How can you compress sound without losing something?

All man-made (and most natural) signals are at least a bit redundant (i.e. contain less information than they take up bits, so to speak :)

On one end of the range are text files: say that the file contains only 1000 different words, on average 4 letters each. We can code 1000 different symbols in 10 bits, and 4 8-bit characters take up 32 bits. Hence the room for compression. Most popular algorithms (Huffmann, LZW etc) differ mostly in details such as whether they need a dictionary or not, do they use fixed or variable lenght symbols etc.

On the other end are signals like audio, where we must use our knowledge about it (e.g. the fact that the difference between two consecutive samples will in most cases be relatively small). But see those academic papers for details.

Of course, sometimes a signal that does not seem redundant at all actually is. For example, white noise one gets by recording, say, cosmic radiation seems as random as output from linear congruence generator, and yet the later's output can be described by just three numbers: seed, factor and modulus (one can find more on this on cryptography sites).

Sorry for being boring.

Cheers!

Dragi "Bonzi" Raos
Zagreb, Croatia
#5196