For me, genres are a flexible convenience for organization. I often make my own genres according to my impression of the music, not according to musical theorists' proposed definitions of the music.
I agree absolutely but I am not talking about that. My post was probably confuse (thanks to my nice english
)
The objective is to classify by similarities, calculated from the audio signal using criterias to be defined. More than to found predefined subsets the objective is to found subset of audio tracks similar to a selected audio track.
These subsets don't have to map predefined categories, they just identify which audio tracks in your library are similars to one selected.
There would be many way to use it but, for example, you could select an audio track as
your reference for a genre/style (that you could name as you want).
Thumb+ proposes the same concept for pictures: it is able to found similar pictures to a selected one; you are able to parameter the criterias used for the calculation.