On Sun, Feb 21, 2010 at 10:44 PM, Stefan de Konink <stefan at konink.de>
wrote:> Op 22-02-10 04:22, Paul Sykes schreef:
>> If I want to encode two audio channels that are not related, audio from
>> two different sources not stereo, is it best to create two instances of
>> a single channel encoder or use one encoder with two channels?
>
> With two different instances you would never have the benefits of any
> entropy found in the other channel and the single memory structure. So
> unless this will lead in a quality reduction because an assumption is
> made that there will always be similarities (like with joint-stereo) I
> would go for the multichannel variant.
He did say that they were not related.
I would not run unrelated audio through CELT in stereo mode, instead I
would use separate mono channels. In stereo mode the channels in the
pair are constrained to behave the same way in some regards (for
example, short vs long transform selection). This is reasonable and
even beneficial when the channels are related. It is not reasonable
if they are not related.
CELT has practically no container overhead which you avoid by running
only a single instance for unrelated audio.