The video contains a second audio stream, as indicated by VLC here:

portal-01

portal-01

The second audio stream is an SSTV stream containing image information.  We extract the stream with the following command (the same file as the download was “credits.ogv” in the game files):

$ avconv -i credits.ogv -map 0:2 -acodec copy out.ogg

We then install pavucontrol so we can redirect our speaker output as an input, and QSSTV to decode the image… after compiling its dependencies.  Also create a symlink to libhamlib.so.2 so QSSTV can find it.  (Use strace to see where it is looking for the library.)

$ sudo apt-get install libltdl-dev libasound2-dev libjasper-dev pavucontrol
$ cd hamlib-1.2.15.3; ./configure; make; sudo make install; cd ..;
$ cd fttw-3.3.3; ./configure --enable-threads --enable-float; make; 
>     sudo make install; cd ..;
$ cd qsstv_8.1.12; make; sudo make install; cd ..;
$ sudo ln -s /usr/local/lib/libhamlib.so.2 /usr/lib/
QSSTV decodes the image

QSSTV decodes the image

Here is the final image generated by the extracted audio stream:

portal-03

portal-03

Each of the three channels within the image contains text; here is the red channel only in GIMP:

portal-04

portal-04

The text is the contents of an Aperture Science Image!  After painstakingly typing the characters (and making many mistakes), we download the decoder and try it out:

My failed attempt at typing all those characters

My failed attempt at typing all those characters

The transcription was terrible, but it is good enough to figure out what the image is supposed to say.

Flag: Cave Johnson hates Black Mesa