Contents |
|
![]() |
|
Start of a 20-note organette tune sheet. |
This page gives a brief description of some of my efforts to preserve the music that is recorded on vintage paper rolls for mechanical organettes.
The idea of storing information as holes in strips of paper has been around for a very long time. Paper tapes and cards were used as digital storage media in weaving looms in the late 1700s, in telegraph machines in the 1800s, in tabulator cards around 1900, and in digital computers in the 1950s and 60s. The paper music roll uses exactly the same principles, except that the information is usually processed pneumatically rather than mechanically or electrically.
Paper organette rolls have an important place in history, as they were the first affordable media for the mass distribution of recorded music to the general public. They were in everyday use for a period of over 40 years, from their introduction in the 1870s until superseded by the phonograph during the 1910s and 20s. Paper music rolls for the Pianola or "player piano" continued the technology into the 1950s, and are still being manufactured today.
Original organette rolls from the 1880s have not survived in great numbers, and many of those that have are in poor condition. The illustration opposite shows three typical Celestina rolls in various stages of decay. Despite their poor condition, these rolls still contain recoverable music that is an important part of the historical record. They provide a snapshot of the musical culture from the 1850s to the early years of the 20th century, including traditional songs, operatic songs and choruses, popular and comic songs, music-hall songs, dance tunes, religious songs and hymns, and military songs and marches. Oganette rolls provided the music for the Saturday dances in many community halls across the country, and for the church services in the same halls on Sunday mornings. Most importantly, these paper rolls can still reproduce the music of the time exactly as our ancestors heard it in their homes on their mechanical organettes, long before the phonograph became a domestic item during the 1900s.
Since the 1990s I have been working on several hobby projects which aim to preserve and increase my supply of organette music by:
While it has been very rewarding to use computers and modern electronic technology to preserve and re-create this music from the 1880s, it is worth remembering that the craftsmen who produced the original music rolls never had the benefit of anything electrical - not even an electric light!
The "Musical Instrument Digital Interface" or "MIDI" is a well-established standard for handling musical data in electronic form. Its basic features include:
The MIDI system is very easy to use, but the internal workings of its data transmission and storage formats can become quite complicated. The full technical specifications are published on the web site of the MIDI Association. There are many unofficial MIDI explanations and tutorials elsewhere on the Internet.
In order to preserve the music as it was originally recorded, we need to know a couple of things about how the original rolls were made.
The process begins with a music arranger or editor (a real person) who takes a published tune and adapts it to suit the scale and compass of the target instrument. The editor translates the pattern of notes into a corresponding pattern of holes, which are cut out by hand to form a "master roll" or stencil. The master roll is then used to control an automatic perforating machine, which mass-produces multiple copies of the paper "tune sheet".
The perforator consists of a row of punches of a size and spacing to suit the target instrument, mounted at right angles to the path of the roll paper. The machine operates on a repetitive cycle in discrete steps. On each step the master roll selects the punches required, the holes are punched, and both the master and the roll paper advance to the next step. The distance advanced is controlled mechanically and is fixed for each type of roll. A typical 20-note organette roll is about 6m (20ft) long, plays for 4 minutes, and requires about 3000 steps. At a maximum speed of about 10 steps per second the perforator would take 5 minutes to produce the roll.
The Pianola Institute in the UK has some excellent illustrations of this process applied to organ and (later) piano rolls around 1880-1910. Note how the machinery is all driven from overhead shafting, most probably powered by a steam engine in a boiler house outside.
There are two important consequences of the stepwise operation of the perforating machinery:
The most basic method of converting a music roll to MIDI is to transcribe the notes by hand into an on-screen MIDI editor.
The first step is to examine the roll carefully to determine the punch step distance and the number of punch steps per beat. The step distance can sometimes be measured directly by looking for evidence of successive punch strikes along the edges of the holes or slots. If the punches are not visible the step distance can be found by measuring the differences in the lengths of a range of the shorter note slots.
There is usually a simple and sensible reason for the chosen roll dimensions, remembering that in the 1800s most things were measured by proportion rather than by ruler. The designers of the Celestina, for example, initially chose a "tracker bar" spacing of 4 notes per inch across the roll. This was divided in the proportion 3:2 between the the note slot and the separating space, and the slot width and the step advance were both made one-half of the slot length. So the step distance is 1/4" x 3/5 x 1/2 = 3/40" = 0.075"
The number of punches per quarter note (PPQ) can be found from the regular bass notes which usually mark the start of each bar. Say a Celestina tune has about 60mm (2.5") between the regular bass notes. Dividing this by the known step distance (0.075") shows that there are 32 punch steps per bar, most probably representing 4 quarter-notes of 8 steps each. The time signature follows, depending on whether the space between the bass notes divides naturally into 3 or 4 parts. Similar calculations can quickly produce a table of bar lengths and corresponding beats per minute for all of the common time signatures and PPQ values. A Celestina roll with a bar length of about 35mm, for example, indicates a tune in 3/4 time with 6 punch steps per quarter note which plays at about tempo 135.
My next step is to use a computer drafting program to draw three or four bars of the original punch grid, which is then laser-printed onto overhead transparency film. My supply of film is left over from the 1970s, but similar material is still available new. The illustration shows a typical grid for a Celestina roll in 4/4 time at 8 punch steps per beat. (Click the image to download a full-size PDF). The smallest divisions are the single punch steps of 0.075", here representing 32nd-notes. Eighths, quarters, and bar lines are also marked. The note names are shown twice - the outer column gives the actual notes, while in the inner column the notes are transposed up 4 semitones to allow the amateur musician to think in a more familiar scale. The text is reversed for durability, so that the transparency can be used with the printing on the under side.
If this grid is placed over a well-made and well-preserved roll, the note starts will all appear exactly on the lines. The note ends may be just beyond the lines, as the punches are usually made slightly longer than the step distance so as to cut continuous slots on consecutive strikes.
The next requirement is for a computer with a MIDI editing program which allows notes to be entered and edited in a "piano roll" view. The editor must have a "snap to grid" function that allows each beat to be sub-divided into a fixed number of steps, which can be set to correspond to the punch steps per beat on the paper roll. Then (in theory) all that remains is to copy the notes from the paper roll into the same positions in the roll image on the screen. As the notes are drawn they will "snap" to the set divisions, and will thus correspond exactly to the original punch grid. The final result can be saved as a standard MIDI file, which will be a true electronic representation of the original roll.
The MIDI software sub-divides the beat according to an internal
"virtual" clock, which must be set (for each tune) to a fixed number of
ticks per quarter note (TPQN). The actual TPQN number is arbitrary,
but 120 is a convenient value as it allows the quarter note to be
divided into 2,3,4,5,6,8,10 or 12 parts, corresponding to the common
punch steps per beat. The relation between the physical punch steps
and the virtual MIDI clock ticks is:
Ticks per quarter note (TPQN) divided by
Punches per quarter note (PPQ) = MIDI ticks per punch (TPP)
So if a tune has 8 punch steps per quarter note, and the MIDI editor
is set to 120 clock ticks per quarter note, there will be exactly 15
clock ticks per punch step: TPP = TPQN / PPQ
Although the time signature (4/4, etc) is important to a human player, it is largely irrelevant to the MIDI transcription. The differences in perfomance (emphasis and rhythm) between (say) 2/4 and 4/4 have already been encoded into the roll by the original arranger. It makes no difference to the actual pattern of notes or holes whether the tune is transcribed as 2/2, 2/4, 4/4, or even 8/8. It is more important just to find a convenient division and a matching grid, and then to adjust the tempo to suit. Time signatures sometimes need to be adjusted during a tune, for example by including a bar of 5/4 (or whatever is necessary) to maintain grid alignment over a ritard.
In the illustration at the right (click to enlarge) the left-hand side shows part of a Celestina roll placed on a coloured background and overlaid with the appropriate grid transparency. This 4/4 march tune has 6 punch steps per beat, and it can be seen that the notes are (almost) all perfectly aligned to the grid. The right-hand side shows the corresponding pattern transcribed onto the screen of a MIDI editor. Note that the screen display includes empty lines for notes that are missing from the 20-note scale.
There are many suitable MIDI editor programs. I use a version of Cakewalk which is over 20 years old, but which still runs fine on Windows 11. I have purchased later versions of Cakewalk (now called Sonar), but these seem to focus on audio production to the detriment of the MIDI editing functions.
An electronic roll reader or scanner can speed up the process of transcription by automatically copying the contents of a roll directly into a MIDI editor. The scanner contains hardware to detect the presence or absence of holes, and software to translate this information into a standard MIDI file. The scanner will make an exact copy of whatever is presented to it, including all of the wear, damage, and distortion of a typical organette roll. Manual editing is still needed to correct these defects, to realign the scanned notes with the original punch grid, and thus to recreate the original roll.
Music roll scanners do not actually exist as commercial items - they have to be custom-made for particular applications. Many different types of mechanical, optical, and pneumatic scanners or "roll readers" have been built by individual hobbyists in recent times, and these have been used to copy large numbers of (mostly) pianola and band organ rolls.
My adventures with organette roll scanning started in 1998 using a "contact image sensor" (CIS) from a discarded document scanner. The CIS is a small glass-topped module about the width of a sheet of office paper, but only about 15mm square in cross-section. The paper to be scanned passes over and in contact with the glass. A row of internal LEDs provide illumination, and a line of tiny lenses focus the reflected light from the document onto the sensing pixels below.
The sensor used has a resolution of 200 pixels per inch, or 1750
pixels over the total width of 8.75 inches. The paper is drawn across
the sensor by a stepper motor and roller also making 200 steps per
inch, giving a resolution of 0.005" (0.13mm) in both directions. The
sensor and motor are driven by home-made software running on a 486 PC
under MS-DOS, and are interfaced through the parallel printer port
- high technology for 1998!
This close-up illustration (click to enlarge) shows the leftmost
couple of inches of the CIS and the feed roller, with a well-worn
14-note tune sheet passing between them. (For scale, the green LEDs
are 6mm apart). At each 0.005" step the CIS pixels accumulate
electrical charge in proportion to the intensity and duration of
the reflected light. The pixels are then read out in sequence and
are interpreted as representing either black or white. The paper
then advances to the next step and the process repeats. On the 486
PC it takes about 4 microseconds to examine each pixel, or nearly
10 milliseconds per scan line. Storing the data, driving a
pseudo-graphic screen display, and advancing the paper add another
5-10mS, giving a maximum scan rate of about 50 scan lines (1/4 inch)
per second. Using a Pentium 166 doubles this speed.
As the pixels on each line are read out in sequence, the software
on the scanner PC counts the numbers of consecutive black and white
pixels as they occur, and stores these numbers in a file for further
processing on a more capable computer. The sample opposite (click to
enlarge) corresponds to the start of the two short notes in the
picture above, after some of the irregularities in the holes have been
corrected in the software. The first column numbers the scan lines
in sequence, while the remaining columns show the counts in the
alternating black and white areas across the line.
The first count (103) represents the leftmost half-inch of the black
roller. There are no notes in scan lines 263 and 264, so the next 1592
(white) pixels represent just under 8 inches of blank roll paper. The
remaining 55 pixels are the far end of the black roller, off-screen at
the right-hand side. The counts on each line add up to 1750.
A one-minute tune sheet generates about twelve thousand scan lines.
As the paper moves past the line of sensing lenses the first hole starts to appear at scan line 265, dividing the 1592 white pixels into two groups. The margins do not change. The second hole appears three lines (0.015") later. After a further 3 or 4 lines there is a stable pattern showing two black slots each 64 pixels wide, one starting at 275 pixels from the edge of the roll and the second at 275+64+147=486 pixels. Knowing the scale of the instrument, the width of the roll margin, and the width and spacing of the note slots, it is easy to calculate the corresponding musical notes. The length and tempo of the notes is calculated from the numbered scan lines (200 lines per inch) and the nominal roll speed for the particular instrument.
Processing of the raw scanner file is done by home-made software written in the Pascal language and running on a regular desktop PC. The software corrects minor errors in the pixel data, translates the pixel counts into notes and times, adds the necessary file and track headers, and writes the end result in the prescribed (and rather complex) format to produce a standards-compliant MIDI file. The file can then be opened in the MIDI editor for further corrections and adjustments (eg, to correct the different starting times of the two notes in the example above).
The scanner itself knows nothing about rolls or instruments - it simply counts black and white pixels, and so can be used for any type of roll that will fit within its 8-3/4 inch span. The processing software has a configuration file which contains the physical details for each type of music roll, so that the pixel counts can be interpreted correctly for the corresponding instrument. This one unit has been used for more than 20 years to scan rolls for the 14-note organette, the Celestina and Symphonia 20-note instruments, the Triola mechanical zither, the Tanzbaer accordion, and even the 12-note Rolmonica. The same configuration file is also used "in reverse" with my home-made punching machine to translate MIDI notes back into punch positions.
Regardless of the method used to produce the MIDI file, a copy should now be made for checking and editing. Files transcribed by hand usually only need "proof-reading", but those produced by scanning will generally need length and tempo adjustments first.
To proof-read a manual transcription it is necessary to go through the tune again with the alignment grid and check carefully that the notes previously entered on the screen match exactly to those under the grid. As with proof-reading text, we tend to read what we expect to see rather than what is actually present, so the process may need to be repeated several times until we are confident that all errors have been eliminated. At each stage it is important to listen carefully to the MIDI file. I also use a home-made checking program which helps to locate missing or mis-aligned notes, but the majority of the proof-reading and correcting is done manually, one note at a time.
In many cases the transcription and checking will proceed easily. There will be no doubt as to the position or duration of any of the notes on the roll, and any notes transcribed accidentally at the wrong pitch, in the wrong beat, or at the wrong place within the beat will be easily found and corrected. The problems arise when the content of the roll is not clear, and we can not be at all sure of what is actually present or what was originally intended. These problems and puzzles may appear as damaged or distorted paper, notes that no longer align with the grid, "wrong", missing or mis-placed notes, missing or additional paper feed steps, unexpected changes in timing, or differences between multiple copies of the same piece or even between repeats in the same piece. These become matters of interpretation, requiring a subjective decision as to whether what we see is "wear and tear", a deliberate musical choice, a mistake by the original roll editor or stencil cutter, or a malfunction of the machinery or the production process.
Much could be written about these problems and how (or whether) to resolve them. As one transcribes a significant number of tunes one learns many of the tricks used by the original roll editors, and many of the production problems that could and did occur. This experience can help to decide whether an "anomaly" is accidental or deliberate, and how it should be corrected. Reference to a printed score (often freely available through the Internet) can help to clarify the composer's intention. The final decision must always make sense, both physically and musically.
The overall approach to "correcting" depends on whether our objective is to make an exact copy of a possibly defective roll, or a satisfactory musical performance of the piece that the roll is intended to represent. Where questions have arisen with the MIDI files preserved on this site I have generally given priority to the music. If significant edits were required I have preserved both as-found and edited versions. (For example, see Sweet Hour Of Prayer on the 20-note organette music page).
The first step in editing a scanned file is usually to adjust the length and tempo so as to bring the bass notes into rough alignment with the on-screen bar lines.
When processing the CIS files my scanner software assumes default
values 120 MIDI ticks per quarter note and 100 beats per minute. By
a happy coincidence, these values turn out to be correct for any
Celestina tune which uses 8 punch steps per quarter note (8 PPQ):
For the Celestina, 8 punch steps x 0.075" = 0.60"
0.60" x 200 scanner steps per inch = 120 scanner steps
per quarter note.
So with 1 scanner step corresponding exactly to 1 MIDI tick, the
scanned image should align perfectly with the bar lines on the editor
screen (after adjusting the starting position). Likewise, quarter
notes of 8 punch steps (0.6 inches) played at 5 feet (60 inches) per
minute will play at tempo 100. If the tune uses a different PPQ
value, the overall length and tempo will need to be scaled in
proportion in the MIDI editor. For other types of roll it is
generally easier to scale the on-screen bar length "by eye" rather
than by calculating the necessary ratio.
Proofreading then proceeds as described above, using the editor's "snap to grid" function to align the notes.
Regardless of the theory, the paper notes will never stay perfectly aligned under the grid for any significant distance, due to 140 years of stretching, shrinking, and other distortions of the paper. Frequent minor adjustments of the grid position will be necessary as checking and editing progresses.
After all our conscientious transcribing, scanning, editing, checking, correcting, and checking again, we finally arrive at a perfect digital representation of... of what, exactly? All we can say for certain is that we have probably made an accurate copy of the roll in our hand. We can not assume that the roll we happen to have acquired is the definitive version of the composer's work, nor even of the organette company's mass-production. I have sometimes seen three or four different versions of the "same" roll with the same name and number. The differences may be minor, such as occasional changes in note lengths, or more substantial, with alterations to notes or chords. Sometimes there are whole new arrangements under the same roll number.
Some of these variations come about when a master roll is damaged during use, or additional masters are required to run multiple perforators. Replacement sections may be cut and spliced into existing masters, or complete new masters made. There are many ways to arrange a piece of music, just as there are many ways to perform it on an instrument, and it should be no surprise if interpretations vary on return visits.
So while we must always strive for accuracy, we can not afford to be too precious about the authority of our transcriptions. When the content of the roll is clear we can transcribe it exactly, but we can not assume that our one sample is the "correct" or the only version. When the content of the roll is not clear we are forced to make subjective decisions, and there can be many ways to resolve the issues. The best we can hope for is a result that makes physical and musical sense, and preserves (as far as possible) the arranger's style and the composer's intent.
The pages following contain a selection of freely downloadable MIDI files of paper-roll organette music that I have transcribed, scanned and edited using the methods described above.
Comments are welcome via the Enquiry and feedback form.