Background noise is the fastest way to make a good video feel amateur. Viewers forgive soft focus and shaky framing far more readily than a hiss, a fan or traffic that sits under every word. The good news: if the voice was captured reasonably clearly, the noise can usually be removed after the fact, without touching the picture.
This guide walks through what you can and can't fix in a recording that is already done, and how to do it without making things worse.
First, know what kind of noise you have
Not all noise is the same, and the type decides how well it can be removed.
- Steady noise: fans, air conditioning, a whirring computer, fridge drone, distant traffic. It stays roughly constant, so it is the easiest to separate from a voice.
- Changing noise: keyboard clicks, a dog barking, a door, a passing car, people talking in the next room. Modern speech-focused tools handle much of this too, because they listen for what a voice sounds like rather than for a fixed noise pattern.
- Room sound: echo and reverb from bare walls. This is part of the voice itself, not a separate sound on top of it, and noise removal does not take it away. If your room is echoey, fix it next time with soft furnishings or a closer microphone.
- Distortion and clipping: if the recording was so loud it hit the ceiling, the waveform is damaged. Cleaning can reduce the harshness, but it cannot fully rebuild what was lost.
Listen to a few seconds where nobody is talking. Whatever you hear there is what you want gone.
The usual workflow, and why it is painful
The traditional route is: export the audio track from your editor, open it in an audio program, find a quiet section to "learn" the noise, apply reduction, tweak until the voice stops sounding watery, export, re-import, line the audio back up with the picture, and render the whole video again.
That last step is the costly one. Re-rendering a video re-encodes the picture, which takes time and loses a little quality every time you do it.
A faster way: replace only the audio
A cleaner approach is to leave the video stream exactly as it is and swap in a cleaned audio track. That is how our tool works:
- Drop in the video file (MP4, MOV, WebM or MKV).
- Preview a short section, original against cleaned, to check the result before you commit.
- Download the cleaned file. The picture is copied as-is, without re-encoding, and only the audio track is replaced, kept in sync with the video.
Our speech AI is tuned for voices: it keeps speech and removes the sound around it. Because it is not a noise gate, it does not chop off word endings or pump the background up and down between sentences.
Try it on the video noise remover.
Tips for the best result
- Clean the original file, not a copy that has already been compressed for social media. Every round of compression throws away detail the cleaner could have used.
- Clean before you edit, if you can. Then every cut, fade and caption is built on the clean audio, and you don't need to export twice.
- Pick a loudness target that matches where the video is going. Online video platforms are generally comfortable around −14 LUFS; spoken-word podcasts often sit around −16 LUFS. Matching the target saves you from viewers reaching for the volume.
- Keep music separate when you can. Speech-focused cleaning is designed for voice, so add background music after cleaning rather than before.
- Listen on headphones for the final check. Small artefacts that laptop speakers hide are obvious in earbuds, which is how many people will watch.
When noise removal is not enough
Some recordings are past saving: a voice buried under loud music, a phone in a pocket, or speech that is quieter than the noise around it. Cleaning helps, but it cannot invent detail that was never captured. If a take matters, a quick re-record with the microphone closer to the speaker will always beat any amount of repair.
For everything else (the interview recorded next to an air conditioner, the vlog with traffic behind it, the tutorial with a buzzing laptop fan) removing the noise and keeping the picture untouched is a few minutes' work.