@audio/reverb-fdn

FDN reverb — 8 mutually-prime delays, O(N) Householder feedback, per-line damping (Jot topology)

npm install @audio/reverb-fdn

import fdn from '@audio/reverb-fdn'

Feedback delay network: 8 mutually-prime delay lines coupled through a Householder feedback matrix (I − 2/N·11ᵀ, applied in O(N) rather than a full N×N multiply), one-pole damping per line, and Jot's per-line decay gains g_k = 10^(−3·len_k / (T60·fs)) so every line decays at the same rate — a uniform, colorless T60 regardless of which delay you're listening to (Jot & Chaigne, 1991). Output taps use orthogonal ±1 sign patterns per channel for stereo decorrelation from a mono or stereo input. Strictly stereo, not N-channel: only data[0]/data[1] are read and written.

fdn(data, { t60: 1.5, damping: 0.3, mix: 0.3, fs: 44100 })
fdn([left, right], { decay: 0.9, mix: 0.4 })  // stereo, decay-derived T60

| Param | Default | | |---|---|---| | t60 | derived from decay | seconds — decay time to −60 dB; overrides decay when set | | decay | 0.6 | 0..1 convenience knob → t60 = 0.3 + 5·decay² (ignored if t60 is set) | | damping | 0.3 | 0..1 — one-pole high-frequency loss per delay line | | mix | 0.3 | 0..1 — wet/dry blend | | fs | 44100 | sample rate; delay lengths scale to it |

Mutates data in place and returns it. Fresh line state is allocated per call — no state carries across calls, reprocess the whole buffer each time. Measured T60 (Schroeder backward-integrated energy decay curve) lands within ~30% of the configured value — see the family test suite.

Use when: a colorless, uniform-decay tail is the goal — algorithmic reverb without per-mode coloration. Not for: plate/spring character (use dattorro/spring) or the cheapest possible CPU budget (use schroeder).

Part of the @audio/reverb family.

MIT © audiojs