How to remove background noise from audio
- Drop your file. Drop an audio or video file (up to 10 minutes) onto the tool, or press Try a sample.
- Pick the settings. Noise removal strength, hum removal (Auto finds 50 or 60 Hz) and a loudness target: Podcast (-16 LUFS), YouTube (-14 LUFS) or Keep.
- Clean and compare. Press Clean audio. The first run downloads the noise model once (about 17 MB). Then switch between Before and After with A and B while it plays.
- Download. Save a WAV, a compressed M4A (or Ogg in Firefox), or for video, the same video with the cleaned audio.
What it does
DeepFilterNet3, an open speech-enhancement model, removes fans, traffic, hiss and room noise while keeping the voice natural.
Finds 50 or 60 Hz mains hum and notches it out with its harmonics. More on hum.
Measures integrated loudness (ITU-R BS.1770) and sets -16 LUFS for podcasts or -14 LUFS for YouTube, with peaks held under -1 dBFS.
How the noise removal works
Classic noise reduction (Audacity's Noise Reduction, for example) learns a "noise profile" from a few seconds of silence and subtracts it everywhere. That works for a steady hiss but smears changing noise and leaves a watery, "crispy" sound when pushed. This tool uses DeepFilterNet3, a neural network trained to tell speech from noise frame by frame, so it needs no noise sample and copes with noise that changes.
The Noise removal slider sets how far the model may turn any part of the sound down. Around 80 suits most recordings. If the voice starts to sound thin or the ends of words get clipped, lower it; if noise is still audible between sentences, raise it to 100.
- Record as close to the microphone as you can. Noise removal works best when the voice is clearly louder than the noise.
- Turn on Match levels before you compare, so the cleaned version doesn't sound better just because it's louder.
- Keep the WAV if you will edit further; use M4A to share.
Remove background noise from a video
Drop an MP4, MOV or WebM. Your browser reads the audio track out of the video, the tool cleans it, and the video is rebuilt: the original picture is copied over packet by packet (no re-encoding, so no quality loss, and it's quick), and the cleaned audio is encoded as AAC (or Opus for WebM) alongside it. If your browser can't rebuild a particular video (some MOV codecs, for example), the tool says so; download the cleaned WAV instead and replace the audio in your editor. It lines up from the first frame.
Podcasts and loudness
For a podcast, leave hum removal on Auto and choose Podcast (-16 LUFS): one pass removes the hum, the noise, and brings the episode to the level Apple Podcasts asks for (-16 LUFS, true peaks under -1 dBFS). For YouTube choose -14 LUFS, the level YouTube and Spotify play back at. To change only the loudness, set Noise removal to 0 and hum to Off.
| Target | Use it for |
|---|---|
| -16 LUFS | Podcasts (Apple Podcasts: -16 ±1, peaks ≤ -1 dBFS) |
| -14 LUFS | YouTube, Spotify and most streaming |
| Keep | Tracks you will mix later: set loudness on the final mix |
Then caption it
Clean audio also transcribes better. Once it's clean, make captions or get the text in your browser too. Working with music rather than speech? The vocal remover splits a song into vocals and instrumental. No recording at all? Text to speech reads a script in a natural voice, and the voicemail greeting generator makes a business phone greeting.
Questions
How do I remove background noise from my audio?
Drop the file onto the tool above, leave Noise removal around 80 and press Clean audio. Listen to Before and After (press A and B), adjust the slider if needed, and download the WAV or M4A.
Is this noise remover free with no sign-up?
Yes. The cleanup runs on your own computer or phone, so each file costs us nothing and you pay nothing. Files up to 10 minutes are free, with no watermark and no account.
How do I remove background noise from a video but keep the voice?
Drop the video onto the tool, press Clean audio, listen to Before and After, then press Video with clean audio. You get the same video with the cleaned sound; the picture is copied, not re-encoded.
How do I remove noise with the least damage to the voice?
Use the lowest strength that clears the noise: try 60, then 80. Over-processing is what causes the thin, underwater or crispy sound. Comparing with Match levels on makes the trade-off easy to hear.
Can it remove background noise that is louder than the voice?
Only partly. When the noise is louder than the speech, the model keeps what it can recognise as voice, but words buried under the noise can't be recovered. Re-recording closer to the microphone is the real fix.
How loud should a podcast be?
Around -16 LUFS integrated, with peaks under -1 dBFS. That is Apple Podcasts' recommendation, and most podcast hosts and apps work to the same level. Choose Podcast (-16 LUFS) in the tool.
What is the difference between normalizing peak and loudness?
Peak normalization raises the file until its loudest sample hits a level; one loud cough decides the result. Loudness normalization (LUFS) uses how loud the whole file sounds, which is what listeners and platforms care about. This tool does loudness.
How do I remove background noise in CapCut or Audacity?
Both have a built-in noise reducer: CapCut's "Reduce noise" switch under Audio, and Audacity's Effect › Noise Removal and Repair › Noise Reduction (select a few seconds of noise only, Get Noise Profile, then apply to everything). If either leaves the voice watery, clean the audio here first and import the WAV.
Can I remove echo from my recording?
Only a little. The noise model reduces some room sound, but strong echo is part of the voice signal and can't be cleanly taken out. Record closer to the microphone in a smaller, softer room.
Which formats and browsers work?
Audio: MP3, WAV, M4A, AAC, OGG, Opus, FLAC. Video: MP4, MOV, M4V and WebM rebuild in current Chrome, Edge and Safari. Compressed audio export is M4A (AAC) in Chrome, Edge and Safari and Ogg (Opus) in Firefox. A 10-minute file takes about one to three minutes on a laptop.
Why is the result mono?
The noise model works on one channel, so the tool cleans a mono mix of your file and exports it on both left and right. For a voice recording that is what you want; stereo music beds are not kept.
Is my audio uploaded anywhere?
No. The file is decoded, cleaned and encoded in this browser tab. The only downloads are the noise model and its WebAssembly engine, which come from this site and stay in your browser's cache. You can disconnect from the internet after the model has loaded and it still works.
What can't it fix?
It is built for speech. It takes out steady and changing background sound: fans, air conditioning, traffic, room tone, hiss, hum and much of the clatter around a voice. It does not separate two people talking over each other, it treats background music as noise, and it cannot rebuild a voice that is clipped or buried under louder noise. Heavy echo is reduced only a little.
Privacy
Your audio and video never leave your device: the file is decoded, cleaned and encoded in this browser tab. The noise model and its engine are downloaded from this site and cached by your browser. We don't use cookies or analytics on this page.
Credits
Noise removal: DeepFilterNet3 by Hendrik Schröter and colleagues (MIT or Apache-2.0), compiled to WebAssembly. Video and audio encoding: Mediabunny (MPL-2.0) on your browser's WebCodecs. Loudness measurement follows ITU-R BS.1770-4. The sample voice is synthetic (Kokoro TTS, Apache-2.0) with noise added by us.