The default insert-latency budget per role, in milliseconds — the accumulated latency
of a destination's whole insert chain, not a per-plugin allowance.
These are derived, not picked:
monitor = 5 ms. A performer hears their own voice or instrument acoustically (and,
for a voice, through bone conduction) at the same time as the monitor feed, so the
feed's delay is heard as a slap/comb against the direct sound rather than as latency.
Published live-monitoring listening tests (Lester & Boley, The Effects of Latency on
Live Sound Monitoring, AES 123rd Convention, 2007) put the point where performers
reliably notice and object at roughly 10 ms, with the most sensitive — vocalists, and
in-ear users who have no acoustic bleed to mask it — reacting from around 4–6 ms. That
~10 ms is the WHOLE path, and the console has spent much of it before any plugin runs:
conversion plus two quantum crossings is already ≈ 3–4 ms at 64 frames / 48 kHz. Half
the audible boundary is therefore what is left to spend on inserts, and the other half
is reserved for conversion and buffering. It is deliberately the same number as the
catalog's LIVE_MAX_MS live/studio gate, so "a live-tier plugin" and "a plugin that
fits a monitor bus on its own" mean the same thing instead of being two thresholds that
drift apart.
program = 100 ms. A program bus is not heard against the listener's own sound; it
is heard against what the audience SEES on stage, and for a broadcast/recording split
against picture. ITU-R BT.1359-1 puts the detectability threshold for sound lagging
picture at 125 ms and the acceptability limit at 185 ms. 100 ms keeps the program path
inside detectability with margin for the rest of the system — and leaves the
mastering-grade tools that belong on a MAIN entirely usable, because a program-role
overrun WARNS with the number rather than refusing (see DEFAULT_ENFORCEMENT).
effects = 60 ms. An FX return sums back against the dry signal, so its latency is
heard as pre-delay nobody dialled in. Fusion of a delayed copy with its direct sound
holds to roughly 30–50 ms for speech and out to ~80 ms for music (the precedence /
Haas window); beyond that it separates into an audible discrete repeat. 60 ms sits at
the practical edge of that window.
All three are overridable per role and per bus (InsertSuitabilityPolicy) — a rig
with 2 ms of conversion and a 32-frame quantum can afford more on a wedge than one on
USB, and only the operator knows which they are on.
The default insert-latency budget per role, in milliseconds — the accumulated latency of a destination's whole insert chain, not a per-plugin allowance.
These are derived, not picked:
monitor= 5 ms. A performer hears their own voice or instrument acoustically (and, for a voice, through bone conduction) at the same time as the monitor feed, so the feed's delay is heard as a slap/comb against the direct sound rather than as latency. Published live-monitoring listening tests (Lester & Boley, The Effects of Latency on Live Sound Monitoring, AES 123rd Convention, 2007) put the point where performers reliably notice and object at roughly 10 ms, with the most sensitive — vocalists, and in-ear users who have no acoustic bleed to mask it — reacting from around 4–6 ms. That ~10 ms is the WHOLE path, and the console has spent much of it before any plugin runs: conversion plus two quantum crossings is already ≈ 3–4 ms at 64 frames / 48 kHz. Half the audible boundary is therefore what is left to spend on inserts, and the other half is reserved for conversion and buffering. It is deliberately the same number as the catalog'sLIVE_MAX_MSlive/studio gate, so "a live-tier plugin" and "a plugin that fits a monitor bus on its own" mean the same thing instead of being two thresholds that drift apart.program= 100 ms. A program bus is not heard against the listener's own sound; it is heard against what the audience SEES on stage, and for a broadcast/recording split against picture. ITU-R BT.1359-1 puts the detectability threshold for sound lagging picture at 125 ms and the acceptability limit at 185 ms. 100 ms keeps the program path inside detectability with margin for the rest of the system — and leaves the mastering-grade tools that belong on a MAIN entirely usable, because a program-role overrun WARNS with the number rather than refusing (see DEFAULT_ENFORCEMENT).effects= 60 ms. An FX return sums back against the dry signal, so its latency is heard as pre-delay nobody dialled in. Fusion of a delayed copy with its direct sound holds to roughly 30–50 ms for speech and out to ~80 ms for music (the precedence / Haas window); beyond that it separates into an audible discrete repeat. 60 ms sits at the practical edge of that window.All three are overridable per role and per bus (InsertSuitabilityPolicy) — a rig with 2 ms of conversion and a 32-frame quantum can afford more on a wedge than one on USB, and only the operator knows which they are on.