CLEANGRABOpen the tool → ◐

---

Remove background noise from a video

By ShinobiTools Team · Last updated: July 2026

Drop the video in, get the same video back with the noise gone from the audio. No editor, no re-export, no account — and the picture is copied across untouched.

Clean a recording, free →

Why a video editor is usually the long way round

The normal route to clean audio in a video is a chore: open a video editor, detach the audio, export it as a WAV, run it through a noise-removal tool, import the cleaned file, line it up against the original, mute the old track and export the whole video again. Four programs and a full re-export for what is, in the end, a swap of one audio stream.

The re-export is the part that actually costs you something. Every time a video is re-encoded it goes through another round of lossy compression, so fine detail, gradients in the sky and fast motion come out slightly worse than they went in — even when you change nothing about the picture at all.

CleanGrab skips that entirely. It takes the audio out of your file, cleans it, and writes it back into the same container while copying the video stream across untouched. The picture is never decoded and re-encoded, so it cannot degrade. What comes back is your video, at your resolution, at your bitrate, with a quieter soundtrack.

What it removes well from footage

The model is trained on the human voice, so it is at its best when someone is talking and something steady is humming away underneath. In practice that covers most of what ruins home and office footage:

• Air conditioning and fans — the constant whoosh in an office, a bedroom or a server room, including the laptop's own fan spinning up next to the microphone.
• Electrical hum and buzz — the low 50 or 60 Hz drone from a ground loop, a cheap cable or a dimmer switch on the same circuit.
• Traffic and street noise — the low rumble of cars and distant engines behind an outdoor piece to camera.
• Hiss and static — the mush that comes from a cheap microphone, or from a recording made too quietly and then turned up.
• Keyboard clatter and mouse clicks — the typing under a screen recording or tutorial.
• Room tone — the general noise floor that makes footage sound like it was filmed in a hallway.

Full detail on each noise type and which strength suits it is in the guide to removing noise from recordings.

Where it struggles, honestly

Some things are not a noise layer at all, and it is worth knowing which before you upload:

• Heavy echo and reverb. The reflections are the voice, arriving late. Removing them means separating a sound from delayed copies of itself, which is a harder problem than pulling out a steady hum. Light room echo is reduced; a church hall is not.
• Someone else talking in the background. A second voice is speech, and the model is built to preserve speech. It will not reliably pick out only the person facing the camera.
• Music under the dialogue. Background music is treated as something to suppress, and it often comes out mangled rather than removed cleanly. If your video has a music bed, clean the dialogue first and add the music afterwards. To pull a song apart into separate instruments, that is a different job — StemGrab does that one.
• Clipping and distortion. If the audio was recorded far too loud, the peaks were flattened at the moment of recording and that information is simply not in the file. Nothing can reconstruct it.

When almost no clear speech is found, CleanGrab tells you instead of handing back an empty-sounding file. That warning usually means the input was music, near-silent, or so noisy that the voice never really made it in.

Getting the best result out of your footage

Start on Balanced and listen to the before/after in the browser before downloading anything. If a layer of noise is still sitting there, go to Strong. If the voice has started to sound thin, hollow or slightly underwater, that is the model working too hard — drop to Gentle, which keeps more of the original character of the voice.

Consistency matters more than perfection here. If you are cutting several clips together, run them all at the same strength: viewers notice a noise floor that appears and disappears between cuts far more than one that stays quietly present throughout.

If you are cleaning a long recording of a meeting, the Zoom and Teams section of the main guide covers the specific problem of calls where each speaker sits in a different acoustic space.

After the noise is gone

Once the dialogue is clean, a transcript is usually the next thing you want — for subtitles, for a description, or for the searchable text version of a talk. ScribeGrab turns speech into text and subtitle files free, and labels who is speaking when there is more than one voice, which is what makes an interview or panel readable rather than one solid block of text.

Common questions