George Bogatko wrote in MMD 010609:

> This is what Dave Saul and Wayne Stahnke do.  Wayne's method
> determines that the slot is 8 holes, with a 1/4 inch punch and
> a step rate of 1/8 inch.  He creates a computer data file that
> says precisely that.  Dave's perforator reads this data file,
> and cuts the roll as 8 punches, with a 1/4 inch punch and a step
> rate of 1/8 inch.  Done.  Perfect match.  At a competitive price.

I agree with your thinking.  You are right.  This is what is needed
to be able to duplicate the roll reliably.  Anyone who can punch rolls
using the information from the computer data file described above would
come up with the same roll.  But is the computer data file for the
punch an accurate description of the real master?

 [ An equivalent and necessary question is, "Has _all_ the data
 [ been recovered from the old roll?"  The answer is found when the
 [ replicated copy is closely compared with the old roll.  If they
 [ match exactly, hole-for-hole, then all the data has been recovered
 [ to 100% accuracy.  No more is possible.
 [
 [ Since old production copies can have errors, Wayne Stahnke tries
 [ to borrow at least three old rolls, hopefully punched on different
 [ perforators or in different productions.  If the same error occurs
 [ in all three then he concludes that the error actually existed in
 [ the old master roll controlling the perforator.  Should this error
 [ be corrected?  Ah, that's a difficult decision!  ;-)  -- Robbie

I am more interested in replicating player piano rolls.  By using the
example of 1/8th inch steps for the paper advance makes me think George
is making rolls for a smaller instrument.

One example is a Duo-Art roll that I have that needs an extremely quick
acting sustain.  At times, the sustain must be turned off for all notes
in a distance of about 2 chain distances.  It works on the grand piano
but has problems on the upright pumper.  I believe that this roll is a
copy of a copy and that the sustain holes have grown together because
of copying errors.  I don't have this problem with any of my other
rolls.

I don't know if all master rolls use tracker feed holes on the side of
the paper for registration or not.  I will assume that they do for this
discussion.  The holes in the master roll are punched in such a way
that for every cycle of the punch, the master roll will increment to
the next set of holes.  The next set of holes are positioned exactly
in front of the reader for the next read and punch cycle.  There is no
guessing where the holes should be.

Software can be made to take this master roll, read it, and convert
it into a very accurate replacement of the master roll.  Making a copy
from this master roll computer file would create a copy of the roll the
same as the original master roll would.  This is what is needed for
replicating music rolls.  The problem is finding these master rolls.
Because of this, 1st or 2nd generation copies will be used for scanning
many of the rolls.

Take a look at one of the long slots found on a piano roll.  The punch
used to make the holes is usually round.  The edge of the slot might
look smooth, but look closer.  You might be able to tell how many punch
cycles were made to punch that slot if the number is 24 cycles/inch or
less.  Try to count the partial moon shaped sides if you can see them.
Compare old rolls to new ones.  This indicates to me that the punched
paper is crawling at a snails pace compared to the master.  To get back
the accuracy that the master had, you must rely on the scanner
accuracy.  But you are scanning the copy -- how good is that?

The other variable on punching of the copy is now far the copy is
incremented for each punch cycle.  This will dictate the tempo of the
roll.  It also varies the slot lengths for each note or track.

Now you try to make a computer data file from the 1st generation copy
for punching new copies.  You are relying on the software programmers
and the roll reader accurately to create the computer data file.

1.  Your scanner supports 100 Dots Per Inch (DPI), or .001 inch per
scan line.

2.  Hole 25 should start playing at position 12.12 inches.

3.  Your punch increment is 20 steps per inch, or .050 inch per step.

A fudge factor has to take place because the punch increments are
not fine enough to duplicate it exactly.  The punch slot is started at
12.1 inches and not 12.12 inches.  This is .02 inches earlier than the
original.  This is what the programmer would have to do since there
isn't an exact alignment.  This also happens at the end of the slot.
It might be another .02 inches later than the other, making the slot
.04 inches longer.  With a normal player piano punch hole size of .076
inches, that is over a 1/2 punch hole difference.

Is this accuracy needed?  Probably not on 88-note rolls.  But I would
like to know that a more accurate reproduction was made.  If this was
the accuracy seen for a reproducing roll, I might think again and not
buy the roll.  In the above example, the duplicated roll timing might
be off so much that it won't play on 90% of the players but the
original will.  The problem is that I will never know the differences
unless I have the original scanned roll to compare the copied roll too.

Ideas:

Making the cutting knife razor thin is not a viable solution.  Changing
the punches feed rate to match the scanner's DPI rate would be best.
The software for the punch could be written to still advance the paper
at a good clip but go into a micro-step mode in areas that didn't line
up.  Paying extra attention and time where the holes started and
stopped by punching the holes at exactly the position seen on the
scanned image.  This would make it possible to punch the reproducing
rolls accurately.

As far as a punch with a fixed paper advance value, you might be able
to slow the playing tempo of the master roll through software while
punching at the normal rate.  This would effectively cause the tempo of
the duplicated roll to be increased such that the fudge errors would
not be as large.

Other things that could be done is write software for the specific
player type (i.e., Duo-Art, Ampico, etc.).  Put rules around the punch
logic that would flag problem areas.  If the parameters for accenting
the specific notes were out of specification and didn't match the
original accent from the scanned data, show the problem areas and maybe
just punch 10 seconds on paper around the problem area.  That way you
could hear if it is a real problem.  This assumes that your player is
operating like 90% of the known players.

I know that these types of issues are being discussed by the roll
scanning group but this thin thread on the MMD seems to overlap the two
groups.  Taking time to read all of their archives and seeing what has
been defined is difficult for me to follow at this time.

I understand all of the hardware issues but have not found any
documentation on the software systems being developed.  I haven't
found any file format specifications for the rolls yet.  I would have
thought this would be some of the shared information for understanding
these types of issues.

Pete Knobloch (Tempe, AZ)

 [ I understand that the Rollscanners group has software production
 [ activities underway.  I invite a report to MMD!  -- Robbie