Pre-delay is the gap between the dry signal and the first output of the reverb. On most plugins it ranges from 0 to about 100 ms, with a practical working range around 10 to 60 ms. Set it to zero and the reverb starts the instant the source does. Push it past the echo threshold and the reverb detaches into a separate event. Somewhere in between sits the parameter that decides how present a vocal feels.
The reason a few milliseconds matter this much was settled in a 1949 dissertation by Helmut Haas, work that had nothing to do with music production.
What Haas actually measured
Helmut Haas submitted his thesis "Über den Einfluss eines Einfachechos auf die Hörsamkeit von Sprache" in 1949 (the work was published in Acustica in 1951). The motivation was practical and not particularly musical. Public address systems in lecture halls were producing amplified voices that sometimes sounded doubled or unintelligible, and nobody had cleanly characterised when a delayed copy of speech would help or hurt comprehension.
Haas placed two loudspeakers at 45 degrees left and right of a listener, played speech through one, then played a delayed copy through the other, and varied the delay. He found three regimes that have shaped every reverb plugin since:
- Delays under roughly 1 ms shift the perceived stereo image without producing two distinct sounds. This is also the regime where mono summing creates comb filtering.
- Delays between about 1 and 30 ms (for clicks, narrower) or up to about 40 to 50 ms (for speech) fuse into a single perceived event. The listener locks onto the first arrival and uses the second to infer space. This is the precedence effect, sometimes called the law of the first wavefront.
- Above the echo threshold, the second sound separates into a discrete echo. For speech that boundary sits around 40 to 50 ms. For music it can stretch toward 80 to 100 ms.
The transition point depends on the signal. Transient material like a snare hit or a click separates earlier. Sustained material like a sung vowel hangs together longer.
Why pre-delay unmasks vocals
Without pre-delay, the reverb starts building energy during the consonant transients and the initial formant of every word. The reverb output overlaps the spectrally rich part of the vocal at exactly the moment the listener is trying to identify it. The result is forward masking from the louder direct sound and simultaneous masking across overlapping frequencies. The vocal sounds like it is fighting through fabric.
With 30 to 50 ms of pre-delay, the attack and initial formant clear out before the reverb energy arrives. The listener perceives a clean onset followed by ambience. The decay tail is still the same length, so the room feels just as large, but the source feels just as close.
This is the regime engineers reach for on lead vocals: enough delay to clear the transient, not so much that the gap becomes audible as a separate event.
A starting point that actually starts somewhere
Ranges that show up consistently across iZotope, PreSonus, and Audio Sorcerer references:
- Lead vocals (pop, intelligibility first): 20 to 30 ms.
- Lead vocals (ballad, intimacy): 40 to 60 ms.
- Snare reverb: 10 to 20 ms. Drums have sharper transients, so the fusion window is shorter.
- Sustained synths and pads: 50 to 100 ms is fine. Long onsets keep the connection to the source.
- Plucked or staccato material: shorter pre-delay, otherwise the reverb sounds detached from the note.
A standard troubleshooting move: if a mix is muddy and the vocal sits in the soup, the first knob to try is not shorter decay. It is longer pre-delay. The vocal stays present and the space stays large.
Tempo-synced pre-delay
Many engineers set pre-delay to a musical subdivision so the reverb onset feels rhythmically integrated rather than arbitrary. The math is straightforward: divide 60000 by the tempo in BPM to get the millisecond duration of one quarter note, then multiply by the note fraction. At 120 BPM, a 1/32 note works out to 62.5 ms and a 1/64 note to 31.25 ms.
The 1/64 value at moderate tempos lands almost exactly on the Haas threshold for speech. That is not a coincidence, more a happy alignment between musical subdivision and perceptual fusion.
If a reverb has no tempo sync, the workaround is to insert a tempo-synced delay before the reverb, set its time to the desired subdivision, and set feedback to zero. The first echo becomes the pre-delay; everything downstream behaves normally.
Where the architecture lives
Pre-delay sits at the front of a three-stage signal flow that most algorithmic reverbs share. First a delay line. Then early reflections, often a sparse pattern of discrete echoes representing first bounces off walls. Then a diffuse tail built from a feedback delay network or Schroeder-style comb and allpass structure. Early reflections carry most of the spatial information about source distance and room size. The decaying tail mostly conveys room character. Pre-delay separates the source from both.
In a convolution reverb, pre-delay is either baked into the impulse response or offered as a separate parameter that delays the convolution output.
The contradiction nobody fully resolves
A small irritation worth flagging. Pop and studio mixers often describe long pre-delay as making vocals closer, reasoning that the cleared transient brings the source forward. Orchestral mixers often describe long pre-delay as making the source farther, reasoning that more delay implies more space between the singer and the surrounding walls. Both communities are correct within their conventions. The simulated space is what differs, not the perceptual mechanism. If you find yourself confused reading two tutorials that contradict each other on this point, you are not missing anything subtle. They are describing different intents in different rooms.
What to take away
Pre-delay is one delay line in front of a much more complex network, but it carries a lot of the perceptual weight of a reverb send. The values that work in practice (20 to 50 ms for most vocals, less for percussion, more for pads) sit inside the fusion window that Haas measured for delayed speech in 1949. Setting it well is less about following a preset and more about deciding whether the listener should hear the singer first or the room first.