[philicorda[_9_]]

On Mon, 14 Feb 2011 09:27:13 +0000, John Williamson wrote:

geoff wrote:
Bill Graham wrote:
I and a few of my musician friends can recognize a song from just
the
first three or four notes.

Occasionally I can from the first fraction of a second.

There is no reason why a computer couldn't do the same thing, but with
more accuracy. The problem is, getting those few notes into the
machine. A lot of people can't read music, so you would have to give
the computer the ability to translate from a whistle or something to a
score, or, give it a variety of ways to input the music.

Audio isn't a score. You'd need to deconstruct the audio into
frequencies, turn that into a scored arrangement, and then compare.
Could be done, but rather hugely complicated methinks. Not quite as
simple as Gracenote.

The smple way is to get a profile of the tune, much as is stored in an
mp3, then compare that profile with it's database of stored signatures.
that's why you can't do it without a large data store, which is why all
the portable apps use a backend store and processing.

There is a book from the '70s called "The Directory of Tunes and Musical
Themes" which identifies most tunes by whether notes are higher (u),
lower (d) or the same (r) as the previous note. So you can take the first
six notes or so and identify a surprisingly large amount of tunes. "White
Christmas" becomes "*udduuuu".