Compression in audio: what it does and when to use it

Search for audio compression and you get two unrelated subjects wearing the same name. Knowing which one you need saves a lot of confusion.

Dynamic range compression is a processing effect. It reduces the distance between the loudest and quietest parts of a recording so that a voice stays at a consistent level instead of jumping between a whisper and a shout. This is what engineers mean when they say a track is compressed, and it is what the rest of this page is about.

Data compression is a file-size question. It is what turns a large WAV into a smaller MP3 or AAC by discarding information the ear is unlikely to miss. It changes the size of the file, not the shape of the performance. If you arrived here wondering why your export is enormous, that is the subject you want, and the fix is a format and bitrate choice rather than an effect.

A compressor watches the incoming signal and waits for it to cross a level you set. Below that level it does nothing at all. Above it, the compressor turns the signal down by a proportion you choose. The loud moments come down; the quiet moments stay where they were.

That alone makes a recording quieter overall, which is not usually the goal. So the last stage of a compressor is make-up gain: you lift the whole signal back up now that the peaks are no longer in the way. The net effect is that the quiet parts get louder relative to the loud parts, even though the compressor only ever turned things down.

This is the single most useful thing to hold on to. A compressor does not make anything louder. It makes the loud parts smaller so that you can safely raise everything afterwards.

Threshold is the level at which the compressor starts working, measured in decibels below full scale. Set it too high and nothing is caught. Set it too low and everything is squeezed, including the quiet detail you wanted to keep.

Ratio decides how hard the signal is reduced once it crosses the threshold. At 2:1 a signal arriving 10 dB over the threshold leaves 5 dB over. At 4:1 it leaves 2.5 dB over. Ratio is the control people misjudge most often, so it is worth understanding how compression ratio works on its own terms.

Attack is how quickly the compressor responds once the threshold is crossed. A fast attack catches the sharp front edge of a consonant; a slower attack lets that edge through and clamps down just after, which usually sounds more natural on speech.

Release is how quickly it stops working once the signal falls back below the threshold. Too fast and you hear the level pumping between words. Too slow and one loud syllable holds the whole sentence down after it.

Speech does not need heavy treatment. A gentle ratio between 2:1 and 3:1, a threshold set so the compressor is only working on the louder half of your delivery, an attack in the region of 10 to 30 milliseconds and a release around 100 to 250 milliseconds will handle most voices without drawing attention to itself.

Watch the gain reduction meter rather than the numbers. On a conversational recording you want to see a few decibels of reduction on the loudest phrases and the meter returning to zero between sentences. If the meter never comes back to rest, the threshold is too low.

Trust your ears over any preset. The correct amount of compression is the amount at which you stop noticing the level and start following the words.

Compressing a bad recording rarely rescues it. Compression raises everything that sits below your loudest moments, which includes room tone, air conditioning and traffic. A recording with a noticeable noise floor will sound worse after compression, not better, because the noise comes up with the voice.

Deal with the noise first and compress afterwards. A noise reducer run before compression gives the compressor much less unwanted material to lift.

Stacking compressors without noticing is the other common trap. A compressor on the recording, another on the mix and a loudness stage at export can add up to a flat, breathless result. If your finished audio sounds tiring rather than clear, look for how many stages are squeezing it rather than reaching for another one.

These three get used interchangeably and they are not the same thing.

Normalization applies one fixed change in gain to the whole file so that it lands on a target level. It does not alter the relationship between loud and quiet moments at all. If your episode is simply too quiet from beginning to end, normalizing is the correct fix and compression is not.

Limiting is compression taken to its extreme, with a ratio so high that effectively nothing gets past the ceiling. It exists to stop peaks, not to shape a performance, and it sits at the end of a chain rather than the middle.

Compression sits between the two. It changes the internal dynamics of a recording so that a voice holds a steady place in the mix, and it is the only one of the three that changes how a performance feels rather than only where it sits.