Narrow the gap between the loudest and quietest parts of a file. The compressor watches the signal, and once it crosses the threshold it turns the level down by the ratio you set. Makeup gain then lifts the whole result back up, which is why compressed audio sounds more even rather than merely smaller.
Threshold and ratio
Threshold is the level, in dBFS, where the compressor starts working. Anything below it passes untouched.
Ratio decides how much of the excess survives. At 4:1, a signal arriving 8 dB over the threshold leaves 2 dB over it. At 1:1 nothing is reduced, which is the setting to use when you want makeup gain alone.
Peaks are measured against the loudest channel at each moment, so a stereo file keeps its image instead of drifting when one side gets louder.
Attack and release
Attack sets how quickly the reduction arrives, release how quickly it lets go.
A fast attack of 1 to 5 ms clamps down on the front of a word or a drum hit, which controls peaks but dulls the attack of the sound. A slower 20 to 30 ms lets the transient through and catches the body behind it.
Release is the one to watch on continuous material. If it is shorter than the gaps in the audio, the level audibly lifts between phrases, which is the pumping sound that gives heavy compression away. Longer releases stay out of the way but respond more slowly to a genuinely quieter section.
Soft knee
The knee is how many decibels wide the transition around the threshold is. At 0 dB the ratio applies the instant the signal crosses, which is precise and easy to hear working.
A 6 to 12 dB knee starts easing the ratio in below the threshold, so the compressor is already partly engaged when the loud part arrives. On vocals and acoustic sources that reads as smoother.
Example: an uneven voice recording
A voice track peaking at -3 dBFS with quiet asides at -30 dBFS has 27 dB of range, which is too much for a phone speaker.
Set the threshold to -20 dBFS, the ratio to 4:1, attack to 10 ms, release to 200 ms, and a 6 dB knee. The peaks land near -15 dBFS. Add 8 dB of makeup gain and the quiet asides are now clearly audible while the loud parts have stopped jumping.
Gain reduction is not loudness normalisation
A compressor reacts to the signal moment by moment. It cannot know the loudness of the whole file, so it will not hit a target LUFS figure on its own.
If the goal is a specific delivery loudness, compress first for consistency, then set the final level. Pushing makeup gain until the meters look right usually pushes the peaks past full scale, and a limiter is the safer way to stop that from clipping.
Applying it to part of a file
The range control confines the compressor to one span of the timeline, with a short crossfade at each edge. Audio outside that span comes back untouched, which is what you want when one section of a recording is much louder than the rest and the rest already sits well.