> I work for eTantrum, the company that created this. :)

Any chance you can post a description of the algorithm? I'm having a hard time figuring it out from the source code.

I actually had an attempt at doing something like this some time ago. I used voc26 from SFS, and created coarse vocoder samples from the first 15s of a tune. The problems I immediately observed if I downloaded a tune from napster from several different people (of something I already owned of course ;-) ) were that tunes were of different amplitude, differing, or no leading silence, different frequency responses (maybe due to bit rate) and inverted wave forms. It seemed that some normalisation phase was required prior to generating a signature. How do you handle these kind of things?

Some links that might be of interest:

Tuneprint
A Patent
Srinivasan
Musclefish
Musclefish paper

Paul