Hi!
This is an awesome software. Thanks for making it.
I also encountered this problem. The 100% similarity problem. It was present in 0.71, and it is also in 0.9.
There are some mp3 files, which differ completely in length and music from other mp3s, but Similarity detects them as 100% similar in music content. And this becomes very horrible as this mp3 file can be similar to hundreds of mp3 files. Of course it's not true.
I tried to look why this can happen.
1. The 100% similar mp3 file has very strange beginning. No ID3 tag at all.
2. The 100% similar mp3 file cannot be played in Total Commander by pressing F3 (View).
Normal mp3 files can be played.
3. Audacity, an mp3 editor crashes while trying to load this 100% similar mp3 file.
(I looked at another 100% similar file, and it can be loaded, but only 5
seconds is editable in Audacity)
It seems like this file is broken somewhere, but it can be played in normal music player.
It would be great if Similarity would ignore these buggy 100% similar files. And maybe even show them as buggy, so I can delete it.