Audacity compressor: settings that work for spoken audio

Nobody speaks at a constant volume. You lean in for an aside, you get animated on a point you care about, you trail off at the end of a sentence. Live, the ear follows all of it without effort. In a recording played through phone speakers on a train, the quiet parts vanish and the loud parts startle.

A compressor narrows that gap. It turns down the parts above a level you set, leaves everything below alone, and then you raise the whole thing back up. The result is a recording with a smaller distance between its loudest and quietest moments, which is easier to listen to almost everywhere.

In Audacity it lives under Effect, Volume and Compression, Compressor. There is also a separate Limiter, which is a compressor with an extreme ratio used to stop peaks, and the two do different jobs.

Threshold is the level above which the compressor starts working. Audio quieter than this passes through untouched. Set it at minus 20 dB and only the parts louder than minus 20 dB get turned down.

Ratio is how hard it turns them down once they cross. At 2:1, a signal 10 dB over the threshold comes out 5 dB over. At 4:1, it comes out 2.5 dB over. Higher ratios mean more aggressive flattening.

For one voice speaking normally, a threshold around minus 18 to minus 20 dB with a ratio between 2:1 and 3:1 is a sensible place to start. That is enough to even out an ordinary performance without making it sound squashed.

If your recording swings wildly, an interview where one person is much louder than the other, you may want 4:1. Past about 6:1 you are limiting rather than compressing, and speech starts to sound flat and lifeless.

Attack is how quickly the compressor reacts once the signal crosses the threshold. Short attack times catch the very start of a word, which controls plosives and sharp consonants but can dull the natural punch of speech. Longer attacks let the first moment through untouched, which usually sounds more natural.

Release is how quickly it stops working once the signal drops back. Too short and you hear the background noise rise and fall between words, a pumping effect that is very hard to unhear once you have noticed it. Too long and it never lets go, so a loud word suppresses the quiet phrase that follows.

For speech, an attack around 0.2 seconds and a release around 1 second is a reasonable starting point. Audacity's defaults are in a usable range, and these two are worth adjusting only after threshold and ratio are right.

Make-up gain, labelled as making the gain normal after compressing, raises the whole signal back up after the loud parts have been reduced. Without it, compression just makes everything quieter and appears to have achieved nothing. Leave it on.

The second option compresses based on peaks rather than average loudness. For speech, the average is generally the better guide, because a single plosive should not set the behaviour for a whole sentence. Leave it off unless you have a specific reason.

After compressing, run Normalize or Loudness Normalization to set the final level. Podcast platforms target around minus 16 LUFS for stereo, and Audacity's Loudness Normalization can target that directly, which is more useful than normalising to a peak.

Compression raises quiet things, and background noise is a quiet thing. Compress a noisy recording and you will hear the noise come up with the voice, sometimes dramatically. Remove the noise first, then compress.

It also cannot repair clipping. If the recording was already too loud going in and the peaks are flattened, compression works on a signal that is already damaged and the distortion stays.

And it cannot fix distance. Someone two feet from the microphone sounds distant, and compressing that brings up the room along with the voice, which makes it sound worse rather than closer.

If tuning six controls per episode is not how you want to spend the evening, Hilite levels spoken audio automatically as part of enhancing a recording. We make it, so read that accordingly. Audacity is free and does this properly once you know which two sliders matter.