Remove background noise from audio

Drop in a recording and get the voice back without the hiss, hum and room noise behind it. It runs in the browser against our GPUs, so there is nothing to install, and the first minutes are free.

What it removes, and what it keeps

Noise removal here is separation, not muffling. The recording is split into the voice and everything steady underneath it — hiss, electrical hum, air conditioning, fans, room tone, distant traffic — and you get the voice back on its own. What was said or sung stays exactly as it was.

The honest limit: a sudden one-off sound sits in the same instant as the voice. A door slam or a cough in the middle of a sentence is much harder to take out cleanly than a floor of steady noise, and no online tool removes it without a trace.

Where this earns its keep

• Podcast and interview recordings — a kitchen fridge, a laptop fan or a hotel air conditioner sitting under an hour of speech comes off in one pass.
• Video voiceovers — narration recorded at a desk rather than in a booth usually carries room tone. Removing it is the single biggest step toward a produced sound.
• Voice notes and lecture recordings — phone recordings are noisy by nature; the voice inside them is usually recoverable.
• Vocals you plan to work with further — anything a later tool learns from, it learns noise and all. Clean first.

Clean it, or record it again?

If you still can re-record in a quieter spot, that beats any cleanup. Noise removal is for the recording you cannot take again: the interview that already happened, the take that was the good one, the file someone sent you. That is most real recordings.

How it works

Under the hood this is a source-separation model, the same family of models that splits a song into vocals and instrumental. Instead of guessing which frequencies are noise and turning them down — the old approach, which dulls the voice along with the noise — it learns what a voice is and reconstructs it, leaving the noise behind. That is why the result does not sound filtered.

Common questions

Is it free?

Ten free minutes on the noise remover, and separate free minutes on the stem separator and the reverb remover. They do not share a pool. One job costs the length of its audio once, and every output of that job downloads at no extra cost. A subscription removes the limits.

What files can I use?

WAV, MP3, FLAC, M4A and OGG, one or many at a time. You choose the output format on the tool, including lossless WAV or FLAC.

Will it work on speech, not just singing?

Yes. Spoken voice separates the same way singing does.

The noise is louder than the voice. Will it still work?

It will try, but there is less voice to recover. The result is honest to the source: a quiet voice under heavy noise comes back thinner than a clear voice under light noise.

Do my files stay private?

Files are processed for your job and are not shared. See the privacy policy for retention details.

The rest of the toolkit

Noise is one of three things that sit on a recording. The other two have their own tools here:

• Remove reverb — when the problem is the room's echo rather than a noise floor.
• Vocal remover — when the voice is buried under music rather than noise.
• Mix vocals — add space and polish back after cleaning, free in the browser.
• Denoising before voice training — why this matters if the recording is destined for an AI voice model.