How to remove background noise from a video

Background noise in a video can be cleaned up with Video by Zaweba's noise reduction. It learns the noise profile from each clip's quietest half-second and removes that noise from the whole clip, and voice EQ presets then add clarity to speech. Everything runs on your own PC and is processed once, then cached, so playback stays smooth.

Steady noise is the kind it handles best: the hiss of a cheap microphone, a fridge, an air conditioner, the rumble of a room. These sounds are the same all the way through, which is exactly what a noise profile captures.

How the noise profile works

Every clip has a moment where nobody is talking, even if it's only a breath between sentences. Video by Zaweba finds the quietest half-second, treats what it hears there as the noise, and takes that sound out of the rest of the clip. It uses FFmpeg's afftdn filter to do it.

Because the profile comes from each clip, a film shot in three rooms gets three different profiles automatically. That's a real help with interviews or vlogs recorded in several places, where one global setting would suit none of them properly.

It also explains the one thing to watch for. If a clip has no quiet moment at all, with music or talking from start to finish, there's no clean sample of the noise, and the result will be less good. Recording a couple of seconds of room tone at the start of a take, with everyone silent, gives the filter something honest to learn from.

Adding voice EQ

Noise reduction removes. Voice EQ shapes. The voice presets lift the parts of speech that carry the words and calm the muddy bits underneath, so a voice sounds clearer after the hiss has gone. Try a preset, listen, and switch it off and on to compare.

Both are processed once and cached. The first time you switch one on there's a short wait while it renders; after that the preview plays the cleaned sound without strain.

Other tools for other noises

Not all noise is steady hiss. The sound effects library in Video by Zaweba has more specific fixes: a hum remover for mains hum, high and low pass filters for rumble and fizz, a de-esser for sharp S sounds, de-reverb for echoey rooms, click and clipping repair, a gate to quieten the gaps between phrases, and a voice clarity effect. There's also a dialogue repair group once you tag a clip as dialogue.

Use them one at a time and only where they help. Stacking every repair on a voice usually makes it sound thin and processed, and a little remaining room sound is more natural than a voice in a vacuum.

Getting the mix right afterwards

Once the speech is clean, the rest of the mix falls into place more easily. Auto-ducking drops music under the dialogue's actual speech, so you don't have to fight the music for clarity. On export the whole mix is measured and levelled to -14 LUFS, the loudness most video sites aim for.

Listen on headphones for the noise itself and on a phone speaker for the overall feel. If the voice is clear on both, you're done.

No tool brings back speech that the noise has completely buried. For everything short of that, Video by Zaweba gives a recording a noticeably cleaner sound.