Dear Jim, thanks for your letter.  You wrote:

> ... This had the advantage of being just about as unique but,
> more importantly, didn't require nearly the effort or musical
> experience to do a transposition.

I'm counting on the computer's input recognition routine to try to make
sense from an amateur user playing the keyboard and, most important, to
provide feedback to the user so that he can check his input.  I suppose
an optional feedback at this point could be the alphabet format of Barlow
and Morgenstern, or the differential variation you describe.  I feel
it's important to have timing information, too, which is conveyed in
the MIDI format.

> Alas, we eventually abandoned the project for a couple of reasons ...
> The first and most obvious reason was the tremendous amount of work
> required for error-free data entry (reduced considerably, I admit,
> through synth keyboard entry) ...

Who can say what an error is, other than the user comparing what's in
his head with what he entered?  My premise is that the computer process
searches for several close matches; I guess that's a form of fuzzy
logic.

> ... but, most importantly of all, and this really stands out with
> musical box arrangements, is the problem of sorting out different,
> but similar, arrangements of the same tune.  Which notes belong to
> the tune and which have been added by the arranger?

Aye, 'tis troublesome, and the same problem occurs with lush piano roll
arrangements.  Just think of the confusion that might be caused by the
counter-melodies in the songs by John Philip Sousa!

> For a simple example, take the first couple of bars of the hymn,
> "Amazing Grace."  Really a simple tune and what could be easier?
> WRONG!  If you do a search on the title you'll come up with more
> than one tune, which is a problem but beside the point.  Worse,
> you'll come up with different arrangements (note sequences) for
> the same tune.  The current popular version includes something
> line seven notes in the first two bars.  The standard version
> includes four.  Does U*UDDU look much like UU* ?  Same tune!
> (In B&M notation we have GCCEDCE vs GCEE.)
>
> Many a time it can be quite difficult to know which notes belong
> to the basic tune and which belong to the arrangement or decorations
> to the tune.

The goal of the search is to discover the name of the song.  In
the example of my article, the primary data bases are presumed to
be of public, published melody copyright depositions, therefore only
registered tune titles will be returned.

The computer process is a success if several similar tunes can be
presented for the user's evaluation.  The _search_ is a success if
the user finds the name of the tune that's in his head.

> Just thought I'd make you aware of a very basic frustration in
> data entry and analysis you may not have considered.

I surely appreciate your thoughts.  Few people have had the actual
development experience as you did.

Robbie Rhodes