The two comments about Welte that followed my previous post [120601
MMDigest, "Piano Roll Arrangements vs. Transcriptions"] missed the
point of what I was trying to say, so I'll spell it out more fully!

What I said was "Aeolian were almost certainly the only reproducing
roll manufacturer that tried to create a playable roll in real time;
this included perforating both the notes and the expression coding."

This statement is backed up by hard evidence in written form by those
who performed the process, and in the surviving 'Original' rolls.
The key point was "playable roll in real time".

The responses included:

(a) "In 1904 Welte could and did record full expression while the
master roll was being marked," and

(b) "The only truly 100 percent hand-played reproducing piano rolls
I know of are those marvelous red paper Welte-Mignon gems, where
mistakes [...] are reverently preserved".

Now, what are we to make of this?  Firstly, although you'll hear (a)
said quite often, I really don't think it rises much above urban myth.
What might "full dynamics" mean?  There is nothing to suggest that Welte
recorded individual note dynamics, and even if they had it would have
been little use to them because they'd have had to convert them to
dynamic coding.

To do this we have the evidence from the Ampico B some 25 years later,
or from AMICA's Leo Podolsky project, that such conversion was a
heroically complex undertaking.  The general inference from the rate
of Welte's roll issues is that coding can only have taken perhaps a day
per roll: this strongly suggests that if they did make use of recorded
dynamics (for which there is no evidence) it was in a form that
moderately closely matched the coding.

Every reproducing roll mechanism posed the same musical dilemma to roll
editors: how to average out the dynamics of each note played into just
two streams -- bass / treble or Theme / Accompaniment -- depending on
the system.  This process was made explicit by Aeolian's recording
device: how better, they must have figured, to do this averaging than
with a trained musician when the pianist was playing?  They found out
soon enough that significant post-recording editing was needed, which
I've written about before.

We know from the Nelson Barden interviews with Ampico roll editors
that, for almost all its rolls, Ampico didn't record dynamics at all
and had the editors create the dynamic coding later (this fact is
seldom acknowledged).  The latter spark chronograph device to record
individual note dynamics suffered from the need to ditch almost all the
data it captured and at a guess was not made much use of, given that it
took a huge time to achieve results little if at all better than Edgar
Fairchild could manage without external assistance.  It's odd how the
existence of this later gizmo is not infrequently used to validate the
earlier recordings!

There are probably loud tut-tutting noises coming from some quarters
about my "no evidence" statement for Welte, but from a hard historical
point I think it's quite correct.  The whole debate about Welte
recording seems to hinge totally around comments published by Richard
Simonton; notably, Edwin Welte, who really did know, never said a word
on the subject beyond that the process was "quite simple"!

There is no patent, equipment or 'recording' roll to refer to.  There
has been huge speculation for 60 years now, but this has no value as
historical evidence.  Much effort has gone into proving that certain
hypothetical devices could work; this doesn't inform us in any way
about Welte's actual activities.  (My guess, and I'm not alone in this,
is that whatever device existed was much simpler than these hypotheses.)
Oddly, it seems that we need some technology to validate the evidence
of our ears that these are superb rolls.

As for (b), all it tells us is that the companies had differing editing
policies: if a fluff is evidence of accurate recording, all I'd have to
do is edit some in.  The fact that some mistakes are apparent in Welte
rolls is not evidence that all fluffs were left unedited.  Whether the
other companies edited out the musical personalities of their pianists
is not for this mostly technical debate.

But, to take you back to what I said to start: the point was the
creation of a playable roll in real time.  Was this a good or bad
thing?  Did it make for better or worse results?  Aeolian decided to
work that way, but the other reproducing roll companies decided to use
devices that marked rolls of paper with ink.  What was the state of
play around 1910?

The advantage of a playable roll is obvious: it could be played to
the artist and corrected there and then, with all the advantages of
immediacy.  One, perhaps subtle, disadvantage is that to make it into a
master roll requires it to be re-sampled to the production perforator's
step rate -- what today we'd call quantisation -- but of course the
recording had already been quantised to the recording perforator's
rate.  Therefore, there was an inevitable loss of timing accuracy in
the process from this double conversion.

This loss of accuracy was compounded in early days by selection of
coarse perforation steps that magnified such errors (but churned rolls
out faster so kept costs down).  Thanks to the roll-scanning of recent
years it's easy to see that German hand-played rolls typically use 50
steps per inch, while American ones use either 20 or 30.  I suspect we
sometimes blame the recording process for what is really a production
shortcoming.

An ink-marked roll has the advantage of being quantised only once, but
the disadvantage that somebody has to decide where the start of each
mark is.  This is actually quite hard to do, because the marker takes
a finite time to operate.  When QRS reintroduced their recording piano
in the 1970s the results were rather variable, and indeed some such
rolls were re-mastered in later years.  It's a moot point whether
either system is innately superior: I suspect it's the skill and
experience of the operators that counts.

However, original companies wouldn't have based their decision on such
matters: if you read the patents on the subject, the issue was more of
a technical one.  The solenoids needed to operate reiterating
perforators from electrical contacts under the keyboard were 'sticky'
because these early designs were bad at letting go when the current was
turned off.  Pneumatics were easier to deal with (the Hupfeld studio
photographs seem to show that Hupfeld's recording piano was pneumatic),
but the smaller solenoids needed to operate a stylus were faster than
those to operate a perforator's interposer.

Also, reiterating perforators are heavy and noisy while many companies
(Welte included) wanted a portable recording device that could sit
besides the piano.  Indeed, QRS made great play of taking their
recording piano around the country and having all the local pianists
come in to record in each town it stopped at; there are many articles
about this in the Music Trades Review.

All this was spurred by a question about whether companies were really
recording hand-played rolls in these early days: they definitely were,
and indeed it was a well-established area.  We know quite a lot about
it thanks to direct evidence from the time, other than for the
Welte-Mignon system whose secrets seemingly went to the grave with it.

Julian Dyer -- who doesn't have to go to work tomorrow thanks to
the Queen's Golden Jubilee holiday, hence the great length of this
posting...

P.S.: As has been said before, American-made dance rolls are almost
never 'hand played' in the sense of the above, but are in strict tempo
and were mostly produced on a marking piano regardless of the name on
the label!  A performance was certainly involved in some cases, but was
used less directly in the arranging process to the degree of fidelity
the editor preferred.  On the other hand, German-made dance rolls for
Welte and Hupfeld are often truly hand-played, and sound very different
as a result.