theora - Apr 2010 - Ogg/theora a/v sync issue with liboggplay

Hi,

The following url is being played with incorrect a/v sync in Firefox:

http://195.10.10.216/flumotion/videodemo/med.ogg?token=c965dab73e5ed982821e01174e87925a4bb3134e4bb9aaec
(it is time limited, but should work a few days, I could provide)

Tested with 3.6.3pre on Linux (with and without PulseAudio) and
Firefox 3.6 on Windows XP.

Looking at firefox source code, I found the library that is supposed
to do a/v sync for ogg is liboggplay.
glut-player, compiled from git master (3901c1c38), shows the same a/v
sync issue.

I ran oggplay-info and noticed that the theora overrun histogram is
quite important.

Theora: Track 0
       Worst overrun: 271 frames
       Average overrun: 263.202 frames
       Histogram bucket size: 13.550
       Histogram: 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 42 2022
       SD of overrun: 3.162278

Vorbis: Track 1
       Worst overrun: 23552 frames
       Average overrun: 10787.287 frames
       Histogram bucket size: 1177.600
       Histogram: 94 109 109 116 107 108 156 118 107 115 109 119 117 128 118
107 96 77 39 13 2
       SD of overrun: 6028.785367

Is the stream bad, or is liboggplay doing incorrect a/v sync?

It looks to me like liboggplay should do correct sync, given the output of:

$ oggplay-info demosite.ogg
Reading 2 tracks ...
       Theora: Track 0 ... ok ...
       Vorbis: Track 1 ... ok ...
track 0 required 1 times [1892128590] track 1 required 2 times
[1892128593, 1892128614]

thanks for your help

-- 
Marc-Andr? Lureau

> Is the stream bad, or is liboggplay doing incorrect a/v sync?
It appears that audio is substantially leading video in the stream.
Although both are timestamped, this isn't strictly correct behavior; a
stream is supposed to be multiplexed in time order.  I'm guessing that
liboggplay is simply assuming correct time multiplexing and assuming
that each stream's start offset is to be corrected to a time of zero.

So... there's a bug in both.

Many live television streams start out as MPEG2, and MPEG2 can have a
substantial A/V offset; whatever is doing the encoding is not
accounting for it, and so the stream multiplex is way out of sync.

BUT A decoder that is working properly should still account for that
because it has the timestamps.  It will have to buffer way more data
then normal to cope, but it should still be capable of sync.

Monty