How do upmixers like Dolby Surround and DTS Neo/Neural work? They attempt to produce rear/center/vertical surround sound channels from a normal stereo source—and often seem to do quite a good job. Explanations I can find are all marketing mumbo-jumbo: “identifies spatial cues” 🤷♂️
Conversation
Replying to
Do they really start from stereo?
Stems and submixes are sometimes available to old songs, for instance.
At any rate, one possible tool is to move different parts of the frequency spectrum to different speakers.
“Let’s put the highs - mostly a scream - in that speaker.”
1
Oh, I was thinking manually.
Algorithms might be related to the active research problem known as source separation.
If you’re going to mix everything together again anyway, maybe you don’t have to separate things perfectly.
Or maybe the answer is just “neural network magic” 🤷♂️
1
Replying to
Tangential factoid--Ray Dolby ('51) & Gordon Moore ('46) had the same math teacher at Sequoia High in RWC, the great Donald Kauffman.
1
1
Replying to
purely speculative, but you might be able to compare a sound in the mix with the diffuse version of itself (the reverb/room-sound). a sound which is very similar to the reverb-sound is more likely to be environmental, adirectional, and farther from the viewer
1
1
1
a direct sound (not similar to the reverb) is more likely to be on screen, and therefore in front of the viewer. this doesn't hold up for nearby-behind-you things, but those seem like the exception
Replying to
This article was a good primer on it, I thought.
3
“To produce spatial audio that lets you hear things in precise 3D locations through a set of headphones,...
Replying to
I think they use differences in phase and time delays to estimate lateral localization. I don’t know how they do source separation, though
1
1