Basilica Audio

Apotheosis — user manual

The final ascension — a lookahead true-peak brickwall limiter for the master.

What it is

Apotheosis is a lookahead, oversampled true-peak brickwall limiter for the master bus. It is the last processor before export/streaming, not a general-purpose compressor: its one job is to guarantee the output's true (inter-sample, reconstructed continuous-time) peak never exceeds a Ceiling you set, while doing as little else to the sound as possible - or, if you dial in some Clip Mix, giving you a second, more aggressive "clipper" character on top of that guarantee.

Where it sits in a heavy production chain

Apotheosis belongs at the very end of the master bus, after every other processor has done its job:

Mix bus -> EQ / bus compression / saturation -> Apotheosis (true-peak limiter) -> export / streaming platform

It is not meant for individual tracks (guitars, drums, vocals) - use it once, on the master, as the final safety net and loudness stage before bouncing. Because it works in the true-peak (oversampled) domain rather than just the sample domain, it also protects you against inter-sample overshoot that a plain sample-peak limiter would miss - the kind of overshoot that shows up as clipping/distortion after a track is transcoded to a lossy format (MP3/AAC) for streaming, even though the original file looked "safe" on a normal peak meter.

Signal flow

Input -> Input Gain -> [4x oversampled] true-peak detect (Stereo Link-weighted) -> lookahead min-gain envelope
                                          -> Attack classifier -> Release (curve-shaped, Auto Release-modulated)
                                                                          |
     Output <-- Dither (Flat/Shaped) <-- ceiling clamp <-- Clip Mix blend <-- apply gain to lookahead-delayed signal <-+

Input Gain, true-peak detection, the lookahead gain-reduction envelope, the Attack classifier, the Release curve, the Clip Mix blend, and the final ceiling clamp all happen inside the same 4x-oversampled domain, before downsampling back to your project's sample rate. This is what makes the never-exceed-ceiling guarantee hold structurally rather than by inference: the limiter never detects a peak at high resolution and then tries to fix it after throwing that resolution away. See docs/architecture.md for the full engineering breakdown, including the latency model and the internal headroom-margin rationale.

v0.2.0 adds four new controls - Attack, Auto Release, Stereo Link, Dither Shape - described in the parameter table below. All four are research-derived: sourced from a competitor's own published help documentation (FabFilter Pro-L 2), a competitor's own published manual page (iZotope Ozone Maximizer), general audio-DSP dithering literature, and the ITU-R BS.1770 true-peak specification's documented behaviour, cited here for the underlying design principles and terminology only (the two-stage attack/release architecture, program-dependent release, noise-shaped dither) - not implying endorsement, sponsorship, affiliation, or DSP equivalence, and not measured against, benchmarked against, or reverse-engineered from any competitor's actual binary/DSP. See docs/design-brief.md and docs/research-notes.md for the full sourcing. Every new control's default reproduces v1's exact prior behaviour bit-for-bit - none of them change what Apotheosis sounds like unless you deliberately move them.

Parameter reference

Parameter Range Default Unit What it does
Input Gain -12 – +24 0 dB Trim applied before true-peak detection - "how hard are you hitting the ceiling". Raise it to drive the limiter harder (more gain reduction, louder/more compressed result); lower it if the incoming mix is already close to the Ceiling and you only want Apotheosis as a transparent safety net.
Ceiling -12 – 0 -1.0 dBTP The never-exceed true-peak target. The output's true (inter-sample) peak will not exceed this value, regardless of any other setting. -1.0 dBTP is a conventional mastering safety margin that leaves room for downstream lossy-encoding overshoot; use -1.0 to -2.0 dBTP for most streaming-platform targets, or push closer to 0 dBTP only if you control the final delivery format and know it won't re-encode.
Release 5 – 1000 50 ms How quickly gain reduction relaxes back towards unity once the programme material no longer requires it. There is no separate Attack control - the attack is always instantaneous and click-free, made possible by the Lookahead delay (see below), not by a time constant. Faster Release (short values) tracks transients more closely and can sound punchier but riskier (more audible pumping on sustained material); slower Release smooths gain reduction out at the cost of holding it longer after a peak.
Lookahead 0.1 – 20 5 ms How far into the future the limiter can "see" an oncoming true peak before it reaches the output - this is the mechanism that makes the instantaneous, non-clipping attack possible at all. This is a "setup" parameter, not a live-automatable one: it directly sizes real-time buffers and changes the plugin's reported latency, so a change only takes effect the next time your host re-initialises the plugin (sample-rate change, transport stop/start in most hosts, or reopening the project) - not instantly while audio is running. Larger values catch faster/steeper transients more reliably at the cost of more reported latency (which your DAW compensates for automatically on every track, so it is not something you need to manually correct for).
Release Curve Exponential / Linear / Smooth Exponential – Shapes the release (recovering) phase only - attack is always instantaneous regardless of this choice. Exponential is the classic one-pole ramp: fast initial recovery that tapers off, generally the most transparent and "musical" default. Linear recovers at a constant rate instead of tapering, which can sound more mechanical/obvious but is very predictable. Smooth is a two-stage cascade that gives a softer, overshoot-free onset to the release - useful if Exponential's initial recovery speed is causing audible pumping on sustained material - at the cost of an overall slower perceived release for the same Release time. Switching Release Curve is a discrete change (like a stompbox toggle), not a smoothly-automatable control.
Clip Mix 0 – 100 0 % Blends the transparent gain-reduction limiter path (0%, default) with an alternate tanh soft-clip "clipper" character (100%) applied directly to the signal rather than via gain reduction. At 0% Apotheosis behaves exactly like a pure lookahead limiter. Raising Clip Mix adds an increasingly present, more aggressive/saturated top end to anything hitting the ceiling - a common modern loudness-maximiser technique that trades some transparency for extra perceived loudness and a more "glued"/aggressive character, appropriate to heavy music's dense, high-energy masters. Every blend still passes through the same final hard ceiling clamp, so the never-exceed-ceiling guarantee holds at any Clip Mix setting.
Dither Off / 16-bit / 24-bit Off – Adds TPDF (triangular-probability-density-function) dither noise at the very end of the chain, at the output word length - standard practice when your final export/bit-depth-reduction target is 16-bit or 24-bit, since it decorrelates quantization error into noise rather than harmonic distortion. Leave it Off if a downstream stage (e.g. your DAW's own dithered bounce, or a 32-bit float delivery pipeline) already handles dithering - stacking dither doesn't help and unnecessarily raises the noise floor twice. Its amplitude is tiny (at most 1 LSB at the chosen bit depth, roughly -90 dBFS for 16-bit and -138 dBFS for 24-bit) and does not meaningfully affect the true-peak ceiling guarantee.
Attack (v0.2.0) 0 – 50 0 ms A transient/sustain classifier, not a gain-reduction ramp - the attack itself is always instantaneous regardless of this setting (see Release, above). Attack sets the duration below which a gain-reduction event is treated as a short transient, whose recovery then happens near-instantly (independent of Release/Release Curve/Auto Release) instead of following the normal Release-governed path. At 0 ms (the default) every event is treated as sustained and Apotheosis behaves exactly as v1 did. Raise Attack to let brief transient peaks recover their gain faster than sustained, dense material - the reference class's documented "short lookahead + long attack + fast release" recipe for maximising perceived loudness while preserving punch (see the Punchy Master factory preset). Research-derived from a competitor's documented two-stage transient/sustain architecture - the 0-50 ms range and 0 ms default are this project's own reasoned choices (no competitor default numeric value was recoverable from the sources fetched), not copied from any spec sheet.
Auto Release (v0.2.0) 0 – 100 0 % Blends in a program-dependent modulation of the effective Release time: a slow (multi-second), running average of how deep/sustained recent gain reduction has been biases the effective Release time longer on dense, sustained material (avoiding pumping) and shorter on sparse, isolated peaks (recovering quickly between them) - layered on top of (not replacing) whichever Release Curve you have selected. At 0% (the default) this is a complete no-op and Release behaves exactly as v1's. This is a from-scratch, reasoned implementation of the qualitative principle documented in the reference class's program-dependent-release lineage - it is explicitly not a reverse-engineered copy of any vendor's specific proprietary algorithm, whose internals are not publicly documented. Shipped off by default because no sourced default intensity exists for this control in any reference product; try it on dense, high-energy masters (see the Dense/Loud Modern and Adaptive Riding factory presets) if you want Apotheosis to ride sustained passages more gently without also slowing down its response to isolated peaks.
Stereo Link (v0.2.0) 0 – 100 100 % How tightly the two channels' true-peak detection is linked. At 100% (the default) both channels share a single, max-linked detector - v1's only behaviour, and the safest choice for most mastering work, since it never lets the stereo image shift under gain reduction. Lowering it lets each channel detect and limit progressively more independently, down to fully independent detection at 0% - useful when a hard-panned peak in one channel would otherwise pull the opposite channel's gain down with it (see the Wide Image Preserve factory preset). Mono and dual-mono host layouts are unaffected (there is no second channel to link against).
Dither Shape (v0.2.0) Flat / Shaped Flat – Only has an audible effect when Dither (above) is not Off. Flat (the default) is v1's existing plain TPDF dither, unchanged. Shaped runs the dither noise through a simple, fixed noise-shaping filter that pushes quantisation-noise energy toward the top of the audible band, where hearing is less sensitive - a general dithering-literature technique, not a copy of any vendor's specific shaping curve (the reference class documents a four-way Noise Shaping choice; this is a deliberately simpler two-way one). Dither's own guidance applies here too: this is genuinely one of the least consequential decisions in mastering, bounded to the least-significant bits either way - reach for Clean Export (Dithered) as a starting point when Shaped is worth trying (a final, fixed-bit-depth bounce).

Metering (engine-side; GUI display is a later milestone)

Apotheosis's DSP engine continuously computes and exposes (via the processor's getGainReductionDb(), getOutputTruePeakDb(), getMomentaryLufs(), getShortTermLufs(), and getIntegratedLufs() accessors) the current gain reduction, the output's true peak, and Momentary (400 ms)/Short-Term (3 s)/Integrated (session-running, absolute-gated) LUFS loudness readings. A visual meter surfacing these values in the plugin's UI is planned for the custom-GUI milestone (M3); until then, this data is available to any host or test harness that queries the processor directly. See docs/architecture.md for the K-weighting/gating implementation notes and its documented deviations from the full ITU-R BS.1770-4 two-pass algorithm.

Presets

A preset bar sits at the top of the editor ([<] [PresetName] [>] [Save] [Save As...] [Delete] [Import...] [Export...]). Apotheosis ships with eight factory presets covering the v1-compatible default plus starting points for each new v0.2.0 control - see docs/presets.md for the full list and intent of each. Your own presets save to ~/Library/Audio/Presets/Yves Vogl/Apotheosis/ on macOS (%APPDATA%/Yves Vogl/Apotheosis/Presets/ on Windows); the preset menu's "Set current as default" makes any preset (factory or your own) the one that loads automatically the next time you open a fresh instance. Import/export works with single .basilicapreset files and .zip preset banks.

Language

The preset bar's own interface text (button labels, menu items, dialog messages) automatically appears in German if your system language is German; otherwise it stays in English. There is no manual language switch yet. Parameter names and units (Attack, Release, dB, %, ...) always stay in English regardless of system language, since they are technical terms, not translatable UI copy.

Tips