Vehicle audio -- engine loops, RPM crossfade, slip SFX

The method for vehicle engine audio that does not screech: two-layer RPM crossfade with narrow pitch bands, code-side looping, a slip-driven skid feedback loop, and the persistent ConVar trap.

Core facts (engine behavior)

• SoundEvent has no loop flag. A looping sound is held code-side: keep the SoundHandle, update handle.Position every frame to follow the emitter, and re-trigger (Sound.Play(...)) the moment !handle.IsPlaying. A longer source sample makes the re-trigger seam rarer.
• Sound.Play needs the full resource path INCLUDING the .sound extension -- Sound.Play("sounds/engine_idle") silently fails; Sound.Play("sounds/engine_idle.sound") works.
• Pipeline: .mp3 source + a minimal .sound event resource next to it; the editor's asset watcher compiles both (.vsnd_c/.sound_c, gitignored). 128 kbps mono is fine for a positional game loop.
• Free recordings: freesound CC0 preview MP3s download without OAuth. License hygiene matters in a public repo -- CC0 > CC-BY (+ATTRIBUTION.md) > everything else; reject NC/ND and "no redistribution" packs.

Engine audio that doesn't screech (the two-layer recipe)

Never pitch one loop across the whole RPM range. Mapping idle-to-redline onto pitch 0.8-2.0 doubles frequency at redline -- chipmunk, instantly rejected. Narrowing to 0.7-1.3 still sounds synthetic. One sample stretched is one sample stretched.

The recipe:

1. Two REAL recorded layers -- a deep idle/low-RPM loop and a rev/high-RPM loop, ideally from the same recording session so timbre matches across the blend.
2. Equal-power crossfade by normalized RPM t: low layer fades 1->0 across t=[0.0, 0.7] (smoothstep), high layer 0->1 across t=[0.3, 1.0], then normalize so lowGain^2 + highGain^2 = 1 -- combined loudness stays steady through the overlap.
3. Each layer pitch-shifts only a NARROW band around its recorded pitch (low 0.9-1.25, high 0.85-1.2). Because each layer's extreme pitch falls where that layer is nearly silent, nothing is ever heard far from how it was recorded. This constraint is the entire anti-screech mechanism.
4. Per-class data on the definition: EngineSoundEvent (+ optional EngineSoundEventHigh; absent = legacy single-loop path) and EnginePitchBase multiplying the band (truck 0.85 sits deep, kart 1.2 keeps its whine). One code path serves a whole roster.
5. Volume rides throttle from an idle floor (~0.55, so a coasting vehicle stays audible) to full.

Slip-driven skid loop (traction-loss feedback)

Per-vehicle (not per-wheel) is fine for a first pass. Intensity = max over grounded wheels of max(|SlipRatio|, |SlipAngle| / 0.7rad) -- catches wheelspin/lockup AND lateral slides/handbrake.

Silent below ~0.40; volume 0->0.9 and pitch 0.90->1.15 ramp to ~1.10; handle.Stop(0.12f) fade on release. Same code-side loop pattern as the engine sound.

The trap that ate two listen tests

A dev-override ConVar left set on a machine persists across sessions and code reworks and silently forces the old loop on every vehicle -- code green, compile green, feature inaudible.

Warn loudly at component start whenever a non-default override is active, and grep the console for the resolved-event log line BEFORE regenerating assets when a listen test reports "nothing changed."