How to Remove Vocals from a Song

Disclosure: I built StemConsole, one of the tools mentioned. This guide is about the method, whichever tool you use. If you just want the tool, go to the vocal remover.

Four ways to remove vocals, compared

For karaoke, covers and practice, use an AI method. Phase cancellation only makes sense when no AI tool is available to you and you do not mind losing the low end.

Why phase cancellation leaves the song sounding hollow

Most mixes put the lead vocal dead centre, with the same signal in the left and right speakers. Flip the polarity of one side and add them together, and anything identical on both sides cancels out. The problem is that the bass, kick drum and snare are usually centred too, so they disappear with the vocal, while stereo reverb on the vocal stays behind as a ghostly trail. That is why old “karaoke” effects sound thin. AI separation does not depend on panning, so the bass and drums survive.

How to remove vocals, method by method

1. AI vocal remover in the browser

1. Start from the best file you have. A WAV, FLAC or high-bitrate MP3 separates noticeably better than a low-bitrate stream rip.
2. Upload it and choose the two-stem mode. It is usually called Vocal Remover or vocals and instrumental. In StemConsole that is the free vocal remover, which takes MP3, WAV, FLAC, OGG, M4A, AAC and WMA up to 200MB.
3. Listen before you download. Solo the instrumental at the loudest chorus, then solo the vocal to hear what was pulled. Tools with a built-in mixer, StemConsole included, let you do this in the browser.
4. Download the instrumental, the vocal, or both. MP3 is fine for karaoke and practice. Choose WAV for anything you will edit or process further.

2. Ultimate Vocal Remover on your computer

1. Install UVR. It is free, open-source software for Windows, macOS and Linux.
2. Choose the input file and an output folder.
3. Pick a process method and a model. An MDX-Net vocal model gives you a two-stem split; a Demucs model can split out the drums and bass as well.
4. Choose WAV, FLAC or MP3 and start processing. A capable GPU makes a big difference: the project lists an Nvidia GTX 1060 6GB as the minimum for GPU conversions.
5. Try two or three models on the same song and keep the cleanest result.

No computer that can run it? Ultimate Vocal Remover online covers the browser options.

3. Audacity

Audacity has two routes. The free OpenVINO AI plugins, built for Audacity 3.7.x, add a Music Separation effect with a two-stem mode (vocals and instrumental), which works like the AI tools above; stem separation in DAWs covers the setup. The older Vocal Reduction and Isolation effect uses phase cancellation. From Audacity 3.5 it no longer ships with the app and is a separate plugin; see the Audacity effects documentation.

4. Phase cancellation by hand, in Audacity or any DAW

1. Import the stereo song and split it into two mono tracks. In Audacity, choose Split Stereo to Mono from the track menu.
2. Invert the polarity of one side. In Audacity: Effect > Special > Invert.
3. Mix the two tracks down to one. In Audacity: Tracks > Mix > Mix and Render to New Track. Anything identical on both sides, usually the lead vocal, cancels.
4. Export the result. It is mono, and the bass, kick and snare will be thin or gone, for the reason explained above.

A common rescue is to lay a low-passed copy of the original under the result. It brings back some of the bass and kick, along with a little of the vocal’s low end. That is a workaround rather than a fix, which is why AI separation replaced this trick.

How good will it actually sound?

On most modern productions, the instrumental is clean enough to perform over and the vocal is clean enough to remix. On dense, heavily layered or older mixes, you may hear faint artefacts, a shimmer where the vocal used to be, or a little reverb tail left behind. That applies to every AI vocal remover. Listening is how you get the best result: audition the split before you download, and if it falls short, try a better source file or another model.

How to keep the backing vocals

A standard vocal remover can remove lead and backing vocals together, including harmonies you might want to keep for a cover. Some backing vocals may remain, depending on the song and model. You have three options. First, pull the vocal fader down rather than off: a quiet guide vocal often sounds more natural than a hole. Second, use a tool with a lead-and-backing model; Ultimate Vocal Remover has karaoke models built for this, and results vary by song, so try two or three. Third, remove all vocals and sing the harmonies yourself, which often sounds better than an instrumental with leftover lead vocal in it.

How to check the result like an engineer

• Listen on headphones, not laptop speakers. Leftover vocal and artefacts hide on small speakers.
• Check the loudest chorus first. Stacked vocals and reverb are hardest to separate there.
• Solo the vocal stem. If you hear guitar or keys in it, the instrumental has lost some of that part.
• Listen for cymbals. A smeared, underwater hi-hat is the most common artefact of AI separation.
• Start from the best source you have. A WAV or high-bitrate file separates noticeably better than a compressed stream rip.

Is it legal to remove vocals from a song?

Removing vocals for personal practice, study or karaoke at home is generally fine. Publishing or selling an instrumental, cover backing or remix made from a commercial recording needs permission from the rights holders, whatever tool you used. This is general information, not legal advice.

What can you do with the instrumental?

• Karaoke and covers: sing over the real band instead of a MIDI backing. See the karaoke maker for a guide-vocal blend.
• Backing tracks and content: drop a clean instrumental under a video, podcast or live set.
• Remixes and sampling: grab the acapella, flip the beat.
• Practice: mute one part and play along, or isolate the bass or drums free with Basic Split, or the guitar with Full Separation (Pro or one credit), to learn it.

Curious how the separation works under the hood? It comes down to how AI reads a spectrogram. We break down the basics in how AI stem separation works.

Frequently asked questions

Dan Murtagh is a Melbourne mixing and mastering engineer with 20 years making records and a decade teaching audio, and the builder of StemConsole. More about Dan.

More tools