Remove background noise from a recording

How to clean a noisy recording

1. Open the editor and drop your file in.
2. Best results: select a second or two where nobody is talking, just the room, and tick use selection as the noise sample. The start of a recording usually has such a stretch.
3. Pick a strength. Light takes about 10 dB of noise, Medium about 20, Strong about 34. On a normal recording, all three leave the voice within about one decibel of untouched.
4. Press clean out noise. A minute of audio takes a few seconds. Listen, and undo if you prefer the original.

What it can and cannot fix

• Great at steady noise: hiss, fans, hum, rumble, air conditioning, the room tone under everything.
• It keeps tracking noise that drifts, like a fan changing speed partway through.
• One-off sounds are different: a door slam or a dog bark is not steady noise. Select those moments and silence them by hand, which the editor also does.
• Pair it with levelling: clean first, then run volume levelling, and a rough phone memo becomes genuinely listenable.

Which noises can actually be removed

The distinction that matters is not how loud a noise is but whether it is steady. The filter learns a fingerprint of the constant background and subtracts it. Nothing about that approach can help with a sound that happens once.

The single thing that most improves the result

Give it a clean noise sample. Select one to two seconds where nobody is speaking and only the room is audible, tick use selection as the noise sample, and the filter learns your actual background instead of guessing. The start of a recording, before anyone begins, is usually perfect.

A sample with even a faint word in it teaches the filter that your voice is noise, which is precisely the recipe for the hollow, underwater artifacts people associate with noise reduction. Two seconds of genuine room tone beats ten seconds of nearly-silence.

Choosing a strength honestly

Light removes about 10 dB, Medium about 20, Strong about 34. On a normally recorded voice, where the background sits well below the speech, all three leave the voice within roughly a decibel of untouched, so Strong is not the reckless choice it sounds like. The setting governs how hard the background is pushed down, not how much the voice is mangled.

It does get dangerous on a recording whose noise is nearly as loud as the speech. There, no setting produces a clean result, because there is no clean signal underneath to recover.

When it sounds watery or hollow

1. Undo. Every pass is reversible and costs nothing but a moment.
2. Give it a better noise sample, as above. This alone fixes most cases.
3. Drop one strength level and try again.
4. If Light still sounds thin, the noise and the voice occupy the same frequencies and there is genuinely nothing more to be had. A slightly hissy recording that sounds like a person beats a silent one that sounds like a robot.

Order of operations

Clean before you level the volume. Levelling raises quiet passages and raises their noise with them, so cleaning first means there is less noise there to raise. Doing it the other way round is the commonest reason a recording ends up worse after two operations that each seemed sensible.

Common questions