How to remove background noise from a video

The quickest way is the free remove background noise from video tool. It cleans the sound in your browser, on your device: nothing is uploaded, there's no watermark and nothing to sign up for.

1. Choose your video (MP4, MOV, WebM or MKV), or a sound file such as an MP3, WAV or M4A. It takes up to 60 minutes of video, or 3 hours of sound, at a time.
2. Pick a strength: Light, Medium or Strong. It starts on Medium, which suits most videos. The strength is how much of the original sound is mixed back in: Light the most, to keep a little of the room, and Strong none, for loud noise.
3. Compare Before and After. Listen to the voice, not just the noise: the voice is what has to survive.
4. Download it. The picture is copied as it is, not re-encoded, so it looks exactly as it did. A sound file comes back as an MP3.

How it works. The tool runs RNNoise, an open-source noise suppressor: a small neural network trained to keep speech and drop the rest, such as fans, hum, traffic and room noise. Because it's made for voices, it lowers music too, so it's not for music videos. And it can't fix clipping or distortion or two people talking over each other, and it doesn't remove echo.

Editors do the same job under other names: CapCut's noise reduction, Premiere's Enhance Speech, DaVinci Resolve's Voice Isolation, Audacity's Noise Reduction. Where to find each one is further down. First, what any of them can and can't do.

Noise removal works by telling the voice apart from everything else. The steadier a sound is, and the less it sounds like speech, the cleaner it comes out. Here's what to expect, by kind of noise:

Two rules come out of the table. Steady noise is the easy kind; noise that sounds like speech, or that damaged the recording, can't be taken back out. And the cleaner the recording to begin with, the less any tool has to do, which is why the best noise removal happens before you press record.

Every tool works better when the voice is much louder than the noise. Widen that gap when you record and the clean-up is light, or not needed at all. Audacity's own guide says it plainly: you can rescue a noisy file, but it sounds best if you make the room quiet first.

• Get the mic close. It's the biggest single fix: the voice gets louder while the room stays where it is. Halving the distance from mouth to mic makes the voice roughly 6 dB louder against the room. A clip-on mic on the chest, or a desk mic a hand or two from the mouth, beats the camera's own mic across the room.
• Turn off what hums. Air conditioning, fans, heaters, a fridge, a laptop's fan next to the mic, notifications. You stop hearing them after a few minutes; the mic never does.
• Pick the room. A small, furnished room sounds better than a big empty one. Avoid kitchens, bathrooms and bare walls, stay away from windows onto a street, and record when the building is quiet.
• Soften it. Rugs, curtains, a sofa, full bookshelves and a duvet over a hard wall behind the camera soak up echo. A wardrobe full of clothes makes a good voiceover booth.
• Outdoors, stop the wind. A foam or furry windscreen on the mic, your back to the wind, and a clip-on mic under one layer of clothing (checked for rustle).
• Leave headroom. Set the level so the loudest laugh stays out of the red. A recording that clips can't be fixed later; one that's a little quiet can be turned up.
• Record ten seconds of room tone. Everyone silent and still, before you start. Audacity's Noise Reduction learns the noise from a stretch like that, and it fills gaps in the edit.

If you're editing the video anyway, your editor may already have a noise tool. Here's where each one is, as its maker describes it:

Whichever you use, take the noise out first, before compression or loudness, which raise everything quiet, the noise with it. If your editor has no noise tool, clean the video with the free noise remover and import the cleaned file. Or save the sound as a WAV with the video to MP3 converter, clean it in Audacity, and put it back under the picture in your editor.

Use the lightest strength that gets the noise out of the way. Noise reduction always takes a little of the voice along with the noise; turned up, it takes a lot. Audacity's manual gives the same advice for its own setting: the lowest value that brings the noise to an acceptable level. What too much sounds like:

• Underwater or robotic. A voice that sounds hollow, metallic or as if it's coming through a phone.
• Warbling in the gaps. Short, watery tones bubbling where the noise was. Audacity calls them artifacts: random bursts of very short tones.
• Missing edges. Soft word endings, s and f sounds, and breaths that vanish, so the speech sounds clipped.
• Noise that breathes. The background drops away when someone speaks and swells back between words.

How to find the right setting:

1. Start light. In the free tool, that's Light; in an editor, the lowest amount.
2. Listen on headphones to a stretch with both talk and a pause. The pause tells you about the noise; the talk tells you about the voice.
3. Go up a step only if the noise still distracts, and stop at the first setting where it doesn't.
4. Compare at the same volume. A cleaned track can come out a little quieter, and quieter always sounds worse.
5. Check the hard parts: a loud laugh, a soft answer, and the quietest person on the recording.

Leave a little of the room. A voice in perfect silence sounds odd, and a jump between a cleaned clip and an uncleaned one is more noticeable than faint, steady noise. That's why the tools mix some of the original back in: the free tool's Light and Medium, Premiere's Mix Amount, Resolve's Amount. And clean once: running noise reduction twice, or stacking two tools, stacks their artifacts too.

• Electrical hum. Mains hum sits at 60 Hz in North and Central America and 50 Hz in most other countries, Audacity's guide notes, with overtones at multiples of it. A notch filter at that frequency (Audacity's Notch Filter, or a de-hummer like Resolve's) takes it out more cleanly than general noise reduction. A hum from a cable is a fault: try another cable or socket before you fix it in the edit.
• Echo. Echo is the room inside the recording, not a separate sound, so a speech tool keeps it: the free tool leaves it as it is. Tools built for it go further (Descript's Studio Sound, Adobe Podcast's Enhance Speech), but nothing beats a closer mic in a softer room.
• Wind. The thumps are the mic overloading. A low-cut filter takes out the rumble under them; if words are lost, record them again or cover them with a voiceover.
• Music. A speech tool lowers music along with the noise. Resolve's Voice Isolation is described by Blackmagic as removing music playing while someone speaks. If you want music under the finished video, record without it and add it after with the add music to video tool, which can lower the music when someone talks.
• When it can't be saved. Take the sound out with the free remove audio from video tool, which keeps the picture exactly as it is, and record a voiceover instead.

Noise costs more on a short clip than on a long video. A clip plays on a phone speaker, often in a noisy place, and the viewer decides in a second or two whether to stay. Captions carry the clip for people with the sound off; clean sound carries it for everyone else. And automatic captions get more words right from a recording that was clean to begin with: Google's speech-to-text guide says heavy background noise and echo reduce accuracy.

For a podcast, clean the whole episode once, before you cut anything, so every clip comes from clean sound. Then take out the long pauses with the free remove silence from video tool: with the noise gone, the pauses are quieter and easier for it to find. The settings for that are in how to remove silence from a video, and the whole edit, from the rough cut to the export, is in how to edit a podcast. If you're still choosing mics and a room, start with how to start a video podcast.

Then the clips. Upload the cleaned episode (MP4, MOV, WebM or MKV, up to 4 GB) or paste its YouTube link, and Clyper reads the transcript to find the moments that stand on their own, cuts each one on whole sentences, frames it 9:16 on whoever's talking and captions every word in one of 5 animated styles. It evens out every clip's loudness, to -14 LUFS, so a week of clips plays at one level. It doesn't remove noise, so clean the video first. More in podcast clips with Clyper.