Multilanguage compression

[TetarZ]

O.K. that's probably a dumb Idea but here it is anyways:
U all how better comression can often be reached using joint stereo.
How about joint languages: When you mux a couple of sound files in a container they may be 90% simillar the size won't be any smaller.In different languages the only thing that changes most of the time is the voices. All the background noise music etc... is the same. Therefor if the various languages where compressed in the same file we might be able to acheive better overall size.
And again it could be neglectable. I don't know.
Anyone who knows is welcomed to get me out of my stupidity.

[Atamido]

No, thats perfectly accurate. You could encode to nearly-identical audio tracks in something like stereo and have excellent compression. The problems are that then both audio tracks would have to be in the same stream, and there is no implementation that looks like this so it would be hard to get support for.

[vinouz]

Not sure this would really be a good idea. Since It would need the different language tracks to be perfectly correlated in time. And it often appears they're not perfectly. Even, in many translations, the background track is filtered and modified by the treatment. So I don't think an encoder would gain significantly from that.
If you wanna try it, take two different language streams you converted in mono ones, and join them in a stereo track. Try some joint stereo encoding on it and compare with pure stereo at the same bitrate......
(I didn't try it...)

Vincent

[RadicalEd]

That really actually is a good idea.. It should be possible that the difference between two language files could be extracted to create 3 sets of audio data. One signal containing the audio data that is the same in both files (bg music, sound effects etc), one containing the extracted difference of language 1, and the third containing the extracted difference of language 2. Its possible that could be implemented and then encoded as one audio file, distributing bits where necessary. You could probably get away with a much lower bitrate in that case, since you're only encoding one set of acoustics and two vocal tracks. On playback the decoder could playback the bgm and dynamically overlay the language chosen. It would take more CPU to encode/decode but the complexity:cpu demand ratio is no new concept. Likewise you could also get a playable BGM/SFX track out of it for no extra bits.
Not sure if that would require a new codec altogether or just a tricky encoder/decoder :\

[bond]

Sounds interesting but i think it will be better to post this issue at hydrogenaudio.org (http://www.hydrogenaudio.org)!

[vinouz]

this would be interesting if DVDs were authored "trackwise". I mean if there was different tracks for instrumental part, fx, and voices. This is something built-in the MPEG-4 norm, and I don't know if it was considered in the design of the MPEG-2 norm.

But for now, even the common part betweeen the two audio wouldn't be usable. Because there often are few milliseconds between "background" part of the different languages tracks, and in a joint stereo coding scheme it wrecks all. Joint stereo is designed to code difference in level, not a delay between two tracks.
And also many background tracks are filtered. In order not to completely repeat me, take Magnolia. I remember clearly noticing this during one part, when there was 'Goodbye Stranger' from Supertramp for the background, while people were talking. If you switch between French and English tracks, the song hasn't the same pitch ! It seems as for the translation they choosed to pitch roger hogdson down for maybe some voice timber reason... (??).
And here I don't even talk about the filtering, which alters completely the track.
Understand that a codec codes the sound, but it doesn't go, until now, as far as saying : "oh, this is Supertramp in the background, here and there too, but here pitched down, filtered and 13ms delayed. And in front of it there are two guy and an off voice talking. Let's code that in this track and this in that track...."
Or else a compressed video file would look like a movie script, and take 50kb :).
BTW I'd really like it, when codecs reach this point :oD.

Anyway, maybe I'm again totally wrong. And yes, maybe in some cases it would work.
But I think we better wait until we get MPEG-4 compliant streams with real distinguished audio tracks, if one day studios decide to use MPEG-4 (not sure at all, see DRM), and to take advantage of all these clean features at their best (not even as probable)...

Vincent