Center-cancel vocal removal: the algebra, the casualties, and the honest fix

You need an instrumental tonight — for karaoke, for a practice session, for a rough edit — and all you have is the finished song. The center-cancel trick is the oldest answer there is: no AI, no upload, one subtraction, done in the time it takes to decode the file.

This page runs that trick and then does something most vocal removers won't: it shows you precisely what the trick destroyed along the way, and gives you a gentler version built for practicing.

§How to remove vocals from a song in the browser

  1. Drop a stereo fileWAV, MP3, M4A, OGG, or FLAC. Everything runs on your device — the song never uploads.
  2. Flip Original to Instrumental while it playsThe A/B is the whole lesson: whatever sat dead-center drops out, everything panned or stereo survives.
  3. Decide about the bassKick and bass live in the center too. Keep-the-bass adds the low end back under 120 Hz so the result still moves.
  4. Pick the version you needFull cancel for karaoke, the −6 or −12 dB practice mix for rehearsal, or the batch mode for a whole setlist.
  5. Download the WAVUncompressed 16-bit, rendered locally.

01One subtraction, no intelligence

A lead vocal is normally recorded once and mixed at equal level into the left and right channels, which is what puts it in the phantom center. Center-cancel computes L minus R: anything identical in both channels subtracts to exactly zero, and anything different survives. That is the entire algorithm — it removes a position in the stereo field, and it has no idea what a voice is.

Everything that makes the result usable or unusable follows from that one line of algebra. Panned guitars survive untouched. A doubled vocal panned left and right survives too, because it was never identical in both channels.

02What else lives in the center

Engineers mix kick, snare, and bass to the center for power and for vinyl-era compatibility, so the subtraction takes them out with the vocal. That is physics, not a bug in this tool or any other center-cancel remover. The keep-the-bass option patches the worst of it: the mid signal is low-passed at 120 Hz and added back, restoring kick weight and bass line while the vocal range stays cancelled.

The other survivor is the vocal's own stereo reverb. The dry voice was center and dies; the reverb around it was wide and lives, which is the ghostly halo you hear on most cancelled tracks.

03Seeing what the trick took

The What got removed mode renders two spectrograms: the removed signal — the original minus the instrumental, literally what the cancel subtracted — and the kept instrumental next to it. In the removed pane you can see the vocal's harmonic stack, and underneath it the regular vertical stripes of the kick and snare that went with it.

It is the fastest way to judge a specific song's suitability. A sparse acoustic mix shows a clean vocal in the removed pane; a dense modern master shows half the arrangement in there, and now you know why the instrumental sounds thin before you commit to it.

04Practice wants the vocal down, not gone

For rehearsal, a full cancel is usually the wrong tool — you lose the drums' punch and the pitch reference with it. The Practice mix mode scales the center instead of deleting it: at −6 dB the center plays at half amplitude, at −12 dB at a quarter, while everything stereo stays exactly where it was.

Worked example: a lead vocal peaking around −8 dBFS in the original sits near −14 dBFS in the −6 dB blend and near −20 dBFS at −12 dB — audible as a guide, quiet enough to sing over. The low end below 120 Hz is left at full strength in both settings, so the groove keeps its floor while the voice steps back.

05Karaoke night in one ZIP

The Karaoke batch mode takes up to ten stereo files, runs the same cancel on each one sequentially, and packs the instrumentals into a single ZIP. Each file gets the same keep-the-bass treatment you chose, and each output is a mono WAV — honest labeling, because a cancelled signal is one channel by construction.

Ten full-length songs is a lot of decoded audio for a browser tab. If the ZIP fails on a phone, run the batch in two halves — the limit is device memory, not the math.

06Positions versus voices

Everything on this page subtracts a position: center-cancel removes whatever sits identically in both channels, casualties included. SongStrip in the studio runs real GPU stem separation — a model that learned what a voice is, not where it sits — and returns a clean vocal and a full instrumental with the drums, bass, and stereo image intact. Run the A/B on this page first, then run the same song through the real thing and keep both stems.

Take it further in the studio — start free with 3 full packs, no card needed.