MOVING PARTS

Recreating The Ballet Mechanique In The Digital Age: Part 1

Published in SOS August 2000
Printer-friendly version

People + Opinion : Artists/Engineers/Producers/Programmers

In 1924 George Antheil wrote a piece of music which required 16 synchronised player pianos playing four different parts, along with a bizarre human ensemble. But it took 75 years before Antheil's dream of a gigantic machine-controlled performance was fully realised, using MIDI, by US-based musician and SOS contributor Paul D Lehrman. In this two-part feature, he explains how this two-year project reached fruition.

When did sequencing begin? Long before computers, long before the crude event generators of early Moogs and VCS3s, there were mechanical instruments that played musical 'programs'. Among these were calliopes, hurdy-gurdys, music boxes, and even mechanical clocks, for which Mozart composed several pieces. Mechanical musical instruments reached their zenith at the end of the 19th century with the invention of the player piano, which soon became a fixture in parlours and saloons all across Europe and North America.

Player pianos could only play solo, so to create larger mechanical ensembles, inventors grafted more and more instruments onto player-piano mechanisms, including violins, bells, organ pipes, and drums. The results were marvellous monstrosities such as the 'Orchestrion'. However, a certain ambitious young composer, and one piano maker, wanted to change all that. Unfortunately, they couldn't. But their failure is one of the great stories of avant-garde music.

The composer was an American living in Paris named George Antheil. He envisioned the music of the future coming from ensembles of mechanical instruments, linked together like looms in a wool mill. In 1924 he wrote a piece which embraced industrial technology with a vengeance. It required 16 synchronised player pianos playing four different parts (four pianos on a part), along with a human ensemble comprised of four bass drums, three xylophones, a gong, two conventional pianos, a siren, seven electric bells, and three aeroplane propellors. He called his creation Ballet Mécanique, and you could say it was the first piece ever written for 'sequenced' instrumentation.

To realise his creation, Antheil was relying on the work of Pleyel, a piano manufacturer in Paris, who had designed and taken out a patent on a system that could slave any number of player pianos to one 'master' instrument, using a complex arrangement of electrical pulses and self-adjusting paper-roll mechanisms (see below). In principle, Pleyel's system bears a striking resemblance to today's timecode-based synchronisation systems. However, the system seems never to have been built, or if it was it wasn't practical enough to be used for public performance. According to one modern player-piano expert, it was just far too complicated to ever work. Unfortunately, it was only after Antheil had finished his magnum opus that he found out that it couldn't be played, so he rewrote it for multiple conventional pianos and a single player piano.

Today, Antheil's original dream of a gigantic machine-controlled performance can be realised. Thanks to MIDI and MIDI-compatible player pianos from manufacturers such as Yamaha, PianoDisc, and QRS, locking together 16 of them is simple. Working with publisher G Schirmer, Yamaha, and a host of consultants, technicians, and musicians, I have helped to fulfil Antheil's dream. It has taken two years, and has been the most interesting music project I've ever done.

Ballet Mécanique is almost unremittingly cacophonous and brutally rhythmic, but in a jagged way that defies dancing, or even foot-tapping (there are over 600 time-signature changes). The xylophones play in parallel major sevenths, while the pianos boom out huge repeated chromatic clusters, the human players sometimes needing to use their forearms. A siren rises and falls over the din, and the bells and aeroplane propellors provide 'pedal points' to keep the noise floor high. Snatches of jazzy melodies occasionally bubble up, only to be subsumed by the sheer wall of sound. It is quite long: 1240 measures which, depending on the tempo, last for 26 to 36 minutes in total. Antheil also introduced a completely new element to musical composition: total silence. There are several sections near the end, some as long as 30 seconds, in which absolutely nothing happens. And this was 30 years before John Cage...

When the Parisians first heard the piece, in 1926 (in the single-player-piano version) it was a roaring success -- if you measure success by the amount of civil disorder a work generates, which was the way things worked at the time in Paris. There were 'boos' and even fights in the audience (and later in the street) between supporters and detractors of Antheil. It was the greatest artistic riot since the premiere of Stravinsky's Rite Of Spring 13 years before, and it made the young composer the toast of the town. "Antheil had Paris by the ear," composer Aaron Copland wrote. He had "outsacked the Sacre." An American promoter tried to do the same thing the following year in New York's Carnegie Hall, but the effort fell flat. New York audiences were apparently too jaded for that sort of thing, and technical difficulties (such as the siren coming in a minute late and continuing to wail long after the piece ended, because someone forgot to tell the siren player that it needed to be cranked for a while before it would sound) induced laughter and derision.

Shortly before Antheil's widow died, in the late 1970s, she gave her husband's manuscripts and files to Charles Amirkhanian, a San Francisco composer, who subsequently published many of the pieces under the name 'Antheil Press'. In the early '90s, Amirkhanian sold the publishing rights to Antheil's work to G Schirmer, the New York music publishing giant.

In 1989 the conductor Maurice Peress revived the single-player-piano version in a concert at Carnegie Hall, but it has not been performed since. Early in 1999, the German Ensemble Moderne performed the multiple-player-piano version for the very first time, but using only two player pianos, not 16. The two pianos were 1920s-vintage pneumatic instruments, customised to respond to MIDI by Dr. Jürgen Hocker, a musicologist and scientist.

Delaying Gratification     Aside from the firmware problem mentioned elsewhere on this page, Disklaviers have another interesting design feature I had to deal with -- they have a built-in delay of 500mS when responding to incoming MIDI data. Why? Well, consider an electronic synth that receives two notes simultaneously at different velocity levels. Although the amplitudes of the two notes will be different, they will sound at exactly the same time, because the synth's circuitry responds immediately to the command, regardless of the velocity. In an acoustic piano, however, 'velocity' refers literally to the speed of the keystroke and the resulting hammer stroke. If you send an acoustic piano two simultaneous note-on commands with different velocities, the solenoid-controlled hammers will start to move at the same time, but the hammer with the lower velocity will actually strike the string later than the hammer with the higher velocity. Yamaha compensates for this by building in a 500mS pre-delay, and then reducing that delay according to the individual note velocities, so that high velocity notes end up delayed more than low-velocity notes. This delay can be switched off, but the MIDI response in non-delay mode becomes extremely 'squishy' and unpredictable. Apparently there are other kinds of parsing of the MIDI data going on during that interval, which is necessary for the instrument to respond correctly. So while the delay feature solved Yamaha's problem, it caused some very interesting complications later on for me.

Bill Holab, who until recently was Schirmer's publications director, saw a great opportunity in the original, 1924 version of the piece, and embarked on a long-term project to create a published version that any contemporary performing group (with sufficient resources) could play. Holab recognised that the piece of the puzzle which Antheil didn't have was MIDI. Knowing about Yamaha's Disklavier line of MIDI-compatible player pianos, Holab contacted that company, and asked if they would be willing to support a first performance of the piece by supplying an appropriate number of Disklaviers to whoever put it on. Michael Bates, director of academic relations for Yamaha's piano division, said yes. And then Holab contacted me.

Holab was familiar with a book that I co-wrote with Tim Tully called MIDI For The Professional. He contacted me by email, and when I received the message asking me if I would be interested in helping with the project, my jaw dropped.

What Holab didn't realise was that I already knew about the Ballet Mécanique. Thirty years ago, when I was at a music camp in Vermont, a percussion teacher told me about the piece -- the 1952 version -- and even gave me a tape of the only recording then available, which I listened to dozens of times. For years, I was dying to hear it performed, or better yet, to somehow be involved in a performance. But I had never even heard of the earlier versions of the piece. Needless to say, I was utterly thrilled to be given the opportunity to work on this incredible composition, and to help Antheil's impossible dream come true -- only three-quarters of a century late.

Holab wanted to hire me to create a set of MIDI sequences, on disk, that could accompany the printed score and the parts for the live instruments when they were sent to groups who were planning to perform the piece. He also knew there would be many other technical issues that these groups would have to deal with, and so he wanted me to research all of these issues, and address them in a booklet that would be included with the other materials. I readily agreed, but then offered to do even more than that. I reckoned that many performing groups might have a hard time procuring sirens, bells, and airplane propellors and so I proposed to find samples of those sounds, loop and format them, and include them in the materials sent out, all of which could be on CD-ROM. Holab took me up on the offer.

We started by deciding whether I really had to enter the entire mammoth score into a sequencer by hand, or whether there was another way. The piece had already been put into a computer, for publishing purposes, by freelance editor George McGuire, using the venerable PC notation program called Score. This program doesn't output MIDI files, but there are third-party add-ons that can, and so Holab and I tried to create a MIDI sequence from McGuire's file. But the insane time signatures (there are bars of 11/16, 7/32, and so on), the huge number of time-signature changes, and the sheer volume of notes (some measures contain clusters of 23 notes in each of the player piano parts), made the MIDI files into a useless jumble, and every sequencer I tried couldn't even open them.

Another possible avenue was opened when I contacted Trimpin, an inventor and composer whose fantastic multimedia installations have been seen all over the world. Trimpin is an expert on historical mechanical instruments and actually owns a set of rolls (it takes three to play the whole piece) for the 1926 single-player-piano version of the Ballet Mécanique. He had invented a scanning device that could convert the paper rolls to MIDI data, and in fact had created Performer files from his rolls. There were, however, three problems which prevented me from taking advantage of his work. Firstly, there was no clean way to separate out Antheil's original four parts from this 'combined' file. Secondly, the bar lines in Trimpin's files didn't correspond to the ones in Antheil's score, so they would have to be put in by hand -- an extremely complicated operation. Finally, Antheil himself wrote that he was never particularly happy with any of the various rolls that had been created for the work, and it looked as if I would have had to check every note against the score for accuracy.

So I resigned myself to creating the files from scratch -- all 1240 measures of them. I started loading the notes into Opcode's Vision, which at the time was at version 3.6. I soon realised that it wasn't going to work as the program had a lot of trouble handling sequences more than 999 bars long. I called Opcode technical support and asked if the next version was going to support longer sequences, but the answer was "We don't know". Fortunately, a couple of weeks later, they sent me v4.01, and I immediately set about making a sequence of 1200 bars. Vision doesn't like bars with more than 32 beats, however, so I had to convert some of Antheil's more expansive time signatures, like 64/8, into their equivalents.

Since nuance of performance was not exactly a high priority, I took great advantage of something which I usually avoid when I'm using a sequencer -- step-time. Squinting at the tiny notes on Antheil's score, I could build the impossibly complex chords, one note at a time, without worrying about being able to actually play the things. Using the copy and paste functions in Vision was something of a mixed blessing. The score, fortunately, had plenty of duplication. Antheil often wrote two bars that were almost identical. One bar might be 4/4 and the next 7/8, and the pattern would repeat, but the half-beat's worth of material that was chopped out of the second bar wasn't the same as the material taken from the fourth. So after pasting, I would have to go in and tweak the rogue measures by hand.

So that I could hear what I was doing while I was creating the sequences, I got Kurzweil to agree to lend me four MicroPiano modules, and installed them in my home studio, rewiring the room with a second pair of speakers behind me. Thus the four player-piano parts would surround me as I sequenced them. This turned out to be a wonderful way of proofing the parts because, despite the incredible atonal cacophony of the piece, if there was anything wrong with the texture at any point (like a wrong note) it stuck out spatially even more clearly than it did musically.

Paul D. Lehrman is a composer and author, who has been writing for Sound On Sound since 1986. More information and sound clips relating to the Ballet Mécanique project can be found at www.antheil.org, and a CD of the piece is available from the Electronic Music Foundation at http://www.cdemusic.org/emfmedia.

Then it was time to test the sequences on 'real' instruments. Modern player-piano mechanisms are very different from the pneumatic systems of Antheil's day, and I wasn't sure if today's pianos could actually handle the Ballet's demands, even though I reduced the note velocities and shortened their durations (see 'The Question Of Duration & Dynamics' box on page 99). In addition, modern pianos are designed, primarily, to be used with their own internal recording/playback computers. They all have MIDI inputs, but they behave somewhat unconventionally when driven by an external MIDI source.

I wanted to make sure that the files would work with all of the major brands of MIDI player pianos, of which there are three: QRS, PianoDisc (both of which are sold under a variety of brand names) and Yamaha. My first test was with an upright Disklavier at the renowned MIT Media Lab. I installed Vision on one of their Macs, and ran my sequence. At the first chords, I knew I was in trouble. Notes were dropped, rhythms were jerky, and the pedal-on commands were left hanging. I called George Litterst, a software developer, teacher, and consultant for Yamaha in my area, and he asked me to read him the firmware revision of the instrument I was playing on. It turned out that it was one of the very first Disklaviers Yamaha made, and the MIDI implementation in the ancient onboard CPU was, to put it politely, a little flakey.

George invited me to his house, where he had a Disklavier conservatory grand in his basement studio, with much more recent firmware. I installed Vision on George's Mac, and the sequences ran much better. But there were still problems. The first had to do with polyphony. The Disklavier has a polyphony limit of 16 notes (more than that, and apparently the drain on the power supply by the solenoids would be too much), but Antheil's score calls for chords of up to 23 notes. I needed to see what happened when the polyphony limit was exceeded. Was there a voice-stealing algorithm, as in digital synths, in which the oldest, or newest, or highest, or lowest, or the note with the least velocity is ignored? I found out that the Disklavier is most definitely not a digital synthesizer: its stealing algorithm was the totally bizarre 'Last Note Only'.

Here's what happened: I would send it a 16-note chord, and would hold it while I followed it quickly with a single note. But when the 17th note was received, instead of one of the keys in the chord going up, all of them went up, and I was left with only one note -- the 17th -- sounding! So it was quite clear to me that the 16-note polyphony limit was to be considered non-negotiable. (Later, George got a firmware upgrade which corrected this anomaly, and the stealing algorithm became true last-note priority, but the polyphony limit didn't change.)

Even with the latest firmware, George's Disklavier couldn't quite handle Antheil's score. Large-chord tremolos and extended fast trills would cause it to miss notes. I experimented with note durations and velocities, and found that the trills behaved better if I increased the velocity to 99. Above that, it sounded as if the hammers might fly off the keys, so I didn't want to push it. In some cases, setting the durations to 60 percent, or shorter, helped as it seemed to give the keys more time to recover. Randomising durations, so that all the solenoids didn't release at once, also helped. Occasionally, however, I found that setting the durations to 120 percent actually did the job better. Having the note-offs occur after the next note's note-on invoked some kind of fast-recovery feature in the Disklavier, which allowed it to play the passages without falling over.

For the really large chords and tremolos, however, no amount of tweaking would get them to sound right, so I simply thinned them out. In almost every case, the very large chords were doubled across at least two and usually all four of the player-piano parts, so I could steal different notes out of the different parts, and the effect would not be audible. To test my files on other MIDI player pianos, I found a dealer who actually had both PianoDisc and QRS systems on their huge showroom floor.

QRS's system behaved very similarly to Yamaha's, with 16-note polyphony (oldest-note priority), the same 500mS fixed delay, and the same problem with repeated chords. But the Pianodisc systems were a different story: they had a polyphony limit of 32 notes, and so could play most of Antheil's chords without complaining. With fast repeating chords, however, they had the same problems as Yamaha's. Their delay time for dealing with incoming MIDI events was much shorter than Yamaha's: anywhere between 50 and 150mS, depending on how the software was configured. I kept careful notes, since all of this information would have to go into my documentation for Schirmer, so that performing groups could understand what they were going to have to deal with.

Alongside checking out player pianos, I was working on getting the samples I had promised to Schirmer. I auditioned dozens of aeroplane sounds from various effects libraries on disc and on-line. A friend told me that Alan Parsons had used some fantastic aircraft sounds on his latest project, and soon after that, at an AES convention in San Francisco, I found myself seated at a breakfast table right next to Parsons. I introduced myself (I'm sometimes cheeky that way), told him briefly about my project, and asked about his aeroplane sounds. He seemed genuinely interested in the project, and gave me the contact information for the owner of the sounds.

The Question Of Duration & Dynamics     Two interesting issues were raised whilst sequencing Antheil's score. The first was duration. Exactly how long is an eighth note? A player piano roll is made of paper, and if you cut a hole in the paper that is exactly an eighth note long, and then repeat the note, the second note won't really exist. In order for the hammer to recover and strike the second, there has to be some paper between the holes. MIDI needs much less time between a note-off and the next note-on -- 1 sequencer 'tick'. Even a MIDI-driven player piano, which uses solenoids to move the hammers and keys, needs time to recover. Since the Ballet Mécanique is full of repeated chords (at one point the same 16th-note chord repeats 300 times), the decision as to how long the notes should sound was an important one. I chose an arbitrary duration of 80 percent for all the notes as I input them in step time, which worked fine for most of the time. The other issue was dynamics. There are no dynamic markings in the piece, except for a few stray accents: in Antheil's writings, he says it is to be "played as loud as possible". Player pianos in Antheil's day were capable of only crude dynamic changes -- there were two pedals, one each for treble and bass, that could move the hammers closer to the strings, thereby lowering the volume of the appropriate half of they keyboard. In the introduction of Antheil's score, he talks about these pedals, but then he never notates where they might be engaged. If I were to take Antheil literally and make all the note velocities 127, I would break the pianos: they aren't designed to take that kind of sustained punishment. So I set all the note velocities to 80, to give me plenty of room to move up or down.

The samples had been made by Brad Miller, who had tragically died not long before. I contacted Miller's widow, who graciously sent me several CDs of incredible sounds, which were part of a library called Sonic Booms. Unfortunately, like all the other samples I auditioned, they weren't going to work. Antheil's score, which stipulates "high wood, low wood, and metal" propellors, requires them to sound for varying lengths of time, from an eighth-note to dozens of measures. I had to be able to loop the samples, which meant that they couldn't change pitch. On all of the library samples, the propellors were flying by, starting up, or doing something active, which made them impossible to loop.

I mentioned my dilemma to Tim Tully (co-writer of the MIDI book that got me in all this trouble to begin with), who had a solution. A friend of his had a plane at a private airfield, and he could go out there with a portable DAT machine and tape a group of aeroplanes sitting on the ground. His recordings were terrific, and I was able to get three very different-sounding prop sounds, which I brought into BIAS Peak and looped, and then loaded into my Kurzweil sampler. I mapped them over three sections of the keyboard, and set the pitch-bend sensitivity at 12 semitones, so that if someone wanted to have some Doppler effect, they could add it themselves in real time.

The siren samples I found in libraries had a similar problem. They were all moving, or in some kind of context, like 'siren in traffic', 'siren in rain', and so on. So I called a friend who was a fireman in the next town to mine, and he agreed to blow the siren on one of the engines while I recorded it on DAT. I put a pair of cheap dynamic mics on a stand 10 feet from the siren, and although I set the DAT's input pad to 'line', and the level control all the way down, I still pinned the meters. I had him blow the siren four times, and when I got home, I found that only one of the four takes hadn't clipped. Fortunately, there was plenty of good audio in that take, and so I was able to create a looped sample, with a dandy release segment for the siren falling. Played over a good sound system, especially in stereo, it was guaranteed to send people running for the exits. The bell samples were much easier: I found an old alarm bell, bought a couple more bells at local hardware stores, and recorded them right to disk in my studio.

As I was finishing the samples, I realised that now I had almost all the materials necessary to produce a complete MIDI-controlled performance of the Ballet Mécanique, right in my studio. All I had to do was to program in the human piano parts and the percussion parts which, considering what I had gone through to get the player piano parts in, was easy. It took about two days, and soon, using a combination of Kurzweil, Roland, and Emu synths, along with my four MicroPianos, I created the first performance ever heard of the original Ballet Mécanique, for which I was the sole audience. It was a pretty amazing moment. But it was nothing compared to what was to come.

Next month: The Ballet Mécanique on stage, on the web, and on CD.