What a compressor does, and what it does to the gaps

The one-sentence description

A compressor turns down anything louder than a threshold, by a ratio, with a chosen speed of response. It makes nothing louder. What makes things louder is the make-up gain you add afterwards — and that gain raises everything, including whatever noise sits between the words.

That last clause is the whole risk. Motion, Sound and Music covers the general principle; this lesson is about the settings for a speaking voice in an edit.

Why dialogue needs it

A person speaking varies by 20 dB or more between an emphatic word and a trailing-off aside. On good headphones in a quiet room that is fine. On phone speakers on a bus, the quiet half is inaudible and the loud half is the only thing you hear.

Compression narrows that range so the whole sentence sits in the band a phone speaker can deliver. It is not an effect; for spoken-word video it is a delivery requirement.

The five controls

Threshold — the level above which the compressor acts. For dialogue levelled to peaks around -12 dBFS, start the threshold around -18 dBFS and adjust.

Ratio — how hard it turns down. 3:1 means an input 3 dB above threshold comes out 1 dB above. For dialogue, 2:1 to 4:1. Above about 8:1 you are limiting rather than compressing.

Attack — how quickly it responds once the threshold is crossed. 5 to 15 ms for speech. Too fast and you flatten the consonant transients that carry intelligibility, and the voice goes dull. Too slow and the loud syllable escapes before the compressor acts.

Release — how quickly it stops once the signal drops. 50 to 150 ms for speech. Too fast and you hear the background noise surge up between words — the artefact called pumping or breathing, and once you can hear it you cannot stop hearing it. Too slow and the compressor never recovers between phrases, so quiet words after a loud one stay suppressed.

Make-up gain — the level added afterwards to compensate. Set it so the compressed signal is about as loud as the uncompressed one was, then compare with the bypass button.

The number to watch

The gain reduction meter. For dialogue you want it moving 3 to 6 dB on the loud words and returning to zero between phrases.

If it is pinned at 10 dB and never releases, the threshold is too low and the voice is being squashed flat. If it never moves, the threshold is too high and the compressor is doing nothing.

This single meter is a more reliable guide than any preset, because it shows you the compressor's behaviour rather than its settings.

The noise floor problem, precisely

Here is the mechanism that catches people.

Compression reduces the difference between the loudest and quietest parts. Make-up gain then raises the whole signal. The voice comes up — and so does the air conditioning, the room hum, the traffic, the computer fan.

So a recording with a poor noise floor gets audibly worse with compression, and the worse the original recording, the less compression it can take. Squash a hissy recording by 12 dB and add 12 dB of make-up, and you have made the hiss 12 dB louder than it was.

This is why the recording lessons come before this one. Compression does not create the problem; it reveals how much of one you had.

Two-stage compression

The approach professionals use, and it is gentler than one heavy compressor:

1. A light compressor on each dialogue track — ratio 2:1, 3 dB of reduction — to even out that speaker.
2. A light compressor on the dialogue bus — ratio 2:1 or 3:1, another 3 dB — to glue everything together.

Two lots of 3 dB sound far more natural than one lot of 6 dB, because each stage is working gently and neither is being asked to catch everything.

The limiter, which is different

A limiter is a compressor with a very high ratio and a very fast attack, used as a ceiling rather than as shaping. One goes on the main output, with the ceiling at about -1 dBFS true peak, to catch anything that would otherwise clip the encoder.

It should be doing almost nothing. If your limiter is pulling down 6 dB regularly, your mix is too loud and the limiter is now your mix engineer.

Free tools

Resolve's Fairlight has a compressor and limiter on every channel and on every bus, plus a dedicated Dialogue Processor that combines gate, EQ, compression and de-essing with sensible defaults. Reaper's ReaComp is excellent. Audacity has a basic compressor; for anything beyond levelling, use one of the multi-track tools instead.

Today

Put a compressor on a voice recording, set the ratio to 3:1, and lower the threshold slowly while watching the gain reduction meter and listening to the gaps between words. Find the point where you first hear the background surge. That point is your ceiling on that recording.