Thursday, April 3, 2008
I know I'm getting off the BCI track a little, but this was too cool to not post.
A new compression format for audio could compress music down the 1/1000th the size of current mp3s. That's not the interesting part. The interesting par tis how they do it: they model every instrument used in the recording and then decode the saved music by essentially replaying it through the encoded. So, there is an encoder for the saxophone, the movement of fingers on the violin string, etc. The linked article on Gizmodo went so far as to suggest modeling the vocal cords of, say, Frank Sinatra, so that old artists could release new music. I'd bet the Elvis estate would be alllll over this.
The two obvious cases where this wouldn't work well would be a) many, many many, layers/tracks/simultaneous sounds (though that would probably have to be in the tens of thousands range), and b) many, many, many instruments/sound producing items coupled with (a). The way the article describes it, each instrument is an individual class, but there are some hints that the decomposition is not related to any 'real' entity, but some type of waveform classification. I tried to think about how this could be done, and really, there are probably about at least a half dozen ways that I could think of off the top of my head, and I'm not an engineer.
Really cool stuff. Also, it is theoretically capable of 1:1 quality matching. Also, and album would be 200kB. And broadband speeds are increasing everywhere*. Night, night RIAA.
*-Verizon FiOS is now available in my area! As soon as I find a real work around for the obnoxiously bad router they provide, which is uber specialized (MOCA and Cat5), I am all over that. I hate my current cable/internet provider (Cox Comm).