I guess in theory you could generate a 'demux mapping' and the client could request byte-ranges corresponding to only one channel, but that seems incredibly complicated, would generate huge requests, and is probably bypassable on the client-side anyway (more of an issue for my idea than yours).
Grooveshark manages to get licenses for all these "gray" mp3s after the fact, but I doubt youtube would be able to do so as easily considering their larger market position would be more threatening to the labels.
From wiki: "Audio in Flash Video files is usually encoded as MP3." On this page:http://en.wikipedia.org/wiki/Flv
FLV is a container, yes? Usually, I'm guessing, MP3 and H.264. H.264 works out well for youtube because they can then also use this on their HTML5 video players.
So are the H.264 blobs and the mp3 blobs stored as discrete files, then packaged when a video is loaded and sent down the tubes? If so, then yeah, obviously, it would be really easy or youtube to serve "audio only"; probably "a few lines of code".
But if the MP3 and H.264 (again h.264 is an assumption) are stored as one [flv] package, they would have to be unpacked before being able to be sent down as their individual components.
Again, this is a shortcoming in my understanding of video containers, so maybe I'm missing the point on this completely. (As in: maybe "unpacking" an FLV is trivial)
I often wonder how much google would save on bandwidth if they let me stream audio only versions of things. The vast majority of the time, I'm not even looking at the visual component of the "video", just listening to the song.
I suppose that this would add computational complexity, as well as additional storage, though. I'm not terribly familiar with the way that youtube stores video.
Can anyone chime in on this? "Extracting" audio from the video containers that youtube uses; how hard is this to do on the server side?