It's perfectly valid to compress the file as a whole by converting it in one huge transform, finding redundant frequencies, and compressing. Course, the decoder would have to do a huge inverse transform to get it back again - that means lots of RAM. It would actually be faster doing a single large transform though.

While it sounds like a great idea, giving you the best statistical compression over the whole, this isn't ideal because you get bad compression over short periods of highly detailed audio where you would want different frequencies removed than other periods. It's a bit of a trade-off finding the best block size for that sort of thing. MP3 uses 1152 sample blocks for compression, which translates to about 1/38th of a second, and probably isnt a bad choice (1024 would have been nicer though :)


- John.

(The above may not represent the views of Empeg :)