I've done it lots of times. It works... partly.
Last time I did it was 12 months ago, when I had to record about 8 passes of a small kids 'choir' and mail the files on.
I used speakers and a pair of U87's. NOTHING was moved, and we stacked passes. Then the kids stood there, while I recorded a silent pass.
I then sent them the 8 passes plus the silent pass, then I sent them one 'bounce' version with 4 passes in + polarity, and 4 passes in - polarity, summed equally.
I've played with lots of cancellation options using this method, and -if you're stacking things like strings or kids, it certainly
reduces bleed.
'Eliminates'? -No. -Not even CLOSE. but reduces? -Yes. Very much so.
HOWEVER:
The cancellation is limited to longer wavelengths, since air convection patterns in a moderately large room, TEMPERATURE DIFFERENCES between passes... (-yes... they DO matter!) and various other factors all conspire to reduce the effective cancelation.
-On the subject of temperature; you can easily demonstrate it by quickly recording two short passes back to back in an empty room, then adjusting the temperature one way or the other, and recording two more passes... then allowing it to return to the original temperature and doing two more.
...Then try canceling the first pair. -It'll cancel fairly well. So will the second, and so will the third. But the first and the second wont cancel, nor will the second and the third, but the first and the third should cancel quite well. Also, if the room is significantly ventilated, air currents will affect cancellation.
Similarly though, if you stack in like-polarity, the low end of the speaker foldback sums strongly, the midrange progressively less so, and the high end doesn't sum much at all. You can sum half a dozen passes and observe this cumulatively.
Now, flipping the polarity of alternate passes DOES indeed cancel, but -as the exact converse of that last quoted example, the LF cancels VERY well, the midrange cancels a little less well, and the HF doesn't cancel all that well at all.
Lessons to draw: -When using speaker foldback and polarity-canceling later on, try to keep the speakers slightly 'dull'-sounding. The less HF you can spray into the room, the less will remain.
Because of the multiple-wavelength distances involved, shorter wavelengths are progressively more easily 'shifted' by any one of several factors: reflection (from moving singers/players etc.) convection, air current speed/direction and air temperature... and even pressure I suppose, although I'd only expect to a TINY degree, and insignificant in practical terms.
There are a few great suggestions already in this thread; 'emrr' gives a great idea for making 'patches' less obvious. -But try to maintain a sense of what's reasonable. -There's no 'magic bullet' to make speaker spill completely disappear; but you CAN make it ALMOST disappear at the low end, suppress it a bit in the midrange... and the rest is up to you.
I also did the 'out-of-phase auratones' trick years ago... but be aware that while the PRESSURE component of the wave between the two will cancel, the VELOCITY component is actually REINFORCED by doing this. An
omni mic between two out-of-phase speakers will produce a significant dip in level. A
Figure-8 mic int he same spot will record an INCREASE in level. -A cardioid probably won't hear much difference in level, (though the tone might change a little).
Nothing's completely simple, nor is anything absolutely foolproof. -But for occasions or locations where headphones are not an option -whether due to the number of singers, or a single insecure vocalist who's scared of headphones- there ARE some tricks. -None of them are perfect, but some of them may help. One of the most educational things to try can be listening to -and metering- the different results from pattern-swapping a multi-pattern microphone, midway between two "out-of-phase" speakers setup facing each other. Set it up and try it sometime.
Keith