Video audio cleanup is the work of turning the sound of a clip into something people will sit through: the fan gone, the level even, the dead air cut, and the whole thing back on the picture where it started. It is one job with five steps, and most of the wasted time in it comes from doing them in the wrong order.
This page is the map. It tells you which problem your clip has, which of the seven video guides here solves it, and the order the steps go in. The controls named below are Gleam Audio’s, which takes video files directly and works on your computer, so nothing is uploaded. The same order works in any editor.
Start with the file you actually have
The first question is not “how do I clean this” but “what am I allowed to touch”. Open the clip in an editor and look at the timeline, because two clips that sound the same need completely different work.
| What the clip has | What you can fix | Where to go |
|---|---|---|
| One track, voice and noise mixed together | the noise, the level, the length | the steps below, in order |
| A voice track and a music track, separately | mute the music and lose nothing | background music |
| One track, voice and music mixed together | separate the voice from the mix, with marks left on it | background music |
| Wind on an outdoor shoot | the rumble, at the mic first | wind noise |
| Sound you want gone entirely | mute it, no processing needed | remove audio from video |
If the music sits on its own track, you are finished in ten seconds and no amount of cleanup will beat it. Everything below is for the mixed case.
Get the sound out of the picture first
Two routes, and both are fine.
Drop the video into Gleam Audio. Every tool takes mp4, mov, mkv, webm and avi files for their sound, so there is no separate extraction step. Pick the tool, add the clip, run it. The result is audio only, saved as an m4a with clean in the name, so the picture stays in your editor untouched and you put the new sound back on it at the end.
Or pull the audio out with ffmpeg, if you would rather keep the video file as the thing you carry through the steps:
ffmpeg -i clip.mp4 -vn -c:a copy audio.m4a-vn drops the video, -c:a copy keeps the audio exactly as it was, so nothing is re-encoded and nothing is lost. The output is audio only.
There is a third route for short clips: the free MP4 to MP3 tool in the browser, for when you only need the sound as a file to hand to someone. The remove audio from video guide covers all three, plus the case where you want the clip silent rather than cleaned.
Remove the noise next, before anything else
Do this before you raise the level. Any gain you add to a noisy clip adds the noise with it, and then you are removing noise that you made louder.
In Gleam Audio, Remove noise has one control that matters at first: Strength, from 0 to 100. It sits at 100, which removes as much noise as it can. If the voice comes back thin or unnatural, run it again lower. Repair clicks is separate and off until you turn it on; it redraws only the samples around a click, so it is the right tool for a chair creak or a keyboard tap and the wrong one for a fan.
Which guide to read depends on where the clip came from:
- A camera or a screen recording, cleaned on a computer: reduce background noise in a video.
- A phone clip, and you want to know what iOS can do before you open anything: noise on an iPhone. Audio Mix on recent iPhones can quiet a clip, but only under conditions the guide lays out, and it stops well short of a proper cleanup.
- A clip already inside DaVinci Resolve: noise in DaVinci Resolve, which covers Noise Reduction, Voice Isolation and De-Hummer, and says when to stop and treat the audio as a file instead.
- A short clip and no appetite for installing anything: remove noise from video online free, which is the browser route with its ten minute limit.
One thing to expect: the harder you push, the more the voice stops sounding like a person. Why noise reduction makes voice sound robotic explains what the model is doing when it gets it wrong.
Even the level while you are in there
Loudness is the step people skip, and it is the one viewers notice. A clip that is quiet next to the next video gets turned up, and the noise comes back with it. Platforms normalize everything to a target, so a clip mastered far above it is simply turned down.
The targets are on the LUFS page: -23 for broadcast, -16 for podcasts, -14 for music streaming. You do not need to memorize them; you need the clip to land on one.
Three ways, depending on your tools:
- In Gleam Audio, Enhance starts with Kind of recording. Pick Voice-over or narration for a video voice track and it fills in the steps: light noise removal, a tight even level, and softer s sounds. Podcast is the one for a talking-head episode.
- In DaVinci Resolve, the level step has its own guide: normalize audio in DaVinci Resolve.
- On the command line, normalize audio with ffmpeg.
Whichever you use, measure before and after rather than trusting your ears on laptop speakers. The free LUFS meter shows the integrated loudness next to each target. What normalize audio means is the longer version if the terms are new.
Cut the parts that should not be there
Now that the sound is clean and even, the remaining job is length. Two different problems wear the same clothes:
Dead air. Long pauses, a false start, the ten seconds of nothing at the top. Remove silence shortens pauses rather than deleting them, so the speech keeps its rhythm. The free silence remover does the same job in the browser on a short file.
Structure. If the clip is a recording with chapters, or several takes that belong in one file, split a recording into chapters covers Cut, Split and Join. In the app, Cut and Split show a picture of the sound so you can find the exact spot, and they work on one file at a time while the other tools take a batch.
Do the cutting before the loudness pass if you are changing the length a lot, and after it if you are only trimming a few edges. Cutting changes what the average loudness is measured over, so the order matters when the trim is big.
Put the sound back on the picture
The cleaned file is audio only. Getting it back onto the clip is one command, and it does not touch the picture:
ffmpeg -i clip.mp4 -i audio.m4a -map 0:v -map 1:a -c:v copy -c:a aac -shortest out.mp4-map 0:v takes the video from the original, -map 1:a takes the audio from the cleaned file, and -c:v copy means the picture is copied rather than re-encoded, so there is no quality loss and no long export. -shortest ends the file with whichever stream is shorter.
In an editor, the same job is a drag: put the cleaned file on the timeline under the picture, line up the first frame, and export. Line it up at the start and it stays in sync, because none of the steps above change the length of the audio.
The reduce background noise in a video guide walks the whole loop, extraction to export, if you want it step by step.
Check it before you upload
Three checks, two minutes:
- Listen on a phone speaker. Laptop speakers hide rumble and flatter bad leveling. A phone is what most of the audience is on.
- Measure the loudness with the LUFS meter and compare it to the target for where it is going.
- Watch the whole thing once with headphones, listening for the moment the noise comes back. A model that removes noise in bursts leaves a change you can hear at a seam, even when each part sounds fine alone.
If the level still moves around a lot, the fix is usually the recording rather than the edit. The podcast recording checklist is about mics and rooms, and most of it applies to video just as well.
When cleanup will not save the clip
Being honest about this saves hours. Cleanup cannot:
- Rebuild a clipped recording. If the peaks were cut off while recording, the detail is gone. A limiter makes it quieter, not better.
- Remove music that was mixed into the voice without leaving marks on the voice. That is source separation, a guess by a model, not an edit.
- Fix a long echo. A short room tail is fine. A hall is the voice itself arriving late, and no noise remover puts it back where it started.
- Add detail the microphone never captured. A cheap mic in a loud room gives you a clean recording of a bad take.
In those cases the honest move is another take, or a different room. Cleaning a good take is quick. Rescuing a bad one is not a job that ends.
Where to start, by what you have
| Your situation | Start here |
|---|---|
| A noisy clip and ten minutes | the free noise remover in the browser |
| A whole folder of clips | Gleam Audio on the desktop, batch every tool |
| A clip already in Resolve | noise in DaVinci Resolve |
| A phone clip | noise on an iPhone |
| Music you need gone | background music |
| Wind on an outdoor shoot | wind noise |
| You want the clip silent | remove audio from video |
| You want the order, start to finish | this page, top to bottom |
Gleam Audio is free for personal use, with every tool, up to 20 files a week and no account. A license is for work: Basic is $10, and Max is $20 at 20% off its $25 list price (November 2026). It runs on macOS, Windows and Linux, and the version is 2.2.0.