Hacker News new | ask | show | jobs
by hurtigioll 6 days ago
people said "it's impossible to separate tracks (voice, bass, ...) after they are mixed". true in theory, but neural-networks can separate them in practice

same here, but there is no real market for somebody to bother yet
2 comments
svelle 6 days ago
> but neural-networks can separate them in practice

That's a massive stretch.

They are able separate them sort of, but they are nowhere close the original quality of the individual tracks.

link
a96 2 days ago
The result is a convincing fake. Enough for many uses, but it's not the "source".
link
urbnspacecowboy 6 days ago
> people said "it's impossible to separate tracks (voice, bass, ...) after they are mixed". true in theory,

and true in practice too. what you get out of an ai demuxer isn't an exact match for what went into the mix. it's a plausible approximation, much like ai upscalers, upsamplers, colorizers, and such.

link