[xiph-rtp] Re: new RTP development list

[illiminable]

----- Original Message ----- 
From: "Tor-Einar Jarnbjo" <Tor-Einar at Jarnbjo.de>
To: "illiminable" <ogg at illiminable.com>
Cc: "Ralph Giles" <giles at xiph.org>; "Aaron Colwell" <acolwell at real.com>; 
<xiph-rtp at xiph.org>
Sent: Thursday, October 21, 2004 5:02 PM
Subject: Re: [xiph-rtp] Re: new RTP development list


> Torsdag, 21 oktober 2004, skrev "illiminable" <ogg at illiminable.com>:
>
>>There's no gaurantee that those few promo packets will arrive before
> song
>>2's start packets is there ? And since they are not time continuous,
> you
>>have no way to know they are out of order. So potentially... you
> could see
>>song 2's packet... start up a new chain... then recieve the promo
> packets,
>>think it's a new chain... then get more song 2 packets and be totally
>>confused about what order you should be playing things in.
>>
>>Or do i not fully nuderstand how it works ?
>
> You're more or less right, but streaming Vorbis over RTP means replacing
> the Ogg conatiner with the RTP header, and with chained Vorbis streams,
> calculating the RTP timestamps would have to be done using a more
> complex mechanism than simply copying the granulepos fields from
> the Ogg pages. It would not work to allow the RTP timestamps to skip
> back and forth in time.
>

What i don't get is why there is going to be different protocols for each 
codec.. wouldn't a single protocol for all ogg codecs be better ?

Then people can stream non-standard combos like speex+theora.

And when it comes to theora... trying to generate a consistent timestamp is 
going to be hard because theoras wierd granule pos scheme.

ie the granule pos is not of consistent intervals... and the conversion 
granpos-->time is non-reversible. So it would be hard to chain files of two 
different keyframe intervals and find a consistent granule pos scheme.

Zen.