I can't imagine anything you could get out of audio data encoded in any format that would be appropriate for a level generator. Maybe with some DSP work like beat detection or something, but certainly not raw bytes. You may as well just use random seeds based on a hash of any file's contents in that case.
Yeah, I have to agree with this.
In this case you wouldn't even need to read it as an ogg file, but you would just read in the bytes until you get and EOF and generate the level from there. I don't see any real need to go into any extra work to read in the stream perfectly.