George Bogatko is quite right: electronic media are quite probably
the worst archival means since Emile Berliner's chocolate discs!
Replica paper rolls are unquestionably the way to go to achieve
100+ more years.

What my earlier posts on MIDI files (MMDs 030415 and 030416) were
getting at was a means to make this possible in a sharable manner, so
anyone's scans can operate any perforator.  It's fine to be able to
scan a roll and recreate its master, but how much more valuable (to
the hobby, not individual businesses) for any scanned roll to be able
to drive any suitable perforator!  Widespread availability gives much
greater chance of rolls being replicated as we would wish.  Many
members of Rollscanners will freely share their scanned rolls for
such purposes.

I still think that use of MIDI format files, as proposed by Richard
Stibbons, is a good way to achieve this.  True, a roll master isn't
quite what MIDI files are designed for, but it's pretty close, and
it's possible to represent a roll in MIDI with no ambiguity and 100%
accuracy if a few simple rules are followed.  This isn't a 'mistake'
-- it's building on good practice in an intelligent manner to achieve
the goals stated above.

Because MIDI is a widespread standard with reams of freely-available
file-handling utilities, it makes a good base for a roll master format.
It will last a long time to come, and no vendor's whim will render it
obsolete.  It's easy to read and write (my software is 1988 DOS QBASIC
running on a 2002 Windows XP machine, but I could also read the same
binary file with pretty well any programming language on any computer
hardware of the last 30+ years).  Every perforator operator is likely
to be able to use 'ordinary' MIDI files already, so they're a long way
towards being able to use these roll-master files as well.  And, as
stated earlier, roll-master MIDI files transform very easily into
e-rolls MIDI and simulated-dynamic 'proper' MIDI.

Whatever the above advantages, MIDI is certainly not the only standard
file format that could be adopted.  For instance, the generic
self-defining file format of the moment in XML, which (I understand)
contains details of its data formats as well as the data itself.  Roll
masters are 90% there in MIDI but would have to start from scratch in
XML.  However, XML is readable as plain text so is more up-front than
MIDI's efficient but complex binary format.  Any other suggestions
of general formats that would suit? The test of whether a file format
is doing the job that's required is that a roll can be translated
ad infinitum between it any other roll-master file format with no loss
of data.

The aim is not to preserve the roll data in perpetuity using electronic
master rolls, but to propagate the data widely enough and for long
enough for the rolls to be replicated by perforating.  However, if the
files do survive, they need to remain readable from first principals
without access to original reading software.  MIDI can certainly meet
this requirement because the underlying format is widely documented in
print.  The 'roll master' rules could even be given in text comments in
each and every master roll.

What I didn't spell out in those posts was that my interest in all this
is in the perforating of new rolls! I was fortunate enough to obtain an
88-note Themodist / Duo-Art perforator a few months ago, and have every
intention of using this to cut replica rolls.  I have so far held back
from making any offerings solely because, although the machine produces
superb analogue copies from Rollscanners scans, they are still prone to
all the rounding errors that are well known.  The higher scan rate means
that chords don't get 'spread' randomly like older direct mechanical
copies, but they still have to be rounded to the nearest punch row.
The result is beautifully even chords at the wrong point in time!

Even when a replicated master roll is available, the perforator
clearly has to be able to make use of it.  In my case, the perforator
is stepper-motor driven at 215 steps to the inch.  It has UK
Themodist-size (i.e. small) punches, and is operated at 6 steps to
the punch row, giving 36 rows per inch.  If it is driven at 7 steps
per inch it gets to 2.5% of the '63-ratchet' Duo-Art rows (63 steps in
2 inches), which is close enough to be well within wind-motor tempo
setting tolerance.  However, the punches are sized for bridges created
by skipping two 6-step rows, and a 7-step bridge is a little too big
even though a trial roll plays on all instruments so far tested.

So, replicate the pattern with the right bridge size but with the roll
15% short (so requiring a roll of tempo 70 to be played at 60), or get
the length much closer but with wide bridges? Or redrill the punch head
for Duo-Art replication? When answering, remember that an accurate roll
opens tracker bar ports at the right time relative to others to achieve
the correct performance: the pattern is more important than the exact
length, because lengthways scaling of the pattern can be compensated by
a comparative scaling of the roll speed.  Absolute length does matter,
of course, in matters such as inter-note gaps vs. port height, and
paper acceleration, but on the whole is a second-order effect
musically compared to the perforation pattern.  I know, the answer is
to redrill!

Julian Dyer

 [ I would guess that the data used to punch 90% of the piano rolls
 [ recut in the last 10 years are stored in the disk file formats
 [ created either by Richard Tonnesen or Wayne Stahnke.
 [
 [ The formats are very straighforward (and quite similar, actually)
 [ and they are not dependent upon any special reading program, nor
 [ upon any special perforator.  The files contain only the mechanical
 [ distance data needed to recreate the original roll (within the
 [ limitations of the roll transcribing system).  File format details
 [ are available from the authors on request.
 [
 [ Published and mature programs which read and operate on these files
 [ include "Wind" and "WindPlay", by Richard Brandle, and "View" and
 [ "Audition" and "Punch" by Wayne Stahnke.
 [
 [ -- Robbie