Remove music from a video

Remove music while keeping speech

1. Choose a video. MusicWipe extracts its soundtrack on your device and uploads the first 30 seconds of audio for a free preview. The picture stays on your device.
2. Compare the original and cleaned sound. Listen to the speech and check for leftover music or missing sounds. Move Ambience toward Voice only for less background sound, or toward Full ambience to include more of the effects track. Changing the mix does not use another preview.
3. Clean and save the full video if the preview works for you. MusicWipe replaces the soundtrack and copies the picture without re-encoding it. The full clean uses minutes; the button shows the cost before you start.

What you get back

The same video in the format you opened (MP4, MOV, M4V, MKV or WebM), with a new soundtrack. The picture is copied without re-encoding. If you choose lossless audio, the export uses MOV. This is the synthetic toast from the listening demo above after MusicWipe:

MusicWipe separates the soundtrack into dialogue, music, and effects and ambience. It builds the replacement soundtrack from the dialogue plus the amount of effects and ambience you choose. This can retain room sound, but it cannot guarantee that every effect survives. In the second synthetic demo above, much of the chopping and sizzle is removed with the guitar.

Keep the room sound, or don’t

The Ambience slider decides how much of the effects and ambience track goes back in, from Voice only to Full ambience. In the toast demo, the second and third players use the same separation with only that slider moved.

For interviews and voice-overs, try a low setting if speech is what matters most. For footage where the surroundings matter, compare higher settings against the original. A higher setting may bring back some music along with the room sound. Moving the slider happens on your device and never uses minutes.

Save the separate tracks

To finish in an editor instead, tick Also save stems as FLAC when you export. You get three lossless files, dialogue, music, and effects and ambience, each the same length as the soundtrack, so they line up at zero. The most common use is laying a track you’re licensed to use under the original dialogue and effects.

Where it struggles

Separation is harder when music is louder than the voices, when someone is singing, or when audio has been compressed many times. Music can remain, and wanted sounds can be removed. The cooking demo above shows the latter. Use the free preview to check your own footage before using minutes on a full clean. Full videos are billed by the second.

No tool can promise a platform won’t flag a video, and removing a song doesn’t change who owns the footage. With Pro, the copyright check tells you which known songs were in the original and whether each one is still detectable after cleaning.

How we tested it

On September 26, 2026 we ran the model MusicWipe uses (checkpoint bandit_v2_real1, stereo processing, as in production) on our own test set: 95 one-minute clips, each a commercial song mixed 3 to 12 dB under dialogue and sound effects from the DnR dataset. The songs are by artists we kept out of training entirely.

• Music left behind: the median was 17.05 dB below the original music, so what’s left is about 17 dB quieter.
• Dialogue: the median signal-to-distortion ratio of the dialogue track was 16.24 dB. In plain terms, the speech that comes out is close to the speech that went in, with errors about 16 dB quieter than the voice.
• Could a song matcher still find the song? We ran audfprint, an open-source audio fingerprinting tool, against the 95 songs. It identified the song in 87 of the 95 original mixes, in 3 of the 95 cleaned soundtracks, and in none of the 95 true dialogue tracks.

This is our own evaluation, not an independent benchmark, and audfprint is not the matcher YouTube, Instagram or TikTok use.