[theora] Video showcasing Theora in Firefox 3.1

[Gregory Maxwell]

On Sat, Feb 7, 2009 at 2:07 PM, Gregory Maxwell <gmaxwell at gmail.com> wrote:
[snip]
> That is very strange— and I wonder if the audio was encoded by some
> broken Vorbis encoder. (There is only one that I'm aware of— FFMPEG
> has a home grown Vorbis encoder that is seriously worse than the
> Xiph.Org reference encoder,  I'd never seen anyone use it until
> Archive.org managed to encode all their Ogv files with it. :( :( :( )
>
> Now checking I see that the broken ogv files on archive.org also have
> the same missing vendor string.  This is really terrible.  The output
> of the ffmpeg vorbis encoder is what I'd use if I were trying to make
> Vorbis look bad. :(


Just so no one is left thinking I'm being a nit-picking audio geek in
this case, I've posted a couple 11 second audio examples:

When you request 128kbit/sec from the FFMPEG Vorbis encoder you get this:

http://myrandomnode.dyndns.org:8080/vorbis/sample-ffmpeg128kbit.ogg
(it's really about 64kbit/sec)

When you request quality 0 (64kbit/sec typical) from Xiph.org
libVorbis, you get this:
http://myrandomnode.dyndns.org:8080/vorbis/sample-libvorbis64kbit.ogg
(it's about 64kbit/sec, as expected)

The FFMPEG result sounds like crud. The libVorbis result sounds fine.

Even at 32kbit/sec, a proper encoder (this exaple using the quality -2
mode of libVorbis based aoTuV encoder) is incomparably better than the
FFMPEG output:
http://myrandomnode.dyndns.org:8080/vorbis/sample-aoTuV32kbit.ogg