Before it was done with "state of the art AI" vocal removal was done by attenuating vocal frequencies in the center of the stereo field because in traditional stereo mixing the vocals are in the center of the stereo field and similar frequencies are moved left and right for clarity.