[FFmpeg-devel] Audio Video Sync issue with dvr-ms files

[Jonathon Elwood]

[...]
> > >
> > Unfortunately, I'm stuck?  With the attached patch the fps is now
> > detected correctly on all samples.  Also the AV sync is correct on 80%
> > of my samples.  This is a nice improvement, but I still have the
> > problem where the video is "stuck" on the first frame (this accounts
> > for the 20% where the AV sync is incorrect).  Unfortunately, I don't
> > know where to look.  Any help or guidance would be greatly
> > appreciated.
>
> one thing which is interresting is that asf videos generated by ffmpeg
> are also sometimes having "stuck" frames when played with some versions
> of wmp, so finding the exact cause of this might help .asf muxing too
> maybe ...
>
>
> > If anyone would like to see a sample that exhibits the "stuck frame"
> > problem, I uploaded the sample on April 3rd that demonstrated this
> > issue.   ftp://upload.mplayerhq.hu/MPlayer/incoming/ NBA
> > Basketball_WPTA-DT_28_01_2007_14_59_45.dvr-ms.  If this sample doesn't
> > exist on the ftp site any longer, please let me know and I'll upload
> > it again.
>
> yes, please do, also upload a .txt file with the same name which says
> what the sample does

I believe the NBA sample is still there.  I tried to upload it again
and I got a message saying "can't create file".  I'm guessing it can't
create the file because it already exists.  Just incase I uploaded
another sample that shows the problem.  The file ABC_SD.dvr-ms stays
on the first frame for roughly 8 seconds.  The NBA sample stays on the
first frame for roughly 2 seconds.
>
> [...]
> > Index: asf.h
> > ===================================================================
> > --- asf.h     (revision 8977)
> > +++ asf.h     (working copy)
> > @@ -114,7 +114,10 @@
> >      int packet_time_delta;
> >      int packet_time_start;
> >      int64_t packet_pos;
>
> > +    int know_frame_time;             //used for dvr-ms files
> > +    uint64_t avg_vid_frame_time;     //used for dvr-ms files
>
> unused
>
>
> >
> > +
>
> cosmetic
>
> but i guess the patch wasnt intended for svn anyway ...
You are correct, I didn't intend to submit this for svn.  Until the
"frozen frame" problem is figured out, I don't know that my changes
are correct.  Also, I would like to clean up my code that accounts for
the segment_marker moving positions.  I think there is a cleaner way
of accounting for this, but I just threw together the change quickly
just to move onto the "frozen frame" issue.  I'll remove these items
that you noticed so they won't be there next time I submit the patch.