engineVoices

engineVoices gives cars their sound: eight engine voices that follow the revs from idle to limiter and how hard the engine is working, plus gear changes you can hear, a rev-limiter stutter, turbo spool and blow-off, overrun pops, tyre skids that change with the surface, and crash crunches. Every sound is spatial and Doppler-shifted, so a car passing you drops in pitch. The values that drive it replicate, so every player hears every car.

It reads only the shared driving contract: the car.* named values on a car's chassis entity (torqueCars writes them, and so can any other car). Install it, put an EngineVoice on the chassis, and the car has a voice.

Quick start

Paste this into any script that runs in play, such as a scene's entrypoint.luau. It needs no car: a stand-in entity writes fake car.* values, running through idle, a full-throttle pull through the gears, the rev limiter, a lift with pops and blow-off, a slide and a crash, then moves on to the next voice.

local started = false

function update(dt)
    if started then return end
    local cam = layers.active.camera
    if not (cam and cam.ready and cam.entity) then return end       -- wait for the play camera
    if audio.listener().count == 0 then
        cam.entity.component.add("AudioListener")                -- the ears
    end
    local demo = entity.spawn("demoCar")
    demo.position = { 0, 1, -12 }
    demo.component.add("EngineVoiceDemo", { voice = "cycle" })  -- writes car.rpm, car.load, ...
    demo.component.add("EngineVoice")                           -- plays them
    started = true
end

To hear all eight voices plus a fly-by in one line, use require("engineVoices.showcase").start(), or load the EngineVoicesShowcase scene in this package.

On a real car, put the component on the chassis entity (the one with the car component, which writes the car.* values):

local Voices = require("engineVoices.voices")
Voices.attach(carChassis)            -- voice from the car's car.voiceId
Voices.attach(carChassis, "v12")     -- or force one
Voices.set(carChassis, "rotary")     -- switch it live; "auto" goes back to car.voiceId

Or set it in the Inspector: EngineVoice.voice is one setting.

The voices

How it sounds the way it does

• Engine loops. Each voice has six seamless one-second loops: three rpm bands, each with an on-load and an off-load take. All 48 were synthesized by engineVoices.synth from the engine's own firing pattern, using pressure pulses and combustion noise per cylinder firing, per-cylinder imbalance and jitter, exhaust-pipe resonances per bank, diesel knock, valvetrain ticks and misfires. A loop holds a whole number of engine cycles, so it has no seam. The spectral peak of every loop sits on the engine's firing frequency at the band's rpm, which Synth.analyze measures.
• Following the revs. Each band loop is pitched to rpm / band rpm and the bands crossfade at equal power in log-rpm, so the note rises smoothly from idle to limiter without any loop being stretched far from where it was made.
• Following the load. car.load crossfades the on-load take (louder, harsher, more low harmonics) with the off-load take (quieter overrun). Lifting at high rpm on a v8, rotary or turbo fires random exhaust pops. Lifting at boost on the turbo blows off.
• Gear changes. When car.shiftCount changes it plays a manual clunk, a sequential cut-and-bang (with a short ignition cut on the engine layers), or a soft automatic nudge. The style is read from the Car's gearbox, or set with shiftStyle. Electric makes no shift sound. The rpm drop itself is heard through the loops.
• Limiter. While car.limiter is true, the on-load layer is chopped at about 16 Hz (the fuel cut), with crackle on the popping voices.
• Turbo. The whistle's volume and pitch follow car.boost.
• Skids. The skid loop's volume follows car.skid, and the loop follows car.surfaceId: squeal on tarmac, concrete, cobbles, rumble strips and oil (quieter on wet tarmac and ice), crunch on gravel and dirt, scrub on mud, grass, sand, snow and puddles.
• Crashes. When car.impactCount changes it plays a light tap, a medium crunch or a heavy smash, chosen and scaled by car.impactSpeed.

API

• EngineVoice component, on the chassis. Fields: voice (auto or a name), volume, maxDistance, shiftStyle, skids, crashes, doppler. :mix() reports the voice, smoothed rpm, load and boost, band weights, on/off-load gains and its source ids.
• engineVoices.voices: names(), has, get, nameOf(id), idOf(name), attach(car, voice?, volume?), set(car, voice), detach(car), clips(), surface(id), firingHz(voice, rpm), loopName.
• EngineVoiceDemo component: a stand-in car that writes the car.* values (voice = "cycle" or a name, flyby, center, distance, speed).
• engineVoices.showcase: start(opts?), stop().
• engineVoices.synth: bakeAll(), bakeVoice(name), bakeShared(), render(voice, band, onLoad), analyze(clip). It re-bakes every clip deterministically after a parameter change.

Cost

Per car: six engine sources, a skid source, a one-shot host, and a whistle or inverter source for the turbo, diesel and electric voices. That is 7 to 8 spatial voices, so 8 cars use about 60 of the mixer's 128. The per-frame script is about 0.04 ms per car, around 0.3 ms for eight. The clips are 32 kHz mono PCM loops, 256 KB resident each.

Clips

The engine loops, turbo whistle, inverter, shift sounds and pops are synthesized. The crash crunches, blow-off and the gravel and soft skid takes were made with the engine's audio generation, and the skid takes were cut into seamless loops. The tarmac squeal comes from soundstage's generated skid take, copied in so this package stands alone. They are all in clips/.

Known limits

• #20654 — a car held at full throttle with assists off can spin its wheels without limit; the voice follows the engine's rpm, so that burnout sits on the limiter chop.
• #20569 — the skid sound follows car.surfaceId, which torqueCars works out by ray-casting under its wheels because the wheels do not read the ground's named surfaces yet; a car that writes no car.surfaceId gets the tarmac squeal everywhere.
• Doppler and distance — the pass-by pitch drop is the Doppler shift on the voice's spatial sources (doppler, 1 = physical, 0 = off); voices beyond maxDistance are silent, and with 8 cars about 60 of the mixer's 128 voices are used, so very large fields will steal voices.
• #21241 / #21242 — nothing of the voice is synced: each peer builds the voice from the car's replicated car.* values, so a remote car sounds as its owner drives it.
• #20946 — heard and measured on the native desktop engine only; the web build has not been run.