The Duo-Art recording process
How were Duo-Art rolls made? Here's a brief overview of how classical ones were made. Popular music was treated differently, see later.
Making the recording
When the pianist sat down to play at the recording piano, the notes were captured directly onto a paper roll using a reiterating punch (in Aeolian's terms, the "original" roll). At the same time, the recording editor would manipulate controls to create the dynamic codes for the accompaniment and theme - the musical background and the accented notes respectively. This can be seen here, where Reginald Reynolds is recording Max Darewski's playing in the London studio. The two dynamic control levers are clearly visible on the console.
The dynamics, as far as the editor could capture them, were therefore cut into the roll directly in the form that the playback instrument would require (of course, there could have been a number of "trials" for the editor to learn the performance before the final recording attempt, although evidence tends to suggest this wasn't much done).
Once the Duo-Art recording session was ended, then, all that was missing on the Original were the "accenting" controls that would switch the system from accompaniment to theme level. There is some indication that attempts were made to record these during the performance but that such efforts were rapidly abandoned, leaving the accents to be edited in by hand afterwards.
Incidentally, Aeolian were unusual, if not unique, in even attempting to produce directly-playable rolls from their recording machine. All other reproducing piano systems chose to add all dynamic coding to the rolls later as part of the editing process. Aeolian were also unusual in using a reiterating perforator for recording, the normal approach being to record with pencil or ink marks on a paper roll which were subsequently punched out by hand.
Editing the recording
The means of editing an Original Duo-Art roll is illustrated below (this being the Duo-Art Original roll of Rudolph Reuter playing Grieg's "Wedding day at Troldhangen", recorded on 5th September 1919, but never issued). Here, the notes to be accented have been identified in blue pencil, and marks made in the margin to indicate where accent perforations are to be placed. These have subsequently been punched in by hand, using a single standard-sized punch, not the small "snakebite" paired-up punches seen on the final production roll.
The black patches are characteristic Aeolian corrections, here blanking-out unwanted perforations where the pianist didn't lift the keys enough to break the electrical contact that triggered the reiterating punch which cut the original roll during the performance. Some work has also been done to the pedalling, altering its onset to avoid catching the wrong notes and extending notes that need to be held. Other than this, little has been done to the actual performance on this particular roll (but it's not necessarily indicative, because it was never issued so may well not have been finished). It was usual to modify the dynamic coding to a greater or lesser degree to achieve the correct reproduction on a studio piano.
The steps between punches gradually increase on an original roll indicating how its speed accelerated as paper built up on the takeup spool: this automatically compensated for the same acceleration at playback, but required the roll to be converted for production purposes where the punch step rate remains constant throughout the roll. This conversion introduced some timing error, particularly if the production step rate was too large. It was done when making the "Stencil" (Aeolian's term, more commonly known as a master roll) for the rest of the production process. Stencils were much longer than production rolls, so that each individual perforation was discrete instead of being overlapped to form slots, with edge sprockets to permit synchronisation with the production machinery.
Although it would seem more sensible for the editing to have been performed on a Stencil, in case it was necessary to revert to the Original, evidence such as the above shows that at least some of the editing was routinely performed on the Original.
The roll artist was expected to take part in the editing process to make sure that the final roll was correct. Some did, some didn't, although Aeolian's contracts required them to do it, and was clearly a substantial help in getting the resulting rolls to sound right. Here, Percy Grainger can be seen with W. Creary Woods in a somewhat contrived picture showing him checking one of his rolls. Any edits that resulted at this stage would have to be performed on the master Stencil.
The final step was, of course, to perforate the rolls. This was done either in America (at Meriden, Connecticut) or in England (at Hayes, just west of London). The picture shows the Hayes factory a few years before the introduction of the Duo-Art. The perforators read the Stencils mechanically and wore them out, so new production Stencils would have been made from time to time from the masters, which could be done exactly without introducing any errors so rolls remained consistent throughout their production life. Aeolian also sent Stencils from New York to London as the majority of roll were made in both factories: some recordings made in London were exported back to America, but there is no way of telling which these are. The Hayes Stencil store was a separate fireproof building, with each master Stencil in a separate earthenware pipe - if the Stencils were lost, the company would be out of business.
Sadly, almost none of the Original rolls or Stencils have survived for the Duo-Art, either in America or England (the same is true for almost every make of piano roll, come to that). This is the reason that it has been necessary to create machinery and software that can take a production roll and "reconstruct the master roll" from it - which is not as easy as it might sound at first - so that modern replicas can be every bit as good as the original production rolls.
Popular Duo-Art rolls were made in a totally different manner to the above. In the era of the player piano, 'Popular' music meant 'Dance' music, which was indeed danced to. Dancers need an absolutely regular beat, and after a very short period it was clearly decided that using the recording piano as described above gave poor results as the length of musical beats would inevitably waver by at least one row when the music was divided into perforator-advance rows (in modern terms, the process of creating the Stencil 'quantised' the music into perforator-advance rows).
For a roll, the dancers' need was that the beat was always the same number of rows long, so the practice was for specialist popular-roll editing staff to create stencils directly. This was done either by marking the notes directly onto blank Stencils, or by use of a special 'marking piano' that created stencils one row at a time. This was exactly the same method that was used to make rolls for ordinary player-pianos (and indeed for almost all other roll-operated instruments). The illustrations are from slightly before the Duo-Art era and show the production of 65- and 88-note rolls at Aeolian's London works: here, the roll editors can be seen marking the stencils, but there is no indication of how the marking piano was used.
Rolls produced using this method are usually known as 'arranged' rolls, and are musically perfect with each beat divided into a specific number of rows, the number selected so all the smaller note values divide exactly into whole numbers of rows. Every note therefore starts at exactly the right time, exactly what is needed for dance rolls. A common value is 24 rows per beat, which usefully divides into 3s and 4s. Perforating such a roll with 20 rows per inch typically gives a good musical tempo for dancing, so this pattern in the most commonly seen. By comparison, Duo-Art rolls of hand-played classical music mostly used 30 rows per inch to give better time resolution. This additional resolution was not needed for arranged rolls because of their exact conversion of the music into punch rows.
Whether these arranged dance rolls represent an individual's hand-playing is a moot point. The general impression given by such evidence as exists is that for 'name' artists it was not uncommon to make a true hand-played recording and then construct a strict-tempo roll using the recording as a guide. The authenticity of the result was entirely in the hands of the roll editor, who could interpret the recording as they saw fit. There was no particular need to use a real recording, and many rolls were constructed directly by the roll editor - unsurprisingly, this was a specialist art with few practitioners. For Duo-Art the most famous name was Frank Milne, who was noted for his ability to draw the roll on graph paper at home with no piano present. This production method was used by almost every American roll manufacturer for popular rolls, whether for reproducing pianos or 88-note instruments.
Julian Dyer, May 2007.