on.

From spec sheet to sound

The synthesis is parametric: the same fields you see in the spec table on every switch page drive the audio. Switch type is linear, tactile or clicky, and it picks the event recipe. Silencing is not a fourth type, it is a separate flag, and the Topre family is not a type either: our 5 electrocapacitive entries are filed as tactile switches and carry a family flag the generator reads to give them the dome behavior above.

Force sets weight, and weight sets pitch and impact energy: the heavier the switch, the lower and harder it lands. The awkward part is that 47 of our 126 switches have no published bottom-out at all, because most brands rate actuation, or a tactile bump peak, and stop there. Rather than invent a number, the generator uses a rated bottom-out where one exists and otherwise takes the highest figure the brand did publish: a tactile bump or click peak, or an actuation force scaled up to the bottom-out it implies. That is why Kailh BOX Navy is weighted by its published click peak instead of being read as a light switch on its actuation figure alone.

The click mechanism picks the click model described above. The silent flag swaps hard impacts for dampened, cushioned transients on both strokes, the way real silencing pads work, and takes the whole pack's output level down with them, so a silent switch is genuinely quieter here rather than normalized back up to match a clicky one. Factory lube removes the scratch and spring-ping layers outright, and an unlubed budget switch carries more slider scratch than an unlubed premium one. The last input is the switch's published sound character, the same sentence you read on its library page, and it is deliberately the weakest input in the model: it can bias the voicing by at most about two and a half semitones toward deep or bright, because a one-line description is not a measurement. A switch like the Boba U4T gets its deep, full bottom-out from its weight and its short ring-out from its short stroke, with the documented character as a nudge on top of both, not from us hand-tuning it to match a YouTube video.

This has a real advantage over recordings for one specific job: comparison. Every switch is rendered through the same model under the same virtual conditions, so when you A/B two switches the difference you hear comes from their specs, not from one being recorded on a foam-filled gasket board and the other on a hollow plastic tray mount with a different microphone.

Why the "simulated" label matters

Much of the switch-sound ecosystem runs on community packs with undocumented origins, frequently ripped from typing-test videos that belong to their creators. We decided at the start that nothing on this site would be built that way: no ripped audio, no unlicensed packs, no laundering someone else's recording through an open source repo. Synthesis was the honest way to launch with full coverage, and labeling it "simulated" wherever it plays keeps the deal with you clear. The full licensing position lives on the credits page.

The recording roadmap

Simulations are the floor, not the ceiling. The plan we are executing: buy the switches, mount them in real hot-swap boards rather than lone testers (a naked switch sounds nothing like one in a case), and record per key with multiple takes at varied velocity, capturing press and release separately since the upstroke is half the sound. Recording the same switches across more than one board configuration, a foam-filled gasket build and a brighter metal-plate build, is what will eventually power honest "same switch, different board" comparisons. As studio recordings land they replace simulations switch by switch, clearly marked, and we intend to publish the recorded library under a Creative Commons BY 4.0 license so the community can finally build on a sound set with clean provenance. Until then, you can hear every current simulation in the switch library, or put one under your fingers in the typing test.

Frequently asked questions