Craig Brougher's comments ascribe acceleration in computer-playback
performances to be due to losing the speed drop during playback
produced by wind motors.  It may well be true, and is an example of
the fiddling little details you have to consider in a simulation,
but is surely at worst a marginal effect.

I was actually referring to Wayne Stanke's note in MMD 98.11.01:

 "The recording machine at the Ampico studios in the early days pulled
  the paper using a capstan, or constant-speed roller, with only a
  single layer of paper on its surface.  This arrangement caused the
  paper to feed at a constant linear velocity.  By contrast, the
  finished production music rolls were accelerated on playback."

This sort of acceleration is surely greater than any drop in speed
as the load on the wind motor increases.  Indeed, a good motor should
have enough torque so maintain near constant speed under ordinary load.
I was expressing surprise that such an device should have been used,
given that other companies made acceleration corrections.

It is possible to prove that Duo-Art rolls were recorded onto a spool,
so automatically compensating for speedup (as recording and playback
will have identical acceleration of paper speed, which therefore
cancels out).  The reiterating perforator that recorded the notes
operated at a fixed number of perforations per second, so as the paper
speed increased, the separation between each perforation increased.
There are surviving rolls that show this behaviour.  These variable-
spaced perforations were converted to the familiar fixed-space
perforations for production cutting.

Interesting, though, Duo-Art dance rolls have a fixed spacing per beat,
constant from start to end of the roll.  (Go and measure some.)  They
therefore accelerate as the take-up spool fills.  This matches the
theory that dance rolls were not directly recorded, but used the note
record as the basis of a fixed-tempo 're-creation' by the roll editor.

As strict tempo is needed for dancing, gradual acceleration of the beat
through the roll was perhaps less damaging than randomness from beat to
beat.  (I know how much dancers complain when a band cannot hold the
beat!)  Manually adding acceleration compensation into a master would
have been uncommercially time-consuming, although some Frank Milne
marches actually do it!  (Rex Lawson has counted the punches per beat
of lots of rolls, and I rely on his word for this.)

So, we have evidence that Duo-Art produced both acceleration-compensated
classical rolls and uncompensated dance rolls, in accordance to the
different needs of each market.  I would have expected Ampico to do
something similar.

If Ampico classical rolls were recorded using fixed paper speed, they
will show intrusive acceleration, especially on long pieces.  Do they?
The shortish dance rolls will accelerate a bit, but not (on the whole)
by an annoying amount, as Robbie commented.  The Duo-Art evidence
suggests that _dance_ rolls cannot necessarily be cited as evidence of
the process used to produce _classical_ rolls.

Wayne's description as quoted above is intriguing.  So much of the
writing on Ampico has been about the Model B recording process, but the
vast majority of rolls weren't produced that way, and it would be nice
to know more about the earlier process.

Julian Dyer