Too much noise behind your voice?

Drop a video or a recording. The fan, the street or the hum is turned down, the picture stays exactly as it was, and you see how far the noise went down.

Runs in your browser. Nothing is uploaded, no signup.

Background noise removerRead on this device

Watch an example

13 seconds · a made-up example

Clips made in the studio from your own recordings are cut and levelled to −14 LUFS for you.Join the beta

How it works

Short and checkable: every limit links to the platform's own page, read on 28 Oct 2026.

  1. Voice: a network made for speech

    RNNoise (Xiph, BSD licence) runs in your tab, 10 ms of sound at a time at 48 kHz. It keeps what sounds like a voice and removes the rest — traffic, keys, a room full of people, and music too. Its model (4.8 MB) is downloaded once.

  2. Steady noise: learnt from your quietest moments

    Spectral gating, the method of the open-source noisereduce: the noise of each frequency is learnt from the quietest tenth of the file, and whatever doesn't rise above it is turned down by 12 or 24 dB. Fans, air conditioning, hum and hiss go; music and the room stay.

  3. The picture isn't touched

    Only the sound is encoded again (AAC). The video's own picture is copied packet by packet into the new file, so it keeps its quality and the work takes seconds, not minutes.

  4. Measured on your file

    The noise floor (the quietest tenth of the sound) is measured before and after, and so are the loud parts, so you see what changed instead of a promise.

What it doesn't do

  • Voice mode removes music and sound effects with the noise: use Steady noise when they matter.
  • Steady noise needs a few quiet moments to learn the noise, and doesn't touch sudden noises.
  • Up to 5 minutes per file, so it fits in a phone's memory.
  • Nothing is sent or saved. The file stays in this tab.

Questions

How do I remove background noise from a video for free?

Drop the video here, pick Voice (someone talking) or Steady noise (fan, hum, hiss) and press Remove the noise. The sound is cleaned in your browser and put back in the same video; the picture is copied as it is, so it loses nothing.

Voice or Steady noise: which one?

Voice uses RNNoise, a small neural network made for speech: it removes almost any noise, but music too. Steady noise learns the noise from the quietest moments and turns down only what stays under it: it keeps music, but not sudden noises like keys or traffic.

Is my video uploaded?

No. The sound is decoded, cleaned and written back in this tab. Only the Voice mode downloads its model (4.8 MB) once, from jsDelivr.

Sources

  1. 1Jean-Marc Valin, Xiph.Org — RNNoise: learning noise suppression (BSD licence)run in the browser through Shiguredo's rnnoise-wasm, Apache-2.0 · read 28 Oct 2026
  2. 2Tim Sainburg — noisereduce — spectral gating noise reductionthe method the "steady noise" mode follows · read 28 Oct 2026

More free tools

All 25 tools

Clean sound on every video.

Vertiscale's Clips mode turns your own long recordings into shorts — pauses and fillers cut, levelled to −14 LUFS — and every video it makes is levelled the same way.

Join the private beta Free during the beta, no card.