How to remove background music in your video editor, and what doesn't work

By the time a video is exported, the music and the speech are one waveform. Turning the volume down, EQ and noise reduction all damage the voices before they get rid of the song. What works is separating the soundtrack into its parts and rebuilding it without the music, and you can do that inside the editor you already use.

I make MusicWipe, so it’s one of the methods below.

The editor workflow

The workflow is the same in most editors: detach the audio, clean it, attach it back.

1. Export the audio. In your editor, export the timeline’s audio as WAV or FLAC, not MP3, so no quality is lost before separation. Note where it starts.
2. Separate it. Run the audio through a tool that returns the voice, and ideally the sound effects, as separate files.
3. Put it back. Import the separated voice (and effects) into the project at the same start point as the original audio, then mute or delete the original audio track.
4. Check the sync. Play the start and the end of the clip. If the separated file is a few milliseconds longer or shorter, lips drift out of sync by the end.
5. Export the video. Export with settings that match the original. Most editors re-encode the picture at this step.

It gives you full control over the mix, and it works with any separation tool. It’s also slower, there are more places for the sync to slip, and the last step costs you a generation of picture quality unless your editor can pass the video through untouched.

What the separation tool returns matters most. Many return only the voice, and then the room sound, footsteps and applause go too. Look for one that gives you a separate effects or ambience track.

What doesn’t work

• Turning the music down. There’s no music track to turn down in a finished video. Lowering the volume lowers the voices with it.
• EQ. Music and speech share the same frequencies, roughly 100 Hz to 8 kHz. Cut enough to lose the song and the voice sounds like a phone call.
• Noise reduction. It learns a steady noise, like hiss or hum, and subtracts it. Music changes every second, so noise reduction removes some of it, smears the rest, and leaves the voice warbly.
• Center channel removal, the old karaoke trick. It cancels whatever is in the middle of a stereo mix. In a song that’s the singer. In a video it’s usually the person talking, which is the opposite of what you want, and it does nothing on a mono recording.

The shortcut: MusicWipe

MusicWipe does the export, separation and re-import in one place. It takes the fewest steps, because you get the finished video back instead of an audio file to line up again, and with a minutes pack or a plan the picture isn’t re-encoded. It also keeps the effects and ambience as their own track, with a slider for how much goes back in.

Here are 30 seconds from a demo clip we made for testing: a text-to-speech voice with generated music and room sound. The MusicWipe versions are the production model’s output, not edited.

The first 30 seconds of any video are free to try in the browser, with no account.

Check the result before you publish

Across 95 one-minute test clips of commercial songs mixed under dialogue, the median amount of music left after MusicWipe was about 17 dB below the original, and an audio fingerprinting tool still identified the song in 3 of the 95 cleaned clips (it identified 87 of the 95 originals). The details are in how we tested it . Whatever tool you use, listen before you post:

• Listen to the loudest part of the song, not just the start. The chorus is where traces survive.
• Use headphones. Laptop speakers hide faint music.
• Lower the effects or ambience track if you still hear the song. Leftover music tends to land there.
• Listen for missing sounds. Steady, rhythmic sounds such as chopping can be mistaken for music and removed.

A note on copyright

Removing the song can clear a claim, but no tool can promise a platform won’t match what’s left. Only process footage you have the rights to use. On YouTube, Studio’s Erase song tool works on the claimed part of a video that’s already uploaded.