5.2 KiB
reverb
(no submodules)
Purpose
ReverbPlugin is the universal reverb effect: a Schroeder-style algorithmic engine (four parallel comb filters plus two series all-pass filters) that doubles as a direct-convolution renderer when an impulse response is loaded. The same plugin handles short room ambience, large hall colour, plate-style metallic shimmer, and full cathedral tails — preset by name or sculpted by adjusting preDelay and decay. Pre-delay is applied to the wet path only so the dry signal stays sample-aligned with the source, and wet === 0 collapses to bit-transparency.
The plugin satisfies the spec's "impulse-response convolver with algorithmic fallback" requirement: once loadImpulseResponse(url) succeeds, every block runs direct convolution; before any IR is loaded (or after applyPreset clears one), the algorithmic engine renders the tail.
Features
- Four automatable parameters —
preDelay(ms),decay(s, mapped to per-comb feedback via0.001^(τ/T)),wet(0..1),dry(0..1). - Allocation-free per-block processing once
setup()has allocated comb / all-pass / pre-delay / convolution delay lines. - Bounded feedback (clamped to
0.997) so the algorithmic path cannot diverge regardless ofdecayvalue. wet === 0early-return path: output equalsdry * inputexactly, bit-transparent whendry === 1.loadImpulseResponse(url, { signal })— fetches the IR, switches the reverb into convolution mode. A pre-abortedAbortSignalshort-circuits withAUDIO_FX.ABORTEDbefore any fetch is issued.- Four shipped factory presets —
room,hall,plate,cathedral— applied viaapplyPresetByName. - Round-trips the loaded IR through the universal preset envelope as a plain
number[]so a saved session captures the cathedral.
Usage: Basic Example
import { ReverbPlugin } from "@suaveplan/audio-fx-plugin-reverb";
const verb = new ReverbPlugin({
initialWet: 0.25,
initialDry: 0.75,
});
verb.applyPresetByName("hall");
chain.add(verb);
Usage: Advanced Example
import { ReverbPlugin } from "@suaveplan/audio-fx-plugin-reverb";
const verb = new ReverbPlugin({ initialWet: 0.6, initialDry: 0.4 });
await verb.setup({ sampleRate: 48_000, channels: 2, blockSize: 512 }, ctx);
// Load a cathedral IR from a CDN, but let the user cancel the load
// if they switch presets before it finishes.
const ac = new AbortController();
cancelButton.addEventListener("click", () => ac.abort());
try {
await verb.loadImpulseResponse("https://cdn.example/impulses/cathedral.f32", {
signal: ac.signal,
});
console.log("Cathedral loaded:", verb.getImpulseResponse()?.length, "samples");
} catch (error) {
if ((error as AudioFxError).code === "AUDIO_FX.ABORTED") {
// User cancelled — algorithmic fallback continues to render.
} else {
throw error;
}
}
// Persist with the session: the IR rides along inside the preset envelope.
const snapshot = verb.getPreset();
saveSession(JSON.stringify(snapshot));
API Reference
ReverbPlugin(class extendsAbstractAudioFxPlugin)new ReverbPlugin(options?: ReverbPluginOptions)— throwsAUDIO_FX.INVALID_OPTIONSon malformed options.loadImpulseResponse(url: string, opts?: { signal?: AbortSignal }): Promise<void>— throwsAUDIO_FX.ABORTEDon pre-aborted signal.applyPresetByName(name: ReverbPresetName): void— appliesroom/hall/plate/cathedral.getImpulseResponse(): Float32Array | undefined— current IR or undefined when algorithmic.- Inherited parameter accessors for
preDelay,decay,wet,dry.
REVERB_PRESETS: Readonly<Record<ReverbPresetName, { preDelayMs, decaySec }>>— exported preset table.
Implementation Notes
The Schroeder-Moorer comb/all-pass topology was chosen because every state buffer is short (≈30–45 ms × sampleRate), feedback is closed-form derivable from RT60 (feedback = 10^(-3 · τ / T), clamped to 0.997), and the implementation has zero per-block allocation after setup. Comb buffers and all-pass buffers are per-channel: stereo material gets four combs and two all-pass filters per channel, with the same delay times — engine-level decorrelation (different delays per channel) is a future enhancement, not a current blocker.
Pre-delay is implemented as a single circular buffer per channel sized to the maximum allowed (200 ms at the configured sample rate). The current pre-delay length is read per-block from the preDelay parameter so automation lanes stay live; the buffer itself is never reallocated. When preDelayMs === 0 the read short-circuits (input passes straight through to the wet engine) so the bit-transparent path remains tight.
The convolution path is direct (time-domain) — appropriate for short room and plate IRs (≲ a few thousand samples). Long cathedral IRs would benefit from a partitioned FFT convolver; that is deferred until a real bottleneck demonstrates the need, since the algorithmic path covers long tails efficiently. The convolution delay line is reallocated whenever an IR is loaded or applied via preset, so per-block cost stays linear in the IR length without per-block allocation.