I have a technical question about the future of umixing with SpectraLayers:
About 2 years ago, i tried unmix and was amazed to find out that the summed stems = a bit-perfect recreation of the original sound file. I really hadn’t expected that, and realized it opened up a whole can of possibilities in mastering.
Around a year ago i heard (on the SonicState channel) an interview with the head woman at Audioshake, plugging their product.
It piqued my interest, so i gave it a try, and was amazed at the quality of their process.
BUT, and there’s a bit but, i suspect (although i never tested it because i’m too stingy to pay!), that their process will not sum back perfectly. So they sacrifice perfect summing for better individual stems, which actually makes sense for their target customer (what’s the point of unmix if you just want to perfectly re-sum it again?!)
Anyway, my actual question to the SL devs is: do you think AudioShake-quality splitting with perfect re-sum will be possible in the near future, or are we still a long way from that with the tech?
Cheers!
All I can say is that SpectraLayers unmixing quality gets better over time (see SL7->SL10->SL11), but SpectraLayers always make sure that the sum is always equal to the original file.
OK, but to play devil’s advocate for a moment, it could that you’re missing an opportunity there, because for many, perhaps even most people using an unmix process, they would sacrifice perfect summing for better sounding stems!
2 Likes
You’ve got a point.
Specially because, AFAIK the perfect summing only occurs when no changes done to any layer. Of course as soon as you do any processing, the “perfect summing” is gone. Either because of the very processing or artifacts that appear in the audio.
1 Like
It does not sum up to “perfect zero”, at least, because of numerical issues of floating point numbers.
To see this, just take a song,
- duplicate it,
- unmix the first song,
- reverse the phase of the duplicate,
- add all layers up together
- normalize the resulting layers
here is the noise…
So, no, it does not add up to zero.
But we may hope, that the errors remain bounded in the lower bits.
So, year, it’s fine!
Right, but for any practical purposes, it’s good enough.
Because presumably if you’re unmixing, you’re going to do something to the audio, at which the perfect resum is a moot point.
It’s more about whether you get weirdness like phase stuff or other strangeness, which i suspect (but didn’t actually test) you’d get with a better individual stem ripper like Audioshake.
