Another letter and its answer;  I still think the idea is feasible and
has commercial possibilities.  I think of the old 78 recordings of Arthur
Schnabel playing the Beethoven sonatas.  If these performances could be
extracted to midi files and then recreated on a modern piano, I think the
recordings would be highly marketable.  I also think an "analog to midi"
box would make Disklaviers and Pianodisks much more attractive.  Imagine
playing your records or cd's and having the output come out through a
real piano... Talk about high fidelity!

> From bekosoft@tibalt.supernet.ab.ca Sat Oct  7 15:10:36 1995¶
Subject: Re: Pitch to Midi

Hello again Mr. Pleatman,

On Fri, 6 Oct 1995, you wrote:

> Thank you for your careful consideration of my question.  I have indeed
> thought of the problems you have suggested, and am aware of the
> complexities.  I am interested primarily in the piano, and am aware of
> all the overtones produced by each string.  I had thought that one could
> predict approximately how the high the amplitude of the overtone would be
> in relation to the primary frequency, one could decide whether or not the
> note was being played by artist.  There could be an algorhythm (sp?) to
> detect overtones and discard them if they were less than x % of the
> primary tone, and if they appeared at the same time.

The problem here lies in the fact, that the spectral components are not
fixed entities, but are rather dynamic. You would have to take a good long
look at the spectrum for each note over time to see how easily a given note
(including it's dynamics) can be identified by one component of the spectrum,
and if there could be conflicts between that component and a harmonic created
by a different note. I found this very difficult because you need to look
at a 3 dimensional spectrum to investigate this and I found that very
time consuming and difficult. The 3D spectogram would need to show the 2D
spectrum of amplitude vs. frequency, and in the third dimension, how the
spectrum varies with time.

> I even considered the possibility of using 88 separate electronic
> circuits, each tuned to detect a single frequency.  The computer's job
> would be simplified in that instance to only the task of measuring the
> loudness of each tone and deciding, at each time interval, whether the
> note was being played anew or just sustained from the previous interval.
> Of course, the problem of the overtones would still be there.

This reminds me of my early days with computers, when I thought of connecting
solenoids to a typewriter to turn a typewriter into a printer :)
What most midi keyboards do, is to use two contacts (contained in a rubber
keypad matrix) which close/open when a key is pressed/released. The contact
closures are physically offset so that the keyboard scanning computer can
measure the transit time between closure of the two contacts. This time is
used to set the midi velocity for the note.
This approach could likely be used with a 'real' piano, although I would ask
myself if it were not simpler to buy a synthesizer with keyboard in the first
place. On top of getting a keyboard that produces midi output, you can also
use the synth's voices to get the sounds/instruments you want.

In any event, I personnaly wouldn't consider pitch to midi of great benefit
for turning a real piano performance into midi, because it would be easier
to use a midi keyboard. Trying to turn a piano performance from a CD say into
midi, would of course be of interest to many musicians, since they could use
the midi information much better to learn the piece, print or edit the score
and many other applications.

> Thanks again for your consideration.

I actually enjoyed it. It isn't all that often that one gets to talk with
other people that are interested in the technical aspects involved in
solving the pitch to midi conversion problems.

Sincerely,

Bernd Kohler, bekosoft.