How to Remove Vocals from a Song

The short answer

Load a track into WaveCraft, run vocal separation, and you get two files: vocals and instrumental. The instrumental is the karaoke/accompaniment version; the vocals file is a solo. Processing runs entirely on your device — the audio never leaves your phone, and it works with no connection at all.

What "removing vocals" actually means

There are two different things people mean, and they are not equally possible:

True separation (what this does). A neural network listens to the whole mix and estimates a vocal mask — a per-frequency, per-moment estimate of "how much of this is the singer". It reconstructs two signals from that estimate. This works on full commercial mixes.

Channel cancellation (what most cheap tricks do). Vocals are assumed to sit dead center, so the tool subtracts one channel from the other. This only works on old stereo recordings made with the singer exactly center-panned, and it destroys anything else that lives in the middle — kick drum, bass, and the vocal itself.

If you have ever tried the channel trick and gotten a hollow, echoey result with the bass missing, that is not a settings problem. The technique simply cannot preserve what it cancels. A separation model reconstructs rather than subtracts, which is why it survives a modern mix.

How to do it

1. Open WaveCraft and start a new project.
2. Import the track you want to process.
3. Select the track, then choose Separate Vocals from the processing menu.
4. Wait for the pass to finish. You now have two tracks in your timeline.
5. Mute or solo whichever one you need — instrumental for karaoke, vocals for a remix or a cover rehearsal.

If you are starting from a video rather than an audio file, import the video directly — WaveCraft pulls a clean audio track out of it before you separate.

What survives separation, and what doesn't

Separation is estimation, so quality depends on the source. Honest expectations:

Clean results: lead vocals with moderate reverb, dense mastered pop, most studio recordings, anything where the vocal sits clearly above the mix.

Harder results: heavy vocal distortion or screaming, very dense mixes with several singers overlapping, and extremely compressed masters where the vocal and instruments occupy the same frequencies.

You cannot get a different take. The separated vocal is a reconstruction, so it carries the same timing and the same words. It will not be a cleaner version than what was recorded — it is the performance, minus the other instruments.

Why the two outputs add back up to the original

A detail that matters if you plan to work with both files: WaveCraft's separation is sample-exact complementary. Vocals plus instrumental reproduces the original mix precisely — nothing is silently dropped, so you can layer the instrumental under anything and still recover the original by summing.

This is worth knowing because it means separation is reversible, not lossy. Many tools discard energy in the reconstruction; if you ever export both stems and find the mix sounds thinner than the original, that is a flaw in the tool rather than in your project.

One legitimate use worth knowing

Separation is also how you build a vocal-only practice track for a song you are learning, or isolate a harmony you want to study. Play the instrumental at half speed, record yourself over it, and compare timing — which is far easier to do when the other parts are not competing for your attention.

Where to do it

WaveCraft runs the separation on-device, so it works on a plane, on a train, or anywhere without signal — and your audio is never uploaded to a server. Exports are available as WAV, AAC, ALAC or FLAC, up to 32-bit float, so the stems drop into any DAW or editing app without a second conversion step.