I agree with Bob Taylor that this question has sparked a good
discussion of issues related to the Duo-Art and its regulation.  I have
seen many of the same performance issues over the years relating to
coding and perforating errors in specific rolls, issues with very small
perforations in more recent recuts, or in some cases bleeds that were
reamed too large, so that valves were not properly actuated, etc.  And
I've seen that some modern listeners used to quiet background music
from electronic speakers with a volume knob don't like the full dynamic
range of a properly-functioning reproducing piano.  But the most common
problem I've seen in the Duo-Art that can cause the symptoms of "too
loud when loud while dropping notes occasionally" is improper or
degraded valve facings in the stack that don't seal adequately at low
stack pressure.

Fundamentally, the energy of a note played in a reproducing piano can
be considered as "force times distance" of the striking pneumatic on
the key, where force should be the stack pressure if valves are
functioning properly, and distance should be constant if the piano is
regulated properly and evenly.  In normal playing, "distance" should
only change if the soft-pedal controls the hammer-lift or "half blow"
function as is commonly the case in both Duo-Art and Ampico.  So then
the performance reproduction comes down to proper control of the stack
pressure from the coding.  In the Duo-Art with typical well-regulated
touchweight, stack pressures range from about 5-7 WI (water-inches of
vacuum) at the level 0 setting to just reliably play quiet notes, while
the loudest crashing chords at level 15 should be around 25-30 WI, near
the maximum pump output.  Intermediate levels 1-14 should map smoothly
to intermediate pressures to give a pleasing, realistic performance.
Since only the level 0 pressure is directly adjusted by the knife-valve
setting, the shape of the rest of the plot of "pressure vs. level" is
determined by the correct design and proper rebuilding of the remaining
player components.

For purposes of regulation, two different types of leaks can occur in
the stack.  Random pinholes or gaps generally cause leaks where the
airflow associated with the leak increases as the stack pressure
increases.  But most common are leaks through the lower valve seats,
where the leak can often be greatest at low stack pressures.  The
original valve facings on most Duo-Arts were brushed valve leather
(often sheepskin), installed with the "suede" or brushed side of the
leather toward the metal valve seat.  When new the material sealed
tightly at low stack pressure, because it has a very soft and compliant
surface.  With aging the leather surface can stiffen or become dirty
so that it requires more pressure to deform and seal against the seat.
Also some rebuilds I've seen have used very different valve facing
material such as smooth pouch leather applied over neoprene foam
rubber--almost the opposite of the original material in presenting a
"hard surface over a soft material" rather than the "soft surface over
a firm material" in the original brushed sheepskin.  When the valves do
not seal well at low stack pressures, the level 0 setting will be set
with the knife valve further open than originally designed to
compensate for the leak.  And then as the roll coding increases the
pressure in the stack, the intermediate levels become too loud because
the knife valve is too far open, and because at higher stack pressures
the valve facings deform and seal better, actually decreasing the
airflow due to the leaking valves.  So the effect is that a very leaky
stack might go from about 6 inches of pressure at level 0 with the
knife valve "half open" to compensate for leaking valves, to almost
full loudness of 30 inches at level 8 or so, with the pressure causing
the valves to seat more firmly and sealing off the leak, and the knife
valve fully open at that point.  And then there is little or no
expression difference between levels 8 and 15.

One quick test for this problem in a stack is to try to power each half
of the stack with suction by mouth with a roll leader in place and one
tube to the stack removed so that a note can be "played" by touching a
finger to the tube nipple.  In a well-functioning stack, it should be
possible to power the stack by mouth for 10-15 seconds or more, with
good control of the note loudness.  It is common to encounter stacks
where it is hard to power the stack by mouth quietly or for very long,
because it is more leaky at low pressure.

Ralph Nielsen
http://www.historicpianos.com/