Guide · OtterStudio

Remove background noise from a recording

A voiceover recorded in a kitchen has the fridge in it. A phone recording has hiss. A laptop microphone has the fan. None of that is in the words, and a steady noise like that can be taken out without touching the voice.

Open OtterStudio →

One button

Add the recording to OtterStudio, or a video that has the recording as its sound. Click the audio clip on the timeline. In the card on the right, under the volume fader, there is a wide green button: Remove noise. Press it. It fills up as it works, listening first, then cleaning; thirty seconds of audio takes about two seconds.

When it is done, a new clip appears on a new track directly under the one you pressed the button on, the same length, in the same place. That is the cleaned version. The original is still there above it, muted. Nothing has been thrown away.

Listen to both

Press play. You are hearing the clean one. To compare, mute the clean track with the speaker button at the left of its lane and unmute the original. Go back and forth a few times on a stretch with a pause in it, because the pauses are where the difference is loudest.

If the result is too much, if the voice sounds thin or there is a watery, underwater quality on the quiet syllables, press Undo in the card. The copy goes and the original is unmuted, as if nothing had happened. There is no setting to fiddle with; the method has one strength, chosen to suit speech.

What it actually does

The noise in a recording is roughly the same from one second to the next. The voice is not. So the tool listens to the quietest tenth of the clip, the gaps between sentences, and takes what it hears there as a picture of the noise: how much hiss at each pitch, how much hum. Then it goes through the whole recording and, at each moment and each pitch, turns down whatever is not clearly louder than that picture. A word is far louder than the hiss under it, so it passes. The hiss between words is not, so it drops by about eighteen decibels, which is most of the way to silence.

This is the same method Audacity's Noise Reduction uses and radio producers have used for decades. It is done here with the browser doing the maths, in a background thread, so you can keep editing while it runs.

Where it works and where it doesn't

It works on noise that is steady: hiss, hum, air conditioning, a fan, traffic rumble, the room's own tone. It does not work on noise that changes, like a dog barking, a door, someone talking in the background, or music, because those do not look like the picture it learned. They come through as they were.

It needs a quiet moment to learn from. A clip that is wall-to-wall talking with no pauses, or a music track, has no gaps, and the tool will tell you it found no steady noise and change nothing. Recording two seconds of silence at the start of a take is the old advice and it still helps.

It is not magic on very bad recordings. If the hiss is almost as loud as the voice, the voice is damaged by whatever removes the hiss. Recording a bit closer to the microphone does more than any tool afterwards.

In a video

A video's sound sits on its own audio track from the moment you add it, so the button is right there on that clip. The picture is untouched. When you export, the clean track is what goes into the MP4, and the muted original stays muted.

Just the audio

If the timeline has only sound on it, Export saves an M4A instead of a video: AAC audio in a file that plays wherever an MP3 does. Browsers cannot write MP3 themselves, which is why it is M4A and not MP3, and for a voice recording you will not hear the difference.