Remove background noise from audio
Fans, traffic, AC or hiss behind your voice? fix.fm finds the noise, explains it in plain words and removes it. Free, and nothing is uploaded.
Never uploaded. Everything runs in your browser.
What's happening
Background noise is any steady sound that keeps going under the voice: a fan, the fridge, air conditioning, traffic outside, or the hiss a cheap mic adds by itself when the gain is turned up. You notice it most in the pauses, when the voice stops and the sound of the room is all that's left. It makes a recording tiring to listen to, even when every word is clear.
fix.fm compares quiet moments with louder speech. Our check flags gaps under 30 dB, with the strongest warning under 12 dB; these are tool thresholds, not universal hearing limits. If there is no useful pause, the estimate may be uncertain. Compare noise, echo and hum before choosing the repair.
Want less noise next time? Test your mic on tonal.fm, our sister site, and listen for hiss in the quiet before you hit record.
How far the noise reduction goes
The denoiser is built for steady sound under a voice: hiss, a fan, a hum's broadband cousin. It is not built for another voice, and it is not free when the noise gets close to the speech. The four cases below are the same clean reading with a background added in software, so we know what was there before and can measure what came out and what it cost.
One clean reading with four backgrounds added in software, then run through this page's fixer at full amount. Every clip is matched to -23 LUFS so only the background differs.
Clean reading
The reference. Listen to the pauses: there is almost nothing in them.
Light hiss, 30 dB under
A cheap mic with the gain up. You hear it in the pauses and forget it during the words.
Light hiss, after the fix
The pauses are quieter than the clean reading's. The words are untouched.
A fan, 10 dB under
A fan or air conditioner in the room. The voice is clear, the pauses are full of it.
A fan, after the fix
The fan is gone from the pauses and from under the words. Compare the s sounds with the clean reading.
Nearly as loud as the voice, 3 dB under
Recorded next to the machine. You can follow the words, but it is work.
Nearly as loud, after the fix
The background is gone and the check now reads it as clean. Listen to the consonants and the ends of words: that is what it cost.
Someone else talking, 12 dB under
A second reader in the room. The check does not flag this file at all.
Someone else talking, after the fix
The other voice is a little quieter and still perfectly audible. It is speech, which is what the denoiser is built to keep.
Public-domain LibriVox voices. The hiss and the fan are white and pink noise, not a recorded fan; the second talker is another LibriVox reader mixed in. So the clean version really is what was there before the noise.
| Case | The check | In the pauses | The words | The consonants |
|---|---|---|---|---|
| Clean reading | SNR 64 dB, fine | -71.6 dBFS | -21.5 dBFS | -50.1 dBFS |
| Light hiss | 34 to 66 dB, fine to fine | -56.4 to -76.3 (-19.9) | -21.6 (-0.1 vs clean) | -50.4 (-0.3 vs clean) |
| A fan | 17 to 65 dB, moderate to fine | -36.8 to -68.1 (-31.3) | -21.8 (-0.3 vs clean) | -51.7 (-1.6 vs clean) |
| Nearly as loud as the voice | 11 to 69 dB, severe to fine | -29.8 to -65.6 (-35.8) | -22 (-0.5 vs clean) | -54.1 (-4 vs clean) |
| Someone else talking | 53 to 62 dB, fine to fine | -35.5 to -41.5 (-6) | -22.4 (-0.9 vs clean) | -55.4 (-5.3 vs clean) |
Read the pauses and the consonants separately. Hiss and the fan come out completely and the consonants stay within a couple of dB of clean. The 3 dB case also comes out completely by the check's number, from SNR 11 to 69 dB, but the consonants lose 4 dB against the clean reading, and you can hear it: the voice is cleaner and duller, with a slight swirl under the words. That is the trade every denoiser makes when the noise is near the voice, and the amount control exists so you can stop before it. What the artifacts sound like goes through them one by one.
The other talker is the case to know about before you start. The check reads that file as fine, SNR 53 dB, because there are moments when both readers are quiet and it measures the stillest moments. The denoiser then takes 6 dB off the other voice and keeps the rest, because a voice is exactly what it is trained to keep. No setting on this page removes a second speaker; that is an editing job, and for the kind of recording where one voice has to be pulled out of a room, Adobe's Enhance Speech is the honest comparison. For the gate-shaped alternative, gate vs noise reduction shows why a gate leaves the fan under every word. How we test has the pass marks behind these grades.
How it works here
- 01
Listens to the pauses
It measures the noise between your words, so it knows what to take out and what to keep.
- 02
Turns the noise down
A small speech model, running in this tab, lowers everything that isn't voice. AMOUNT sets how much.
- 03
Stays on your computer
Your file is never uploaded. No account, no queue, nothing to wait for but your own computer.
If you'd rather do it in your editor
Questions
Is it really free?
Yes. The noise fix runs in your browser, so there's no server bill to pass on to you. There's no sign-up and no watermark.
Is my audio uploaded anywhere?
No. The file is read and fixed inside this browser tab, and the audio is never sent to us or anyone else. You can watch the network tab in your browser's developer tools while it runs to check.
What kind of noise can it remove?
Steady sounds work best: fans, air conditioning, computer hum, room tone, traffic and mic hiss. Sounds that come and go, like a dog barking, get quieter but may not disappear. Other people talking in the background count as voice, so they stay.
Will it make my voice sound robotic?
It can if the noise was very loud and you take all of it out. Start at the default AMOUNT of 80%, flip between before and after, and turn it down if the voice starts to sound thin or watery.
Can I remove background noise from a video?
Yes. Drop in an MP4 and it cleans the sound. You download the fixed audio as WAV or MP3, then put it back under the video in your editor.
Does it work on music?
Not well. It's tuned for voices, so it treats instruments as noise and damages them. Use it on speech: podcasts, voiceovers, interviews, lectures and calls.