Reverb is information about where

Reverb is information about where

A completely dry voice sounds like it exists nowhere. That is occasionally what you want (a narrator outside the story, an interior monologue) and almost always wrong for a voice that is supposed to be in a place. Reverb is not decoration; it is the cue that tells a listener the size of the room and the distance to the speaker.

The physical event it models is straightforward. You speak; the direct sound arrives first. A handful of distinct reflections arrive next, from the nearest surfaces, at times set by their distances. Then the reflections multiply until they become a dense, decaying wash.

Every reverb control maps to part of that.

Pre-delay: the control that matters most

Pre-delay is the gap between the direct sound and the start of the reverb. It corresponds to the extra distance the first reflection travelled, and at 343 m/s, 1 millisecond is about 34cm.

A 20ms pre-delay implies a reflecting surface roughly 3.4 metres away: the sound went 3.4m out and 3.4m back, arriving about 20ms after the direct. So pre-delay tells the listener how big the room is and how far from the walls the speaker is standing.

It also has a practical use that has nothing to do with realism: a pre-delay of 20 to 40ms separates the voice from its own tail in time, so the words stay intelligible even with a generous reverb. Zero pre-delay smears the reverb into the consonants and the voice goes mushy. If a reverb is making the dialogue hard to follow, increase the pre-delay before reducing the level.

Decay, early reflections, and the high-cut

Decay time (RT60) is how long the tail takes to fall by 60dB. For a voice in a believable interior, 0.4 to 0.8 seconds. A cathedral is 3 to 8 seconds and will swallow speech.

Early reflections are the first discrete echoes, and they carry room size more strongly than the tail does. A reverb with strong, well-spaced early reflections and a short tail reads as a small real room; a reverb that is all tail reads as an effect.

High-frequency damping is not optional. Air absorbs high frequencies over distance, and soft furnishings absorb them at every bounce, so a real room’s reverb tail is duller than its direct sound. A reverb return with full treble sounds artificial and metallic. Roll the reverb return off above 5 to 8kHz, and high-pass it below 200 to 300Hz too, which keeps the low end of the mix clear.

Send, not insert

Put reverb on a send: the dry voice goes to the output, a copy goes to a reverb bus, and you mix the wet return underneath. Three reasons.

• The dry signal stays untouched, so intelligibility is preserved and the balance is adjustable at any time.
• Several sources can share one reverb, which makes them sound as if they are in the same room, which is the point.
• You can process the return independently: EQ it, compress it, duck it under the dry voice.

An insert reverb with a wet/dry mix does roughly the same thing for one track and gives up the shared-space benefit.

Matching a room you already have

The commonest real task is not “add atmosphere” but “make this studio pickup match the on-location dialogue around it”. The line recorded in a booth has to sit inside a scene recorded in a hallway.

The approach that works:

1. Find a moment of the location recording with room tone and a reflection you can hear.
2. Use a convolution reverb loaded with an impulse response of a similar space. You can capture one yourself: record a balloon pop or a starter pistol in the room, and the recording is the impulse response. Every convolution plugin will load a plain WAV.
3. Match the pre-delay and decay by ear against the location line, alternating between them.
4. Match the tone with EQ before reaching for more reverb: a mismatch is usually brightness, not space.

Capturing an impulse response with a balloon costs nothing and is far more accurate than picking a preset called “Small Room”.

Amounts

For dialogue in a visible room, the reverb should be barely noticeable when you solo the voice and clearly missed when you mute it. That test (mute the return and see whether the voice suddenly floats) is the reliable one. If you can point at the reverb, it is too loud for dialogue.

Music and creative work have no such constraint. Dialogue does, because intelligibility is the job.

The free path

• Convolution: the IR.lv2 plugin, x42 zeroconvo, and Audacity’s built-in convolution all load WAV impulse responses free. Large libraries of impulse responses are available under permissive licences, including from the OpenAIR project.
• Algorithmic: Dragonfly Reverb is open source, cross-platform, and good. The Calf and LSP suites include reverbs. Ardour ships with a-Reverb.
• DaVinci Resolve’s free Fairlight page includes a full reverb set.
• ffmpeg has afir for convolution and aecho for crude echoes.

The limitation

Reverb adds a space; it cannot remove one. A recording made in a bad room already contains that room, and adding a nice reverb on top gives you two rooms at once, which sounds worse than either. Dereverberation tools exist and have improved sharply, and they work by estimating and subtracting the room response: with the same class of artefacts as noise reduction, plus a tendency to hollow out the voice.

The ordering is therefore the same as everywhere else in this module: get it right at the capture stage, because the processing stage is always a negotiation with a loss.