>Because the watermark is going to be different for every copy of a
>particular song this suggests that if you get three copies of a song with
>different watermarks and do bit voting with them you can produce a fourth
>file that contains all the information that is the same in the first three
>(the song) but does not include any of the differences (the marks).
....
>If memory serves the mark Anderson describes is sort of steganographic in
>nature. Most of the marks that I've seen have been FIR filters. This
>complexity is needed to survive analog reproductions. As Liquid Audio
>claims that their mark will survive analog reproduction its likely to be a
>filter. Can you build a filter that removes the mark? Probably. Most of
If you're limited to filters that survive transition to audio and back,
the space available for watermarking becomes smaller but hairier,
requiring more complex math than simple stego techniques for digital environments,
and also is more likely to do things that annoy listeners.
Bit voting may well catch most of the pieces if the watermark
looks like "tweak the value of a few bits", but it's tougher to use
on changes like "add or drop samples at known points", or
"speed up or slow down sections by 1%" or "add different frequencies
with very low amplitude" which may cause huge numbers
of differences for simple comparisons (maybe this is ok,
since you're trying to disrupt watermarks rather than impersonate them.)
But there's a lot of room to hide large numbers of bits, letting you
play spread-spectrum code-division kinds of games that can still
tell you some bits about the user even after voting kills lots of them.
Thanks!
Bill
Bill Stewart, [EMAIL PROTECTED]
PGP Fingerprint D454 E202 CBC8 40BF 3C85 B884 0ABE 4639
