[opus] ambisonics formats and channel mappings

[Michael Graczyk]

Marc,

Thanks for the comments.

On Sat, May 28, 2016 at 5:50 PM, Marc Lavallée <marc at hacklava.net> wrote:
> So, channels will stay uncoupled for Ambisonics? That would be great,
> because considering how good Opus already is, I don't think that more
> optimization is required, and it would put content creators at ease;
> they often believe that any compression or channel coupling could
> break ambisonic decoding.

Yes, the current implementation of the libopus encoder does not couple
channels when configured with the ambisonic channel mapping and I do
not intend to change that.

> I'm certainly suggesting to *include* stereo down-mixing, because it's
> trivial to do, and it would allow compatibility, something very
> important to promote Ambisonics. But I would not include decoding to
> other output schemes like 5.1, because it would be like putting a square
> peg in a round hole; decoding Ambisonics for irregular arrays can be an
> art form, it's not like working with common surround formats.

Thanks for the input. I agree.

> I would suggest to also include support for the UHJ format, which is
> roughly equivalent to 3 channels FOA, but using only 2 channels; even
> soundfield rotation is possible with UHJ (although less good), enabling
> horizontal "VR audio" with only 2 channels.

UHJ is an interesting way to preserve compatibility with non-ambisonic
playback systems. However, I have not seen it generalized to higher
orders. I expect that its popularity will decrease as HOA becomes more
and more common. If UHJ becomes popular in the future, we could
specify rules for encapsulating UHJ in Ogg. I believe that should use
a different channel mapping for simplicity.

-- 

Thanks,
Michael Graczyk