Miserere — user manual (v0.3.0)

Four voices, one prayer — the parallel vocal template in a single unit.

What Miserere is

Miserere packages the documented 2010–2023-era parallel vocal template — the "rough vocal" workflow popularized in public interviews by mixers such as Andrew Scheps — into one plugin: a Direct path plus four parallel return busses (CRUSH, SANDWICH, SPREAD, SLAP), each with its own return fader, Mute and Audition. This is a documented, publicly sourced technique from that era (see research-notes.md), not an endorsement by or association with any named person or brand.

The core idea, and v2's correction over v0.1.0: the Direct path is a wire. Out of the box, every optional section on it is OFF, so the dry vocal passes through essentially untouched — its natural envelope and phrasing survive. Everything else is layered underneath it via the four return busses, which are copies of the direct-path output at unity, processed hard, and blended back in at a modest level. "Even with all that stuff in the mix, you'd probably think the vocal is bone dry" is the calibration target.

Signal flow

in → [In Trim] → DIRECT PATH (serial; every section optional, ALL OFF by default:
                   De-Esser (pre) → FET Comp light → Console EQ → Sat → De-Esser (post))
        │ = "the channel". Output feeds the sum at unity AND all four sends (unity taps):
        ├─→ ① CRUSH    : FET limiter, all-buttons character        → return fader
        ├─→ ② SANDWICH : Passive EQ → Opto Leveler → Passive EQ    → return fader
        ├─→ ③ SPREAD   : dual micro-pitch (≈30/50 ms, ±cents, L/R) → return fader
        └─→ ④ SLAP     : ≈110 ms dark single-repeat delay          → return fader
   Σ (direct + returns) → [Parallel macro trim scales returns ①–④] → [Out Trim] → out

Busses ①/② are minimum-phase and add zero latency, so they stay sample-aligned with the direct path — parallel summing never combs regardless of settings. Busses ③/④ are delays by design (see architecture.md). See research-notes.md for the sourced findings behind every default below.

The Direct path

Off by default, section by section, in signal order:

The four return busses

① CRUSH — FET limiter, all-buttons character

No threshold knob: Input drives the signal into a fixed per-ratio threshold/knee table. Ratio selects 4:1/8:1/12:1/20:1/ALL (ALL is a plateau-shaped curve with a deliberate give-back and a short attack lag that lets transients punch through before clamping — the "snap"). Attack/Release are 1–7 dials where a HIGHER number is FASTER, matching the hardware convention this is modelled on; release is dual-rate and program-dependent (fast after brief transients, several times slower after sustained heavy compression). Style switches between All-Buttons and a softer, fixed 2:1 Gentle voicing. This bus is meant to sound "terrible" soloed (use Audition) and good blended in.

② SANDWICH — Passive EQ → Opto Leveler → Passive EQ

Two independent Passive EQ instances bracket an opto-style leveler. Each Passive EQ offers a shared-frequency LF Boost and Cut (both can run at once — a deliberately non-cancelling curve, not a simple sum to flat), an HF Bell Boost with variable bandwidth, and an HF Shelf Atten. The Opto Leveler has no threshold: Peak Reduction drives into a fixed static curve (soft ~3:1 below −20 dB, hard ceiling above; Limit tightens the soft region toward ~10:1), with a raw-audio detector (no smoothing ahead of the ballistics) and a two-stage release whose tail lengthens the longer it has been working. Emphasis makes the detector progressively HF-selective (up to −10 dB less LF sensitivity), so at high settings it reacts mostly to sibilance/presence, "like a multiband". Residual (default on) keeps the Passive EQ's small, never-fully-flat vintage tilt; defeat it for a cleaner EQ.

③ SPREAD — dual micro-pitch

Two short delay taps (~30 ms pitched up, ~50 ms pitched down), hard-panned L/R. Detune sets the pitch offset in cents (default 6 — deliberately small, so the ear reads "pushed to the outside" rather than chorusing). Time scales both base delays together; Width blends from a fully centred sum (0%) to the full hard pan (100%).

④ SLAP — single-repeat dark delay

Time (50–160 ms, default 110 ms, plain milliseconds — deliberately not tempo-synced). Feedback is fixed at 0 in v2: there is exactly one repeat, and its darkness comes from a built-in bucket-brigade-style voicing (Tone sweeps a progressive HF loss plus soft saturation baked into that single repeat) rather than a filtered feedback loop. Stereo switches from the default mono return (the classic mono slap behind a stereo-widened vocal) to independent L/R delays.

Fader logic

Presets

A preset bar sits at the top of the editor: [<] [PresetName*] [>] [Save] [Save As...] [Delete] [Import...] [Export...]. Clicking the preset name opens a Factory/User menu; a trailing * means the current preset has unsaved changes. Ten factory presets ship in the box (see presets.md for what each one is for); user presets save to ~/Library/Audio/Presets/Yves Vogl/Miserere/ on macOS (%APPDATA%/Yves Vogl/Miserere/Presets/ on Windows). The preset menu's "Set current as default" makes any preset — factory or user — load automatically on every fresh instance; "Import..." accepts both single preset files and zip preset banks.

Starter recipe

  1. Leave the Direct path off, or add De-Ess Pre / a touch of Console EQ if the source needs it. Keep FET Comp and Sat off unless the vocal specifically needs light insert compression.
  2. CRUSH starts at −9 dB by default with the ALL-Buttons character already engaged — bring Input up until Audition shows heavy, "disaster in solo" compression, then trust the default fader level and adjust by ear from there.
  3. SANDWICH starts at −12 dB; raise Peak Reduction until the vocal thickens without audibly pumping in context.
  4. SPREAD and SLAP (−18 dB / −15 dB by default) should each pass the "you only notice it's gone when you mute it" test — if either is audible as a discrete effect, pull it back.
  5. Use Parallel to back the whole layer off quickly on quieter/more organic material.

Known limitations (v0.3.0)