@audio/convolution
Status calibrated
npm1.1.0licenseMIT
Convolution reverb — impulse-response convolution (rooms, plates, cabinets)
import convolve from '@audio/reverb-convolution'
Convolves the input against an impulse response with no latency: the IR's head (its first 128 samples) direct-form, the rest by FFT overlap-save with a frequency-domain delay line in partitions growing with the offset — 128, 1024, 8192 (Gardner 1995, Efficient convolution without input-output delay). Differentially tested against the direct form to 1e-6 of the peak, IRs about each partition's edge, chunks of any length. The wet tail is truncated at the input length (no automatic tail extension — pad data yourself if you need the reverb tail past the input's own end). Mono only. Speaker-cabinet simulation (@audio/amp-cabinet) shares this engine.
convolve(data, { ir: roomImpulseResponse })
convolve(data, { ir, normalize: true })
// streaming: chunks of any length, each back at once (wet only)
import { convolver } from '@audio/reverb-convolution'
let room = convolver(ir)
for (let chunk of input) play(room.write(chunk))
| Param | Default | |
|---|---|---|
ir |
(required) | Float32Array impulse response — throws RangeError if missing/empty |
mix |
1 |
0..1 — wet/dry blend |
normalize |
false |
scale the IR to unit energy first, so output level ≈ input level regardless of the IR's own loudness |
method |
'auto' |
'auto' or 'fft': the partitioned engine; 'direct': the plain direct form, its reference |
Mutates data in place and returns it. A delta IR ([1, 0, 0, …]) is an exact identity; a delayed delta shifts the signal by that many samples — useful sanity checks when building your own IRs.
Use when: the most accurate reverb character available — real captured spaces, plates, or cabinets, at the cost of needing an IR file.
Not for: parametric control over decay/damping (use any of the algorithmic reverbs) or extending the tail beyond the input length automatically.
Part of the @audio/reverb family.
MIT © audiojs