SFX Forge — UI Sound Designer (exports 140 WAVs)

Design a complete UI sound set in your browser, then export all 140 files at once.

The demo above is the real tool. Every action, every voice and every parameter is live, and single-sound .wav export works - try it properly before deciding.

It starts from an action, not from noise

Most sfx generators hand you a randomise button and a waveform. That is fine for a laser or an explosion, and useless for an interface, where the job is not one good sound but thirty-five sounds that belong together.

SFX Forge starts from the action instead. Pick confirm and you get a rising gesture; pick cancel and you get a falling one; hover is one short tick at the edge of perception. Then you shape it.

Four voices, one action set

• Minimal — Clean modern-app. Sine through triangle to saw as you open the brightness, with a contact transient so a tap reads as a press rather than a beep.
• Retro — Chiptune. Pulse waves whose duty cycle tracks brightness, plus real bit and rate reduction.
• Sci-Fi — Holographic. Two-operator FM at non-integer ratios, so the partials land off the harmonic series and read as glass.
• Soft — Warm wooden mallet. FM at 4:1 with a low index and a slow attack - the defining shape of marimba and kalimba tone.

Every voice renders the same 35 actions under the same filenames, so switching your project from one to another is a folder swap rather than a rename job.

Ten parameters, and all of them do something

• Pitch — -24 to +24 semitones
• Decay — 0.25x to 4x the note's tail
• Brightness — harmonic content and filter corner together
• Texture — how much contact noise the attack carries
• FM ratio / index — inharmonicity, on the two FM voices
• Bit crush — quantisation, on every voice
• Drive — saturation
• Room — reverb, scaled per action
• Level — peak target in dBFS

Parameters a voice does not use are greyed out rather than left sitting there inert - Minimal and Retro have no FM operators, so they do not pretend to.

Exports arrive already mixed

This is the part that saves the afternoon. Levels are assigned by role, not normalised flat:

• -24 dBFS — hover
• -20 dBFS — scroll_tick
• -18 dBFS — slider_tick, type_key
• -10 dBFS — click
• -6 dBFS — error, success
• -3 dBFS — complete, level_up

Drop a set in at unity gain and the relationship between a whisper and an event is already correct. Reverb follows the same logic: anything a player can trigger repeatedly - hover, scroll ticks, slider detents, keystrokes - stays bone dry, because tails on those stack into mush the moment a cursor crosses a menu.

Audition the whole set as one file

Build a sequence from any sounds - across voices if you like - set the spacing, and play or save it as a single .wav.

• All 35 actions in one pass, to check the set holds together and the levels sit right against each other
• Current sound x8, the repetition test: a UI sound that survives being spammed is a different thing from one that only sounds good once
• Anything you assemble by hand, for a client preview or a storefront clip

Sounds are allowed to overlap at close spacings, because that is what a real interface does - a player clicks again before the last tail has finished.

What you get

• The full tool, batch export unlocked: 35 sounds per theme, 140 for all four, one zip, with a README documenting the gain structure
• 44.1 kHz / 16-bit / stereo WAV
• Runs in any modern browser. Nothing to install, and it keeps working offline once downloaded - it is a single folder of HTML and JavaScript, about 60 KB
• Sequence rendering and export, for auditions and previews
• Unlimited sets. The licence covers the sounds you export, for unlimited commercial projects, with no attribution required

Licence

Royalty-free. Everything you export is yours to use in unlimited personal and commercial projects, no attribution, no revenue cap. You may not resell the tool itself, or redistribute exported sets as a sound pack of your own.

How this was made, plainly

This page is tagged AI Generated → Code, and the whole product is code, so that tag covers all of it.

No generative model produces the audio. There is no diffusion model, no audio model and no training set anywhere in the chain - every sample is arithmetic: oscillators, FM operators, exponential envelopes, biquad filters, a comb-allpass reverb. The same category of tool as a hardware synth. What AI wrote is the DSP code doing the arithmetic.

The practical consequence is a good one: nothing was sampled, recorded or scraped, so there is no third-party material inside anything you export and no third-party claim available against your use of it.

Prefer a ready-made pack?

The same engine rendered UI SFX Vol.1 — 236 sounds, fixed, WAV and OGG, if you would rather not turn any knobs. There is a free 32-sound sampler of it too.