Clean up your audio in one pass

Take out hiss, hum and room noise, level it to -14 LUFS for YouTube and Spotify or -16 for podcasts, and pull the voice away from the music. Audio or video in; the same video back with clean sound, or the audio alone.

or paste a link

Processed on our own Mac, never uploaded elsewhere

How it works

  1. 1

    Add a file

    Drop an audio or video file, or paste a TikTok, Instagram or YouTube link.

  2. 2

    Pick the fixes

    Noise reduction and its strength, a loudness target, voice or music only if you need it. Preview 20 seconds first.

  3. 3

    Compare and download

    Flip between before and after, check the loudness readings, and download. Files are deleted after an hour.

Features

Noise reduction

ffmpeg's FFT denoiser tuned to each file's own background noise, or the RNNoise speech model. Adjustable strength and a rumble filter.

Platform loudness

Two-pass loudnorm to -14 LUFS (YouTube, Spotify), -16 (podcasts) or your own target, with before and after readings.

Voice or music only

Demucs pulls the vocals away from the backing track: dialogue without the music, or the music bed without the talking.

A/B preview

Hear 20 seconds cleaned before committing, and switch between original and cleaned on every result.

Video in, video out

The picture is copied untouched and only the soundtrack is replaced. Or take the audio alone as WAV, MP3, M4A or FLAC.

Stays in-house

Everything runs on our own Mac with ffmpeg and open-source models. No cloud APIs, no ads, no tracking.

FAQ

Standard or Voice noise reduction?

Standard works on steady noise (hiss, hum, air-con, a fan) and leaves music alone. Voice uses a model trained on speech and also handles changing noise like traffic or chatter, but it can thin out music, so use it for talking. Start around 50 and preview: too strong sounds watery.

Which loudness should I pick?

-14 LUFS for YouTube, Spotify, TikTok and Instagram (they turn louder audio down anyway). -16 LUFS for podcasts (Apple's spec). Broadcast is -23. Loudness is measured over the whole file, so the balance between quiet and loud parts stays as it was.

Why does separation take so long?

Demucs is a neural network running on our Mac's processor: a minute or two for a short clip, about five for 15 minutes. Separation jobs run one at a time in their own queue and your card shows its place in line. Keep the page open and it will finish.

Which files and links work?

MP3, WAV, M4A, AAC, FLAC, OGG and Opus audio, MP4, MOV, M4V, WebM and MKV video, up to 500 MB and 15 minutes, and public TikTok, Instagram and YouTube links. Anything else is refused.

What happens to my file?

It's processed on our own Mac and deleted, with the results, an hour later. Nothing is sent to any outside service.