AI Audio Noise Removal

🧹
Click to upload or drag and drop

Audio files only (Max 10MB)

ℹ️ How it works:
  • Upload your audio file
  • Choose a method (AI is recommended) and process
  • Download clean audio

Remove Background Noise from Audio

Almost every recording made outside a studio has something behind the voice — a fan, traffic, a hiss in the microphone, a room that will not stay quiet. This tool takes that out and leaves the speech. The recommended AI method is a neural denoiser trained to tell a voice apart from everything else, so it copes with uneven noise like passing cars or chatter, not just a steady hum. Four classical filters are there as well if you want a lighter touch. No sign-up and no watermarks.

How to Remove Background Noise from Audio

  1. Click the upload area above or drag and drop your audio file (up to 10 MB).
  2. Choose a method — AI Noise Removal is the right answer for most recordings.
  3. Listen to the preview, then process and download the cleaned file.

The Five Methods

  • AI Noise Removal – a neural denoiser, and the best all-rounder by a clear margin. It is the only method here that handles noise which changes from moment to moment: chatter, traffic, a door, a dog. Start here.
  • Spectral Gate – learns what the steady background sounds like and subtracts it. Very good on a constant fan, air conditioner or mains hum; less good when the noise moves around.
  • Wiener Filter – a fast, general-purpose reduction. A reasonable middle option when the AI result feels too aggressive.
  • Median Filter – aimed at short, sharp intrusions rather than a constant background: clicks, pops and crackle, of the kind you get from vinyl or a bad connection.
  • HPSS + Median – separates the sustained parts of a sound from the percussive ones before filtering, which makes it the one worth trying on music with vocals.

What It Can and Cannot Remove

It can remove noise from audio whenever that noise sits behind the voice and is unlike it — hiss, hum, fans, air conditioning, traffic, rain, room tone, general street noise. Removing hiss from audio is the easiest case of all, because hiss is constant and sounds nothing like speech. It cannot pull apart two things of the same kind. Background music, other people talking, and echo or reverb are not noise in this sense — they are structured sound sharing the same space as the voice, and no filter here will separate them out. Echo in particular is your own voice arriving late, which is why removing it is a different and much harder problem.

Getting the Best Result

  • The voice needs to be louder than the noise. If they are level, there is not enough to work with.
  • Try the AI method first, and only reach for the classical filters if it takes too much with it.
  • Denoise before anything else. Clean first, then set the level and tone with Enhance Audio — doing it the other way round amplifies the noise before you remove it.
  • A better source cleans up better. Noise reduction recovers what is there; it does not invent detail that was never recorded.

When You Might Need This

  • An interview recorded in a café or somewhere with traffic outside
  • A voice memo with a fan, fridge or laptop humming underneath
  • A lecture or meeting recording made on a phone at the back of a room
  • A podcast take that is otherwise good but has hiss on it
  • A video voiceover that needs to sound like it was recorded properly
  • An old tape or vinyl transfer with crackle and surface noise
  • Speech you are about to transcribe, where noise costs you accuracy

Frequently Asked Questions

Upload it, leave the method on AI Noise Removal, and listen to the preview. For most recordings that is the whole job. Only if the result sounds over-processed — thin, or with the voice cutting in and out — is it worth trying one of the classical filters, which take less away. There is nothing to install and no account needed.

Steady background sound comes out most reliably — tape or microphone hiss, mains hum, fans, air conditioning, computer noise, general room tone. The AI method additionally handles noise that comes and goes: traffic, wind, rain, distant chatter, a door closing. What it cannot do is separate two things of the same kind, which is why background music and other people's speech stay put.

The classical filters work statistically: they measure what the background looks like and subtract it, which works well as long as the background stays roughly the same throughout. The AI method was trained on a great many examples of speech mixed with noise, so it is deciding moment by moment which parts are the voice — that is why it survives noise that changes. The trade-off is that a statistical filter is predictable and gentle, while the AI occasionally takes a quiet consonant with it.

Every method here works on a single combined channel, so a stereo file is mixed down to mono before it is cleaned and comes back that way. For speech that costs you nothing — a voice recorded on a phone or a headset is effectively mono already. It does matter for music or anything with a deliberate stereo image, so keep your original if the stereo is worth preserving.

No, and it is worth knowing why before you try. Noise reduction works by recognising what a voice looks like and keeping it. Background music and other speech have the same structure as the sound you want to keep, so nothing here can tell them apart from your speaker. Echo is the hardest of the three, because it is literally your own voice arriving a fraction of a second late. Those jobs need source separation or de-reverberation, which is a different kind of tool.

That is the sound of too much being removed, and it usually means the noise was nearly as loud as the voice. When a filter cannot cleanly tell them apart it takes parts of the speech with the noise, leaving a hollow, watery quality. Try a classical filter instead of the AI — Wiener or Spectral Gate are gentler — and accept a little remaining hiss. A recording with some background left in is easier to listen to than one that has been scrubbed hollow.

This page removes something that should not be in the recording. Enhance Audio shapes the sound that is already there — its level, its tone, its balance — and cannot separate a voice from its background. If the recording is noisy, start here. If it is quiet, muffled, harsh or uneven, that is the other tool. Using both, clean here first and set the level afterwards.

You can upload MP3, WAV, OGG, FLAC, AAC and M4A, up to 10 MB. If you need a different format afterwards, Convert Audio will change it.

Related Audio Tools