Getaway Driver is an automatic post-phrase throw unit built by Sound Detective / Detective 47. The classic throw — a word or a phrase ending hurled into a delay or reverb, blooming in the gap before the next line — traditionally means riding a send fader by hand, all night, on every pass. This unit does the riding for you. It listens to the source, keeps the wet path gated shut while the phrase plays, and when a genuine rest is confirmed, the throw blooms into the gap. The moment the phrase returns, it ducks out of the way.
The brain is a RestDetector: a threshold decides what counts as quiet, a hold time decides what counts as a real rest rather than a breath, and an arm latch requires the source to have actually played since the last throw — so a silent track never ghost-fires. You will not hear a throw the unit did not earn.
Under the hood: seven delay machines (six from DD-FX-Core, level-matched to each other), five reverb types, a dedicated five-voicing octave-shimmer layer that feeds back into the reverb tank, reverse / gated / swell verb effects, ping-pong and motion stereo modes, a Dust 12.47-lineage CHARACTER rack on the wet path only, and an always-safe wet path standing guard at the exit.
ESCAPE is the signature macro. One knob that scales throw level, reverb size, feedback, stereo width and motion together — staged so it moves from subtle ambience to cinematic without ever just getting louder. The width term is squared. It arrives late, on purpose.
Getaway Driver is the back-of-phrase companion to Backtrace's front-of-phrase reverse swells. Backtrace asks how the line began. This unit handles where it went. It is a getaway car idling at the curb — it only moves when the phrase stops talking.
Understanding the order of operations is the difference between using Getaway Driver as a delay pedal with a gate on it and using it as an instrument. Every stage was placed deliberately.
THE DRY LANE. The dry signal never touches a processor. It runs straight to the final sum, where MIX blends it against the wet. Whatever the wet path does to itself, your source walks out clean.
THE DETECTOR. The brain sits in parallel, not in the signal path. An envelope follower tracks the input against the threshold, HOLD confirms the rest, and the detector produces two control signals: throwEnv, which blooms into the gap, and duck, which follows the source's presence. Everything downstream obeys these two.
THE DELAY LINE. The line is always fed — the end of the phrase is already sitting in it when the gate opens. While the source plays, the loop drains, so every throw is built from the phrase that just ended, not from a chorus trapped a minute ago. Above roughly 72% FEEDBACK, an external dub-regeneration tap joins the loop.
THE VERB FEED. The reverb does not just hear the delay. It hears a smoothed handoff of the repeats that rises as they decay, a taste of detector-gated dry, the REVERSE replay when that FX mode is armed, and the shimmer layer's INTO VERB return. The tank charges while the gate is shut — that is why the bloom is there the instant the rest is confirmed.
THE WET CHAIN. Delay and verb are summed per ROUTING, then pass through CHARACTER, the master LOW CUT, the M/S width stage (which grows with the square of ESCAPE), THROW level, and finally the throw gate itself — throwEnv times duck, shaped by ATTACK and RELEASE — before the always-safe wet path clears it for release.
THE EXIT. Dry and wet sum per MIX, the OUTPUT trim applies, and a soft ceiling knee — bit-transparent below −0.2 dBFS — stands at the door.
The brain. A fast envelope follower with 2.5 dB of hysteresis watches the input. When the level drops below the threshold, HOLD starts counting. Only when the source stays quiet for the whole hold time does the throw fire. An arm latch requires the source to have actually played since the last throw — silent tracks never ghost-fire.
The scope draws the input level, the threshold line (drag it), and the throw bloom. Watch it for one verse and you will know everything the detector knows.
With AUTO engaged, the detector ignores the fixed threshold and rides approximately 14 dB below the source's recent level. The loud tracker rises fast and falls at 3 dB per second, so the line follows the performance without chasing every syllable. It never sinks within 6 dB of the quiet bed — room tone cannot arm a ghost throw. The effective line is clamped between -55 and -8 dB.
While AUTO is on, the THRESHOLD knob is dimmed and inert. Drag the threshold line on the scope and the unit performs a manual takeover: AUTO switches off and the line is yours.
The level line the detector watches. Below it, a rest begins. Above it, the source re-arms the next throw. Only active with AUTO off.
Set it two ways: turn the knob, or drag the line directly on the scope. Put it where the phrase lives above the line and the gaps live below it.
How long the source must stay below the line before the throw is confirmed. This is the difference between a breath and a rest. Raise HOLD to ignore syllable gaps and breaths. Lower it to catch quick rhythmic rests.
HOLD also anchors the ping-pong delay: the flip grid is aligned to the hold time plus roughly 45 ms so repeats land on-side. See FILE 06.
One-pole rise time on the wet envelope after a rest is confirmed. The 110 ms default is deliberate: the reverb has been accumulating the whole time the source played, and a fast attack pops the entire stored bloom in at once. The default eases the door open instead.
Shorten it well below the default for tight rhythmic gap-fills — best done in Delay-only routing, where there is no stored bloom to pop. See FILE 09.
One-pole fall time on every wet close — both the duck engaging when the phrase returns and the tail ending. This is why the unit feels like an automated fader, not a hard gate. Shorten it and the throw gets out of the way fast; lengthen it and the tail lingers under the next line.
How hard the wet is pushed down whenever the source is present. The arithmetic: duckGain = 1 − duck × presence, where presence attacks in about 12 ms and releases over about 180 ms. High settings clear the wet almost completely under the phrase; at 0% the throw stays put and the phrase talks over it.
The shimmer layer ducks 1.35× harder than the rest of the wet. The octave air is the first thing out of the room when the phrase walks back in.
The throw's life span. The envelope decays exponentially, reaching -60 dB after 0.4 × 20^tail seconds — a base range of 0.4 to 8 seconds.
Three things stretch it further: ESCAPE (up to ×1.8), FEEDBACK's cubic dub zone (up to ×4 at 95%), and SIZE when the verb is in the path (up to ×2). The documented ceiling sits around 14 seconds and beyond. TAIL hands down the sentence. The other three file appeals.
The one big amber knob in the centre of the panel. ESCAPE never simply raises volume — every stage is scaled together, and the stereo term is squared so width arrives late. The panel hints read SUBTLE → CINEMATIC. They mean it.
One knob that scales throw level, reverb size, feedback, stereo width and motion together, staged so the unit moves from subtle ambience to cinematic without ever just getting louder. At ESCAPE value e (0 to 1), the stages move as follows.
| STAGE | WHAT ESCAPE DOES |
|---|---|
| Feedback | Effective feedback = min(0.90, fb × 0.90 + 0.12e). Up to +12% loop gain, always capped at 0.90. |
| Reverb size | +0.25e — up to a quarter more SIZE at full tilt. |
| Tail life | ×(1 + 0.8e²) — up to ×1.8 longer throws. |
| Delay output | ×(0.85 + 0.15e). |
| Verb output | ×(1.15 + 0.85e) — the verb gains ground as ESCAPE rises. |
| Throw level | +3 dB × e. |
| Stereo width | M/S side amount 0.3 + 1.2e². The term is squared — width blooms only in the top half of the knob. |
| Motion | A whisper of delay-time movement. |
Automate ESCAPE per song section instead of touching five knobs. The bottom half stays sober; the top half blooms wide. See FILE 09.
Three knobs and a switch decide how the getaway reaches the street.
Delay — repeats only. The verb is muted.
Reverb — the tail is reverb alone, fed by a detector-gated dry send. That send is presence-ducked to about 25% while the source plays, so the verb output carries a fixed +8 dB (2.5×) makeup to compensate. Balance this mode with VERB MIX or THROW, not OUTPUT.
Delay > Verb — the classic throw. The verb hears a smoothed handoff of the repeats, rising as they decay, plus a taste of the gated dry.
All sends glide, so mode changes never click. Switch mid-performance if the song asks for it.
Level of the triggered throw itself, plus the 3 dB × ESCAPE bonus on top. It never touches the dry signal. This is the effect-level control — reach for it before you reach for OUTPUT.
Send-style global balance. At 50%, dry and wet both run at unity. Below 50% trims the wet. Above 50% trims the dry. At 100%, wet only.
On an insert, leave MIX at 50% and set effect level with THROW. On an FX bus, set MIX to 100% and drive the unit from a send. See FILE 09.
Final smoothed trim after everything, followed by a soft ceiling knee that is bit-transparent below −0.2 dBFS. Below the knee, the unit does not touch a single bit.
The delay line is always fed, so the end of the phrase is already sitting in it when the gate opens. While the source plays, the loop drains — recirculation is pulled to about 25% — so every throw blooms from the phrase that just ended. The line never testifies about evidence trapped from a minute ago.
Clean Getaway is the house throw engine — a tanh-conditioned loop with TONE-driven damping, built for this unit. The other six are the Backtrace lineup from DD-FX-Core, each output-trimmed to load at the same perceived level. Switching machines is a character decision, not a level decision.
Type changes crossfade equal-power over about 60 ms, with the incoming machine reset clean. No clicks. No leftover evidence.
| MACHINE | CHARACTER |
|---|---|
| Clean Getaway | House throw engine. tanh-conditioned loop, TONE-driven damping. |
| Reel Echo | Tape multi-head. |
| Digital Pedal | Clean digital. |
| Magnetic Drum | Drum echo. |
| Tape Witness | Gritty slap. |
| Cold Rack | PCM studio. |
| Vault Delay | Dirty sampler. |
Tempo-synced divisions computed from host BPM (clamped to a 20 BPM minimum; the resulting time clamped 20 to 4000 ms). The TIME knob is dual-mode: in sync it steps through the divisions, in Free it dials milliseconds. It is never greyed out.
Exact delay time when TIME is set to Free. Time changes slew tape-style over about 80 ms of smoothing, so moves never click — turn it while the throw rings and it bends instead of stuttering.
Repeat count. The loop is hard-capped at 0.90 gain with tanh inside it, so it cannot run away on its own.
Above roughly 72%, an external dub-regeneration tap joins the internal loop and the total gain can reach true unity. It is governed and self-sustain-bled — it sustains musically, but it can never park at the ceiling. This is the dub zone.
High feedback also cubes into TAIL life: 50% is ×1.4, 80% is ×2.8, 95% is ×4.0. When you want long, let FEEDBACK carry the length. See FILE 09.
Level of the delay section inside the throw. Pull it down when the verb should lead and the repeats should only suggest.
Dust 12.47-family high-pass on the delay branch only. Clears the repeats' low end without touching the verb.
Dust 12.47-family low-pass on the delay branch. Darken the repeats here for vintage throws that sit behind the vocal instead of beside it.
At and below 100%, the delay branch folds toward mono. Above 100%, a Haas spread on the right channel opens up to about 8 ms. 100% is exactly the pre-knob sound — the detent is honest.
Ping Pong — each repeat lands on the opposite side. The flip grid re-syncs to the repeat onsets at every throw, so words are never chopped in half.
Motion — a slow sine pan, roughly 4.5 seconds per cycle, drifts the repeats across the field.
Both modes force the M/S stage to preserve their panning. They read even at low ESCAPE.
Five flavors behind one SIZE knob, each with its own size-and-decay mapping so every type stays in its musical range. Type changes crossfade over about 60 ms from a clean tail. The shimmer layer is its own voice: it runs over every verb type, is fed from the wet tail — never the dry — and is throw-gated and extra-ducked so it lives and dies with the getaway.
Plate — dense studio plate. This is Backtrace's Phantom Plate tank. Shimmer — long octave-bloom air with a fixed pitch-in-feedback identity. Room — tight, clean chamber. Dark Verb — the same architecture voiced dark; it stays dark across the whole DAMPING range. Spring — dispersive tank with a softened transient exciter.
All output trims are level-matched to Plate. Audition types at the same loudness and pick with your ears, not the meter.
Reverse — a rolling record of the dry mono, about 2 seconds deep, is replayed backwards (about 0.7 seconds) into the verb at each throw. It rises out of nothing into a fast, clean cut — the classic reverse swell — and re-fires at every throw.
Gated — the tail is chopped dead when the throw envelope falls below 30%. Eighties style, no apologies.
Swell — the bloom fades up over about 0.6 seconds on a squared ramp into the rest.
Drives each type's own size-and-decay mapping, so every flavor stays inside its musical range. ESCAPE adds up to +25% on top.
When the verb is in the path, SIZE also stretches the throw envelope up to ×2 — big verbs are allowed to hang.
The gap before the reverb starts, so the bloom sits behind the phrase ending instead of on top of it. Floored at 0.5 ms and clamped per type: Plate up to 118 ms, Shimmer up to 148 ms, Spring up to 78 ms.
How fast the verb's top end fades, through per-type mappings that preserve each type's identity. Room sweeps 3.3 to 14.8 kHz. Dark Verb sweeps 1.4 to 4.8 kHz — dark at every setting, as filed.
Level of the reverb section inside the throw. In Reverb-only routing this path carries the ×2.5 makeup described in FILE 05.
High-pass on the reverb branch. Clears mud from the tail before it reaches the mix.
Low-pass on the reverb branch. Smooths the tail's top when the source is bright and the room should not be.
THE SHIMMER LAYER
A dedicated octave voice over any verb type, not a separate reverb. It taps the wet tail, never the dry, and its gate follows throwEnv^1.25 — it swells out of the tail rather than arriving ahead of it.
Amount of the octave layer over the current verb type. The signal path: wet-tail tap → 350 Hz high-pass / 9 kHz low-pass band-limit → dual-tap Hann grain shift up one octave → allpass diffusion cloud → tone tilt plus silk ceiling → soft density → bloom regeneration.
The shimmer layer's character. The order has been frozen since v0.7.0, and the slots map to internal variants so the DSP numbering never moves.
| VOICING | CHARACTER |
|---|---|
| Velvet | The dark pole. Soft 6.5 kHz silk ceiling, doubled grains, 8¢ detune. |
| Choir | 14¢ dual-voice detune beating in both ears, 3-tap grains. Stacked-voices feel. |
| Halo | An FDN tail after the shifter smears grain edges in time. Modulated diffuser. |
| Glass | Bright 3-tap read. Sparkle without the comb edge. |
| Air | Brightest. 16 kHz silk, shift fed from higher up. |
Brightness tilt of the shimmer cloud — a one-pole centered per voicing, reaching about ±1.1 octaves around roughly 4 to 10 kHz depending on the voicing. A fixed silk low-pass keeps even the brightest setting out of ice-pick territory.
Grain length from 50 to 130 ms (longer per voicing) plus diffusion density from 0.42 to 0.70. High settings make the pitch shift disappear into pure air. Low settings reintroduce granular flutter — the mechanism becomes audible, which is sometimes exactly the point.
Shimmer regeneration — feedback capped at 0.5, with in-loop damping at 5.2 kHz and 300 Hz. The cloud re-shifts up an octave on every pass and stacks heavenward, dying with the throw. It cannot outlive the getaway.
How much of the octave layer returns into the reverb tank (tanh-bounded) to be re-diffused, instead of being added in parallel. At 0, you get a discrete octave-up copy sitting on top of the verb — the chipmunk. High settings dissolve the octave into the tail as rising air — the classic shimmer topology.
The direct level compensates as (1 − 0.6 × send), so the knob stays level-neutral. Keep it high unless you want the copy on purpose.
Dust 12.47-lineage color, applied to the wet sum only. The dry signal never sees the inside of this room. The rack is deterministic — fixed-seed RNG — so renders repeat exactly: the same session bounces the same dust every time.
Scales WOW, SATURATE and AGE together. At 0, the wet path is surgically clean — bit-transparent, with saturation crossfaded in from zero. High settings are dusty and unstable, and still musical.
Sampler-style zero-order-hold wear, with the hold rate dropping as low as about 0.3× the sample rate, plus a dulling low-pass sweeping from 16 kHz down to 6.5 kHz, plus a colored hiss bed.
The hiss is envelope-gated to the wet signal. The texture lives inside the throw and dies with it — no hiss bed idling between phrases.
A modulated delay line — compound 0.31 Hz and 0.87 Hz LFOs plus a slow random walk — bends the whole tail's pitch and time by up to about 3.5 ms. The right channel runs at 0.83× depth for stereo life. Smooth unless deliberately pushed.
tanh density with unity small-signal gain, crossfaded in — thicker, warmer tails, never clipped-harsh. It also feeds the Clean Getaway delay's output drive upstream, so the house machine grows teeth as you raise it.
One-pole darken sweeping 2.2 to 16 kHz (log) on the whole wet path. It does double duty: it also sets the delay's in-loop damping, 3 to 11 kHz, so repeats age a little more on every pass.
Lower TONE and every repeat comes back darker than the last — tape-style decaying echoes from one knob. See FILE 09.
The Dust 12.47 high-pass — the same DD-FX-Core TPT filter, same range and skew as the 12.47 plugin — on the entire wet path. This is the CHARACTER master filter, the last spectral word before the width stage.
These are operational procedures developed through field use. They are not the only ways to run the unit. They are evidence of what it is capable of.
TECHNIQUE 01 — THE DUB GETAWAY
Push FEEDBACK past roughly 72% to enter the dub zone. The external regeneration tap joins the internal loop, repeats approach true unity, and TAIL life stretches cubically — 50% is ×1.4, 80% is ×2.8, 95% is ×4. Keep TAIL moderate and let FEEDBACK carry the length. The loop still drains whenever the source plays, so each throw is always built from the phrase that just ended — endless repeats, never an endless pile-up.
TECHNIQUE 02 — ESCAPE IS THE ARRANGEMENT
Automate ESCAPE per song section instead of touching five knobs. The stereo width term is squared, so 0 to 50% stays sober while 60 to 100% blooms wide — and on the way it adds up to +3 dB of throw, +12% feedback and +25% size. Staged, never just louder. Verse at 25%, chorus at 70%, final chorus at 100%. One lane of automation, a full arrangement move.
TECHNIQUE 03 — THE REVERSE WITNESS
Set VERB FX to Reverse. The unit replays the actual recorded ending of the phrase backwards into the verb at every throw — a rising ramp into a fast, clean cut. Pair it with Reverb-only ROUTING and a longer HOLD, so the replay anchors to a real phrase ending and not a breath.
TECHNIQUE 04 — THE PING PONG ALIBI
Set MOTION to Ping Pong with a synced TIME division. The flip grid re-syncs to the repeat onsets at every throw, so each repeat lands cleanly on the opposite side — no words chopped in half. Both motion modes force their panning to survive the width stage, so the movement reads even at low ESCAPE.
TECHNIQUE 05 — SHIMMER OVER ANY TANK
The shimmer layer is independent of verb TYPE. Plate with SHIMMER raised is the lead-vocal classic. The voicings run dark to bright roughly Velvet → Halo → Choir / Glass → Air. Keep INTO VERB high — the 70% default — for dissolving air. Pull it toward 0 only when you deliberately want the discrete octave-copy effect.
TECHNIQUE 06 — TAPE DECAY WITH ONE KNOB
CHARACTER's TONE does double duty: it darkens the whole wet path and sets the delay's in-loop damping, 3 to 11 kHz. Lower TONE and every repeat comes back darker than the last — tape-style decaying echoes without touching the delay HIGH CUT.
TECHNIQUE 07 — INSERT OR SEND, MIX DISCIPLINE
MIX is send-style: 50% means dry and wet both at unity. On an insert, leave MIX at 50% and use THROW for effect level. On an FX bus, set MIX to 100% — the dry trims out — and drive the unit from a send.
TECHNIQUE 08 — THE RHYTHMIC GAP-FILL
For tight rhythmic gap-fills, shorten ATTACK well below its 110 ms default and switch ROUTING to Delay only. The default is slow for a reason — the reverb accumulates during the phrase, and a fast attack pops the whole stored bloom in at once. Delay only has no stored bloom, so the fast attack lands as a clean stab of repeats in the gap.
Six situations, six findings. Most of them are the unit doing its job.
NO THROWS EVER FIRE — THE GETAWAY LAMP NEVER LIGHTS
The detector must arm: the source has to cross the threshold at least once since the last throw. An empty or too-quiet track can never fire, by design. Leave AUTO on — the default, with the line riding about 14 dB under the source automatically — or, with AUTO off, drag the scope line so it sits between the source's level and the rest bed. Watch the lamps: ACTIVE → HOLD → GETAWAY is the path. If it keeps bouncing between ACTIVE and HOLD, your HOLD time is longer than the actual rests.
THE THRESHOLD KNOB IS DIMMED AND DOES NOTHING
AUTO owns the line. The knob is deliberately inert — drawn at 35% alpha — while the auto-threshold rides. Toggle AUTO off, or simply drag the threshold line on the scope: the takeover switches AUTO off for you.
HITTING STOP FIRES A THROW
v0.7.1 and later reads the host transport: on the playing-to-stopped edge, the wet fades in about 60 ms, the racks are wiped silently, and the gate returns slowly — about 800 ms — over a bounded window of roughly 2.5 seconds. In a host (or in Standalone) that does not report transport, the unit assumes "playing", so stopping audio reads as a legitimate rest and throws. That is expected there, not a defect. Live monitoring with the transport parked works normally, because the guard window is bounded.
REVERB-ONLY ROUTING SOUNDS LOUDER THAN DELAY > VERB
Intentional. In Reverb-only mode the verb's dry send is presence-ducked to about 25% while the source plays, so the verb output carries a fixed +8 dB (2.5×) makeup. Balance with VERB MIX or THROW, not OUTPUT.
THE THROW ONLY CONTAINS THE END OF THE PHRASE
By design. The delay loop is drained while the source plays, and the verb's dry send is detector-gated, so the bloom is built from the last word — not a wash of everything that played. If you want more of the phrase in the throw, lengthen the delay TIME division or raise FEEDBACK. There is no "capture" control to look for.
RED "GUARD xN" OR "HOT xN" TEXT IN THE HEADER
The always-safe wet path caught and contained something (GUARD), or logged sustained wet loudness (HOT). Audio recovers by itself with a short clean fade — it is a diagnostic counter, not a crash. If the count keeps climbing in one session, note the settings and report them. The number is exactly the evidence needed.
| PARAMETER | SPECIFICATION |
|---|---|
| Formats | VST3, AU, Standalone (JUCE 8.0.7 · VST3 category Fx|Delay|Reverb · AU Effect) |
| Platform | macOS build in this repo (CMake + JUCE; cross-platform framework, only macOS verified here) |
| Channels | Mono→Mono, Stereo→Stereo, Mono→Stereo (mono output folds L+R) |
| Latency | 0 samples reported to the host (no latency declared); declared reverb tail 12 s |
| GUI Size | Fixed 1100 × 960 design canvas at 0.80 scale → 880 × 768 px window (not resizable) |
| Presets | None — single program; settings persist via full APVTS state save/recall in the session |
| Version | 0.7.6 (shown live on-panel, bottom-right) |
| Parameters | 41 host-automatable parameters, IDs frozen |
| Manufacturer | Sound Detective · codes SDet/Gtwy · com.sounddetective.getawaydriver |
| Engine | DD-FX-Core v0.1.4 (pinned tag) for delay machines, verb tanks and Dust 12.47 filters |
| MIDI | None (no input, no output, not a MIDI effect) |
Rest confirmed. Throw executed. Case file complete. The getaway is yours now.