Re: [codec] Conferencing and layering

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

Christian Hoene wrote:
> 1) A central conferencing server that serves all clients.
> - a) which does a efficient single changing mixing - still a challenge if you have multiple operational modes in a codec.
> - b) which does a more intelligent mixing using (at least) two channels to support spatial hearing - this increases the quality of conference call significant. Thus, stereo modes must be supported if the codec shall support state of the art conferencing technologies.
> 2) A mesh-like N-to-N transmission conference call. Then you might benefit from layered coding because you have at a client you have to encode only once. But then again, how to support multiple coding modes?
> 3) One client acts as conference server (similar to 1)
> 
> In total, the support of conferencing calls will require:
> - Layered coding
> - Efficient mixing of streams
> - Support of stereo
> - Good (self-)transcoding
> Is this all achievable?

It's obviously not all "required".  Layered coding is irrelevant unless
the network is heterogeneous and supports efficient multicast routing, and
on the internet this is essentially never the case.  The only places on
the internet where efficient multicast routing is supported are very fast
links where the small bandwidth savings of layering are irrelevant.

Similarly, "efficient mixing" is not really required.  Modern audio codecs
typically require so little CPU time that decoding, mixing, and
re-encoding for a large number of channels in real time is perfectly
possible without high-end hardware.  Certainly, conferencing makes encode
and decode efficiency an important consideration.  Codecs capable of
accelerated mixing would be intriguing, but by no means required.

--Ben