3 ms·

It _can_ be nice, if properly used. The Queen example is lousy at best: instead of having the single audio source wander in space, it could have been used with

It _can_ be nice, if properly used. The Queen example is lousy at best: instead of having the single audio source wander in space, it could have been used with a true mix to actually position (and move, properly) each voices/instruments in space. But of course, you'll need the original tapes to be able to do that. :p

timrichard 8y ago
I agree... going to be limited and annoying without at least the stems. I quite liked this demo, which shows a bit more potential for placement : https://www.youtube.com/watch?v=N_vVU43Kz_w https://www.youtube.com/watch?v=N_vVU43Kz_w
HocusLocus 8y ago
There is a project that does attempt to disentangle musical instruments and voices in the waveform domain that is interesting, http://jordipons.me/apps/end-to-end-music-source-separation/ http://jordipons.me/apps/end-to-end-music-source-separation/ ...such that the results could be flanged and recombined separately. But as you hear in their demos, any recombined result would be a Picasso painting as compared to a photograph, maybe interesting though.
notacoward 8y ago
My thoughts exactly. I'd like to be able to hear the guitar here and the vocalist here and the drums over there. If the individual sounds move around separately that's cool. The Virtual Barber Shop example does a good job illustrating that, with a voice on one side and a phone on the other simultaneously etc. They even play with distance as well as direction. I have only a vague idea how that works, but it definitely adds to the experience. The Bohemian Rhapsody example, by contrast, is just crap. Panning the entire ensemble from left to right, over and over? Meh. Then it's not even stereo like it was originally. It's just sequential mono. It's reduced, not expanded. These effects generally require at least having the original separate tracks, if not specialized recording techniques. However, I still have hope that advanced signal processing can separate the parts well enough after the fact to get a satisfactory result. I'm no ML fanboi by any means, but I'm pretty sure those techniques could be productively brought to bear on the problem as well. I'll bet there's a grad student or two working on it right now, and I wish them great success.

timrichard 8y ago

I agree... going to be limited and annoying without at least the stems. I quite liked this demo, which shows a bit more potential for placement : https://www.youtube.com/watch?v=N_vVU43Kz_w https://www.youtube.com/watch?v=N_vVU43Kz_w

HocusLocus 8y ago

There is a project that does attempt to disentangle musical instruments and voices in the waveform domain that is interesting, http://jordipons.me/apps/end-to-end-music-source-separation/ http://jordipons.me/apps/end-to-end-music-source-separation/ ...such that the results could be flanged and recombined separately. But as you hear in their demos, any recombined result would be a Picasso painting as compared to a photograph, maybe interesting though.

notacoward 8y ago

My thoughts exactly. I'd like to be able to hear the guitar here and the vocalist here and the drums over there. If the individual sounds move around separately that's cool. The Virtual Barber Shop example does a good job illustrating that, with a voice on one side and a phone on the other simultaneously etc. They even play with distance as well as direction. I have only a vague idea how that works, but it definitely adds to the experience. The Bohemian Rhapsody example, by contrast, is just crap. Panning the entire ensemble from left to right, over and over? Meh. Then it's not even stereo like it was originally. It's just sequential mono. It's reduced, not expanded. These effects generally require at least having the original separate tracks, if not specialized recording techniques. However, I still have hope that advanced signal processing can separate the parts well enough after the fact to get a satisfactory result. I'm no ML fanboi by any means, but I'm pretty sure those techniques could be productively brought to bear on the problem as well. I'll bet there's a grad student or two working on it right now, and I wish them great success.