Ref:  Ben Willis 08.08.10 - Creating New Expression Rolls From MIDI
      Bob Taylor 08.08.12 - Creating New Expression Rolls From MIDI

As some of you know, I am proprietor of Meliora Music Rolls
( http://members.aol.com/meliorarol ).

I have had some success in creating Duo-Art and (one) Welte-Licensee
rolls from MIDI files, but my approach has not utilized algorithms or
the like (which I agree would be the better approach).

I have a Steinway Duo-Art XR and a Hamburg Steinway Welte upright
that are each equipped with Spencer Chase valve systems.  I can
therefore add expression coding to a file and then immediately play
back what I have done for evaluation purposes.

I use Richard Brandle's "Wind" software for this work, which was
designed specifically for creating music rolls; it also boasts several
keystroke shortcuts for coding Duo-Art rolls.   Expression coding for
other roll formats is a bit more time consuming, but it can be done:
Wind supports Ampico A, B, Duo-Art, Welte-Licensee, Welte T-100, and
all Recordo formats.  Wind has its own file format, but you can input
MIDI files.  MIDI velocity information is lost on import, but pedaling
information is retained.

My approach might be termed the "W. Creary Woods" method --
I listen carefully over and over to the MIDI performance and attempt
to approximate "on the fly" the dynamics and pedaling of the original
performance until it sounds "right" (or until I give up!).   I also
refer to the MIDI velocity information for notes but not slavishly so.
The results seem good to my ears and to many of my customers, but it
would be far better if someone could develop software that would accept
the MIDI velocity data and convert it to reproducing piano coding.

The obstacle (and a formidable one) is that in MIDI, each individual
note is capable of independent velocity information, while with a
reproducing piano, only one half (bass and treble) of the keyboard
can have a specific velocity at any given point in time.  If you had
a MIDI file wherein a number of notes with different velocities are
being played at the same time, how would the conversion program know
which of the several simultaneous velocities to choose for the
reproducing coding?

We know how reproducing roll editors dealt with this problem; they
slightly advanced or retarded the note on which an accent was intended
that it could be assigned a different velocity than the notes that
otherwise would have "hit" the tracker bar at the same point in time.
It would seem to me that the best a conversion program could hope for
is to approximate the MIDI velocities, but that manual editing would
always be needed to finalize the file.

I agree with Bob Taylor's comment that Duo-Art coding might be the
easiest to begin with in terms of conversion software, but its major
complexity involves compensating for "droop."  The Duo-Art does not
have "regulated" vacuum to the stack such that each additional note
being played on a Duo-Art causes overall vacuum loss to the stack which
must be compensated with by seemingly "overcoding" the expression
coding.  Anyway, a discussion of the engineering shortcomings of the
Duo-Art system is beyond the scope of this posting.

It would be nice to get some activity moving on this front.

Tim Baxter
Atlanta, Georgia, USA
http://members.aol.com/meliorarol