giles at xiph.org
Fri Jul 1 01:50:52 PDT 2005
On Fri, Jul 01, 2005 at 10:14:01AM +0200, Thomas Bosch wrote:
> So, does anyone know the coding delay of ogg-vorbis?
There are two parts to this. First, the reference encoder buffers
some from which to make coding decisions. IIRC it's on the order
of 3 packets or so. It's possible to write an encoder with only
one packet's latency, but the quality/bitrate would be much, much
worse than the reference encoder, so there's not much that can
be done about this.
On top of the encoder latency, the Ogg encapsulation adds a further
delay as enought packets to fill a (by default) 4k page are accumulated
before the data is sent. Depending on the bitrate, this can be quite
some time. Of course you can avoid this by using a lower latency
container like RTP or (less efficiently) flushing an ogg page after
every packet submit.
IIRC the end-to-end latency of an Ogg Vorbis http stream I did in
a lan test a couple of years ago was around 3 seconds, so if you're
doing this in an interactive application you really have to get
rid of the Ogg delay, and then see if you can live with the Vorbis
Hope that helps,
P.S. I suspect FLAC can more easily trade compression efficiency
for latency. You might look at that as well if bandwidth isn't
More information about the Vorbis