VP3 to Theora
Now, this quick guide involves a lot of tools which are either experimental or are known to be buggy. Beware. Use with caution.
mencoder -oac copy -ovc copy -fafmttag 0x55 -o vp31.avi input.nsv
You can skip this step if your input file already is an AVI.
Get the Theora code
Download the Theora source code: svn co http://svn.xiph.org/trunk/theora
Build the avi2vp3 tool
Go to theora/win32/experimental/transcoder/avi2vp3/
Compile the tool which will extract the VP3 data out of your AVI file: gcc *.c -o avi2vp3
Copy theora/win32/experimental/transcoder/transcoder_example.c to theora/examples/encoder_example.c - this will replace the usual Theora encoder example with the application creating a Theora stream from the raw VP3 data.
Configure your copy of Theora (run autogen.sh). This has a few dependencies. The configuration process will guide you. Be patient. Once configuration didn't complain because of missing dependencies: Invoke make.
After everything compiled you should rename the encoder_example binary (which is in the examples directory) to e.g. transcoder_example to avoid confusion.
Process your videos
- Use avi2vp3 to convert your VP3 AVI. The input file *must* be named vp31.avi, the output file will always be outfile.vp3
- Use transcoder_example to create an Ogg Theora stream: transcoder_example outfile.vp3 > video.ogg
- Use e.g. mplayer to decode the audio in the AVI file: mplayer -vo null -ao pcm vp31.avi
- The audio will be in audiodump.wav. Encode it with e.g. oggenc: oggenc -q 1 audiodump.wav
- Now you need to combine video.ogg and audiodump.ogg. Use oggzmerge from the oggz-tools package (at least in Ubuntu it's oggz-tools): oggzmerge video.ogg audiodump.ogg > theora.ogg
theora.ogg now should be a valid Theora + Vorbis Ogg file. Congratulations!
If the VP3 video input has dropped frames (this is common on broadcasted videos with narrow bandwidth constraints) you may notice that the created Theora file is distorted in affected parts of the video (apparently because the information that frames were dropped is lost, thus some video frames predict from the wrong predecessors).