Re: [codec] possible issues to track

["Benjamin M. Schwartz" <[e-mail]>]

Michael Knappe wrote:
> Ben, how would you approach channel counts larger than 2? Joint coding
> between any two L-R pairs, with other channels (like the center channel)
> left singular?

I agree with Stephen Botzko that the mechanism should not be part of the
requirements.  However, we will need to agree on an interface, i.e. the
abstraction barriers.  I think:

1. The codec should provide an interface for coding a pair of closely
coupled channels with the presumption that they represent something like a
stereo pair.

2.  The format should provide an efficient way to pack multiple
independent synchronized codec channels into a single stream.

3. The format should provide a way to specify N named output channels as
linear combinations of the coded channels.

Advantages:
 - Stereo can be specially optimized.  There are mature techniques for
efficient coding of stereo, and plenty of test samples.  CELT can already
code joint stereo more efficiently than independent coding or a simple
mid-side split.
 - Surround sound can support all the various configurations (e.g. 5.1,
Ambisonics, custom speaker sets) without any major work within the spec.
 - The codec does not need to know anything about specific surround-sound
configurations, saving development time and effort.

Disadvantages:
 - This approach cannot achieve maximally efficient coding of
surround-sound, as can be achieved using techniques like those mentioned
in [1] and [2].

--Ben
[1] http://people.xiph.org/~xiphmont/surround/demo.html
[2] http://people.xiph.org/~xiphmont/surround/demo2.html