Noise reduction for voice recordings

Reduce background noise in a voice recording in your browser, without picking a noise profile first: an AI speech model removes steady and changing noise. Free, no upload.

AudioRinse · voice cleanerOn-device · 0 B uploaded

Drop audio or video here

MP3 · WAV · M4A · MP4 · MOV · WEBM — up to 60 min. Nothing is uploaded.

How it works

Drop your recording onto the box above. MP3, WAV, M4A and most video files work.
The speech model reduces the noise on your device. No noise sample to select first: a one-minute clip takes a few seconds.
Switch between Original and Rinsed, and set the strength: lower keeps more of the room, higher takes more noise out.
Download the result as MP3, M4A or WAV.

Before / after

Hear it before you drop anything

“After” is AudioRinse at its default settings. The voice is synthetic, so we can share it freely.

A desk fan was added to this recording.

Noise reduction without a noise profile

Classic noise reduction, like Audacity's Noise Reduction effect, works in two steps: you select a stretch where only the noise plays, the effect learns that noise's profile, and it then subtracts that profile from the whole recording. That works well on noise that never changes, like a steady fan or tape hiss, and it needs a few seconds of pure noise to learn from.

AudioRinse reduces noise differently. Its speech model, DeepFilterNet 3, has learned what a voice sounds like, so it needs no sample of the noise: every 10 milliseconds it keeps what is voice and turns down what isn't. A noise that changes or comes and goes, like traffic, a keyboard or a café, is reduced as well as the steady hum under it.

Profile-based or AI noise reduction

Two ways of reducing noise in a voice recording
	Noise profile (e.g. Audacity)	AudioRinse
What you do first	Select a few seconds of noise alone	Nothing: drop the file
Steady noise (fan, hiss, hum)	Yes	Yes
Changing noise (traffic, chatter, keyboard)	No: it isn't in the profile	Yes
Music	Yes, with care	No: it's built for speech and treats instruments as noise
Where it runs	On your computer (desktop app)	On your computer (in the browser)

Getting a natural result

Set the strength with headphones on. If the voice starts to sound thin or watery, lower it; if you still hear noise between sentences, raise it.
Reduce the noise before you cut or add music, so the background stays even across your edits.
Hum at 50 or 60 Hz from a cable or a ground loop has its own tool, the hum remover, which takes the hum out without touching the voice.
For a voice that is also quiet or uneven, the audio enhancer reduces the noise and evens the loudness in one step.

Questions

Is my recording uploaded anywhere?

No. AudioRinse runs the cleaning model inside your browser, on your own computer. Your file never leaves your device, so there's no upload to wait for and nothing stored on a server.

Do I need to select a noise sample first?

No. Unlike profile-based noise reduction, the model doesn't need to hear the noise on its own: it knows what speech sounds like and reduces everything else, so there's nothing to select.

Is it the same as the background noise remover?

It's the same speech model as the background noise remover. This page explains how it compares with profile-based noise reduction, for people coming from Audacity or a similar editor.

Can I reduce the noise only a little?

Yes. The strength slider goes from a light touch to the full reduction, and the default (70%) leaves a little of the room so the voice keeps sounding natural. You compare with the original before downloading.

Does it work on music?

No. It's built for speech, so in a song it would treat the instruments as noise. For a song, the vocal isolator or the stem splitter separate the parts instead.

Is it free?

Cleaning and listening are free and unlimited. You get 3 free downloads a month for audio files up to 10 minutes (1 without an account). Pro adds unlimited downloads, files up to 1 hour and video.