Tech TangentsWhat every self respecting software engineer is thinking inside as they watch AI wreck everything.
Popular Science
Popular Science gives our readers the information and tools to improve their technology and their world. The core belief that Popular Science and our readers share: The future is going to be better, and science and technology are the driving forces that will help make it better.
"Look! We used AI to remove vocals from songs!"
*meanwhile in the 80s*
"We accidentally got some wires backwards and realized we could usually remove the vocals, so here's that in a box"
From: https://books.google.com/books?id=0AAAAAAAMBAJ&pg=PA122#v=onepage&q&f=false
VWestlife@TechTangents Here is the demo disc for the Thompson Vocal Eliminator: https://youtu.be/3NmUq0CpOuI
Thompson Vocal Eliminator 1980 demo flexidisc
Christopher Snowhill@vwestlife @TechTangents Doesn't that 80s solution only work if the vocals are perfectly center panned, and also ruins the stereo image of the audio?
Lux 
@chris @vwestlife @TechTangents I think that in a karaoke session, you don't care that much for stereo.
@orange_lux @vwestlife @TechTangents True. I think the vocal remover they made with AI was harvested from some tech that was actually made for quality manipulation, but it was so cheap to license, they shoved it in an expensive karaoke box.
Charles ☭ 
is a Green@chris @orange_lux @vwestlife @TechTangents
Huh, I'd always assumed these were band pass filters.
If you have the original left and right signals and the difference, can you reconstruct the stereo by passing the AND of the channel and the difference? This wouldn't work on a sample by sample basis but maybe it would with FFTs?
@celesteh @orange_lux @vwestlife @TechTangents
Ah, yeah, real band pass filter would probably be a lot better.
I was thinking of the dumbass wiring mistake that would make one think of this sort of processing in the first place, where you disconnect the common line and the stereo image collapses. But that would only affect centered vocals if they were somehow inverse phase between channels, now that I think about it.
Man, I really should not try thinking about this sort of stuff.