John Walsh, Phil Taylor and I have been participating in a
discussion about lengths of tunes compared. Just a few
bars or the whole tune? (With an argument going on about
the mathematics).
It occurred to me that John's is sort of doing a JPEG-like
compression of the tune and comparing the compressed
files. (There are a million details that mean JPEG would
be a *very* bad starting point for implementation).
Here's a sort of test case: Take a 32 bar tune. Swap the
A music with the B music and compare with the original.
It should show great similarity. The spectral idea would
since it essentially discards the order of the notes.
The BLAST / Windiff approach would.
A Fourier approach might well, though if there were a
discontinuity (i.e. big jump) at the phrase boundary either
introduced or removed by the swap then this would
create lots of havoc pretty much across the whole spectrum.
(And for other basis functions I suspect this would also
be true more often than not).
Laurie
To subscribe/unsubscribe, point your browser to: http://www.tullochgorm.com/lists.html
