On Thu, Mar 04, 2021 at 02:15:12PM +0000, Augustine Leudar wrote:

> http://www.ambisonic.net/ambimix.html

Or 'how to do FOA on a stereo mixer' 38 years ago...
Amusing but pretty irrelevant today.


Some simple facts:

Horizontal FOA requires three _independent_ signals, W, X, Y.
Binaural only provides two. That means that a correct linear
transformation from binaural to FOA can not exist. That's
just maths you can't argue with.

It could be done using 'parametric' methods, which means
you try to mimic how a human brain analyses the information
it gets from the ears, and uses it to reconstruct a scene.
That involves 'sensor fusing', i.e. information from other
sources (e.g. visual), expectations, previous experience
of 'known sounds', and even cultural bias.

Maybe possible with AI in ten years but certainly not today.

A human also can and will rotate his/her head in order to
resolve ambiguous localisation. This can't be done given
only prerecorded binaural signals, so part of the required
info will be missing anyhow.

That said, not all is lost.

Conversion from binaural to stereo is possible using 
something similar to 

<http://kokkinizita.linuxaudio.org/linuxaudio/zita-bls1-doc/quickguide.html>

This is a linear process and it could be done using
convolution as well. But then you don't have the interactive
controls and that could be a problem in practice.

Given the stereo signals, you could you use a simple
upmixing process to separate direct and diffuse sound.

Finally, using convential AMB methods, pan the direct
sound to the front and the diffuse part to the back.

That should give you 'plausible' rendering of the original
binaural recording, assuming the person wearing the in-ear
mics didn't move his/her head randomly too much.

Ciao,

-- 
FA

_______________________________________________
Sursound mailing list
Sursound@music.vt.edu
https://mail.music.vt.edu/mailman/listinfo/sursound - unsubscribe here, edit 
account or options, view archives and so on.

Reply via email to