Triple Harpsichords, Cross-Staff Beaming, revisited

This is a substantial re-write and clarification of a post made January 4, 2010.

Daniel Spreadbury, Sibelius Senior Product Manager and writer of, kindly posted about my story on using PhotoScore and Sibelius to produce a new edition of Bach’s Triple Harpsichord Concerto, BWV 1064.

In my post, I mentioned problems I had with cross-staff notes in keyboard parts. Daniel explained that I was going about it the wrong way.

I contend that I was going about it in a way that makes sense to a keyboard player or a composer. I have read the reference manual. In the process, I revealed what I think is a bug or deficiency in Sibelius. Setting aside the the age-old “it’s not a bug, it’s an undocumented feature” argument, let me explain how I did it and how Daniel recommends doing it. At the end, I’ll explain why I wish that Sibelius would handle this situation differently.

Let me use a new, more simple example.

Above are two bars from the grainy PDF of the urtext.

The two hands play an unbroken melodic line an octave apart, in parallel. There are no rests in the line played in either hand. Because the melodic figure goes through some low notes, it’s properly notated by having the right-hand part move down to the bass clef and then back up again, following the arc of the melody. There are cross-staff notes in this example, but there is no need for cross-staff beaming, due to the particular note groupings.

Example One

How I did it

I chose to notate this in Sibelius by entering all the right-hand notes in Voice 1 in one unbroken line, and then selecting certain notes and applying the “Cross Staff Notes” commands to move those from the treble staff to the bass staff.

I entered all the notes in Staff 1 in Voice 1 (blue noteheads). Then I entered all notes in Staff 2 in Voice 1 (blue noteheads), and flipped the beams and stems in Staff 2 down, below:

Next I used the “Move Down a Staff” command on the notes that needed to cross staves. This includes two notes that are not only cross-staff, but also cross-beamed. [Postscript: Yes, I realize I’m moving one note that was not moved in the Urtext. You go to all the trouble to make nice canned examples and then at the end you realize you overlooked one little detail.]

The only problem, below, is funky beams on the cross-staff notes in the second bar:

This is not an error: sometimes it’s the shape you want. In my case, I wanted to adjust them like this, below, which can easily be done by selecting each beam and dragging:

And that’s all there is to it. But the problem, as I will detail below, is that in more complex and dense examples, Sibelius does nothing automatically about collisions in many elements: noteheads, accidentals, stems, beams, and slurs. Here is my example from the previous post:

2. What happens when you tell Sibelius to cross staves

The recommended method

This involves setting things up very differently, like this:

In Staff 1, create notes in Voice 1. Enter 8 16th notes, then a half-note rest, then in Bar 2, a half-note rest, followed by 4 16th notes and 2 8th notes.

In Staff 2, create notes in two Voices. In Voice 1, enter a half-note rest, followed by 8 16th notes, a dotted-eighth, 4 16th notes, and a half-note rest.

Move to Voice 2 (green noteheads) in Staff 2, Bar 1. Enter the line shown below.

Next, select only the two notes highlighted in red, and use the “Move Down a Staff” command on those two notes alone.

See below:

Once the notes are moved, just as in the previous example, drag those funky beams into place:

One remaining problem: This is a convenient fiction. What we’ve notated in the recommended method is three voices and a bunch of rests, whereas in what J. S. Bach actually wrote, there should be only two voices and no rests anywhere. To complete the illusion, we have to find the half-note rests and use the Hide command. Below you can see them greyed-out.

Unfortunately, outlined in red above, Sibelius has spontaneously flipped the stem directions on two beam groups as soon as you use the “Hide” command to conceal the half-note rests that were visible before. So in the last step, below, I have manually flipped the beams back where I want them:

When you print out the score, all the note heads will be black, and there will be no indication to the performer that you have performed a little slight-of-hand by using three voices where there should be two.

Evaluating the recommended method, you can imagine that under certain circumstances it would take more work than my method. Mapping out extra voices, switching note input to different staves, lots of extra rests that have to be hidden after the fact, and problems with beams and stems flipping where you don’t want them. So why did Daniel recommend this method? One more example, and then my conclusion is below.

Collisions: Example 2

Here is another example from the Bach Triple Harpsichord Concerto. In these two bars, there are notes that cross staves inside of every beam group.

How I did it

It makes musical sense to me to enter Staff 1, Voice 1, four groups of four eighth notes, and then go to Staff 2 and create Voice 1, four groups of four eighth notes.

Then, in Staff 2, select the last three notes in each group of four:

Apply the “Move Up a Staff” command. You get this mess:

Flip the beams in Staff 1 Voice 1 where you want them, drag on each of the Staff 2 beams to get the correct orientation and adjust their angles, move the staves closer together, and maybe do some more dragging to optimize things:

The recommended method

Create two voices in Staff 1 and no voices at all in Staff 2.

Hide the whole-note rests in Staff 2.

In Staff 1, Voice 2, select only the first notes in each four-note grouping.

Apply the “Move Down a Staff” command, and the result is this:

Now you can drag the beams into position and move the staves closer together.


I now know the recommended method to use Sibelius in order to get the desired results with cross-staff notes. But I don’t like it. That’s why I’ve gone to all the trouble to document this. Software gets improved when users express their opinions about features.

Sibelius advertises collision avoidance as a major feature that make life easier for you when you create scores. It does. But those features are not implemented at all on cross-staff notes. If you actually use the cross-staff note commands where it makes musical sense, you may create further problems that you have to fix manually. My interpretation of the recommended method boils down to this: avoid using the cross-staff commands whenever possible. Only use them on individual notes when there is no alternative.

In the absence of Sibelius addressing this issue, you need to use the recommended work-around of entering notes in the score in a different order than a keyboardist would play them, creating extra rests and extra voices, and then hiding them after the fact.

The Atlanta Baroque Orchestra Brings Ancient Music Back to Life with Sibelius and PhotoScore

L-R: Daniel Pyle, Michael Bauer, Peter DeWitt, rehearsing the score I edited. Photos by Rich Nuckolls.

Lots of musicians us the Sibelius notation program to compose and publish new music. But what you may not know is that some people use it to re-create music that’s hundreds of years old. This is a technical essay written for musicians and scholars on how I worked with the Atlanta Baroque Orchestra to prepare a piece of music by J. S. Bach that is rarely performed anywhere. This may also serve as a helpful tutorial for users of any music notation software.

The challenge

Ever heard an orchestra with a real harpsichord? How about a concerto with three harpsichord soloists? Johann Sebastian Bach wrote one, and the Atlanta Baroque Orchestra (hereinafter the ABO) performed it on November 21, 2010.

Here’s how I described it in the concert press release:

…[It is] a rarely-heard, dazzling showpiece for three harpsichords and strings that J. S. Bach performed with his sons W. F. and C. P. E. Bach. “I call it the Grand Prix of keyboards,” says Wheat Williams, music copyist and volunteer. “It’s a dizzying display of spirited shredding by three virtuosos that would leave today’s rock guitarists in awe. You won’t believe how many notes they can pack into 16 exuberant minutes.”

I had just signed on as a volunteer working with the orchestra in October 2010, intending to provide them with a new Web site, marketing through email and Facebook, and services like receiving online donations and online advance ticket sales. Then Resident Director Daniel Pyle told me about Bach’s Triple Harpsichord Concerto in C, BWV 1064. “A lot of the music we do is long out of print, and we have to go looking for scores,” he said. “Even when a score is commercially available, it is usually an old printing from worn-out old engraving plates, and the legibility is quite low. So sometimes I create a completely new edition of a score in Finale.”

Bach’s Triple Harpsichord Concerto, BWV 1064, frontpiece

I stepped up. Based in Atlanta, Georgia, I’m a musician and consultant working with Roberts Creative Systems out of Franklin, Tennessee, and we are a Sibelius dealer among other computer music and audio services. I offered to create a completely new score for this piece for three harpsichords and string orchestra. It’s 16 minutes long, in three movements, and according to Sibelius’ note-counting plug-in, when I was done, there are 23,828 notes. It was a challenge.

Sibelius 6

Sibelius-certified expert Dave McKay, with Roberts Creative Systems, estimated that my Bach score would cost the ABO $2,500 if they hired an AFM (American Federation of Musicians) union copyist. No chamber group could afford that much for a piece that makes up only 20% of the music performed in a single concert, which is why I’ve discovered that a lot of early music performers are already doing their own work in Sibelius or other notation programs. When Daniel Pyle saw my work, he upgraded his old copy of Sibelius to 6.2 and joined in with the proofreading on his 27-inch iMac.

It goes without saying that virtually all the repertoire of a group like the ABO, which performs the music of not only Bach but all sorts of European composers from roughly the years 1625 to 1750, is in the public domain.

Bach never published any of his scores in his lifetime. It’s safe to say he never expected anybody other than himself and those he hired to perform most of them. Fortunately a lot of hand-copied scores survived. What we have today in published form was mostly engraved and printed in the late 1800s, over a century after Bach died. And his output was so vast that there are plenty of pieces that don’t get performed often, so making modern editions of more obscure pieces is not of much interest to sheet music publishing companies.

So what is a baroque orchestra?

The ABO is an “early music” chamber group, part of the “historically-informed performance” or “authentic performance practice” revival movement, which started in Europe in the late 1970s and has caught on worldwide. The ABO has been performing since 1998, and calls itself the oldest such orchestra in the Southeast of the US. They perform the music of the late Renaissance, Baroque and Classical periods on replicas of the actual instruments used by the composers.

Members of the Atlanta Baroque Orchestra rehearse

Modern symphony orchestra instruments are often easier to play and always louder and harder in their sound, and when a symphony orchestra plays Bach, Handel or Haydn, it’s usually with many more players than the composer would have used. In contrast, in the authentic Baroque ensemble, “sections” are often only one player to a part. Violins and other string instruments use the original sheep gut, not steel strings, and old-style bows and bowing techniques that enable more subtle articulations. Vibrato is used minimally. The fretted, six-string viola da gamba often joins the cello. The lute, theorbo, harpsichord or tracker pipe organ play continuo. Baroque “natural” horns don’t have valves, flutes are made of wood and don’t have keys, recorders are used a lot, and don’t get me started on Baroque bassoons and oboes. As far as performance, much to the astonishment of symphony orchestra players, Baroque-specializing musicians are expected to improvise and embellish their parts according to certain historically-researched guidelines. Everybody tunes down to A-415, not 440, and keyboards don’t use equal temperament, because it hadn’t been invented yet. Overall, the sound that an authentic Baroque orchestra makes is startlingly different than the modern symphony orchestra. You can find many famous examples on iTunes, although the ABO has to date never released any recordings.

Getting started: where to find the music

With regard to finding free copies of sheet music in the public domain online, there are a few excellent resources. Music enthusiasts all over the world have taken old scores, scanned them, made PDF files, and contributed them to Wiki-like sites including:

Here you can find thousands of scores from all sorts of composers from antiquity to the 1920s, presented as PDFs. In addition to the original scores, you may also occasionally find modern arrangements of classics by the great composers whose contemporary editors have released them into the public via the Creative Commons License or some such. I needed to avoid modern arrangements, because I was looking for the authentic original work of the composer.

What I hoped to find was an urtext; that’s a German term that means a professionally-engraved edition that is recognized as authoritative, as close to the composer’s original manuscript as possible. At the IMSLP, I found a grainy PDF of the urtext of the Bach Triple Harpsichord Concerto, published by Breitkopf & Härtel in the late 1800s.

Page One of the original, scanned

Happily, there are lots of folks around the world who are already doing what I’m starting out to do–they’ve already found old scans, they’ve redone the piece in Sibelius or another commercial notation program, and they’ve donated those scores back to the Internet archives mentioned above–sometimes in the source files of the various notation programs, sometimes as PDFs you can print, or if you’re really lucky, the MusicXML format, which can be parsed and converted into the formats of many different notation programs. But there were no such files for the piece I needed.

If you can’t find the score you need in the above sources, there are commercial sites like Sibelius Music where, among the new music, enterprising copyists sell scores they’ve created from old scores. That’s a legitimate thing to do, as you can copyright your own edited edition of a composition in the public domain even though the underlying musical composition belongs to no one.

Finally, there’s scouring or looking for used copies of printed scores from bookstores, purchasing them, and getting them shipped to you. I might add that commercial publishing companies sell new, contemporary printed scores of music that’s in the public domain due to its date of composition, and even if there’s a copyright notice on that printed edition, you are generally free to make your own version of the notes on the page and do with them as you will, without restriction.

If you can find actual printed scores, they’re often enormous. Daniel Pyle provided the original Breitkopf & Härtel scores for three other Bach two-harpsichord concertos that I’m working on. He found them in a used bookstore in London over twenty years ago. These original conductor scores measured about 10-3/4 x 14-1/4 inches (27 x 36 cm). For those, you need a large-format scanner. Daniel and I both own the excellent and affordable Brother MFC-6490CW all-in-one printer-scanner, which can scan and print 11 x 17-inch tabloid or ledger size, and provides a reliable sheet-feeder for scanning dozens of pages at a time. I got mine brand-new last year for only $200.

I do all my work on an Apple MacBook Pro running Mac OS X 10.6 Snow Leopard. I use Hamrick VueScan software to process the scans and make PDFs, and I repaginate and clean up the scanned PDFs in Smile Software’s PDFPen. (When you batch-scan printed scores with a sheet feeder, you often find yourself scanning the pages out of order because of the way the book is bound, so a program that lets you reorder the pages within an output PDF file is very useful.)

Getting down to business, and the tools of the trade

Whether you have to make your own scans, or you can find scans ready-made online, the remainder of the process, getting the music into Sibelius, is the same.

However beautiful the original engravings were, the printed quality of what’s available today is usually poor. Stems may not connect to noteheads in places, and clusters of notes in chords may blob together, where accidentals and noteheads are no longer distinct. The people who made the scans may not have been careful: their images are crooked, distorted, or the music on the edges is out-of-focus if they scanned from a bound book and didn’t lay the whole page perfectly flat on the scanner. All these factors can impact what kind of success you have with the next phase.

The process used to get an old score into Sibelius or another notation program is called optical character recognition, or OCR. In this process, a computer program reads a digital bitmap image of a printed page (made with a scanner) and attempts to convert it into notes and words that a music notation program can interpret. Neuratron PhotoScore is the tool I use. It’s designed to work hand-in-glove with Sibelius. You have PhotoScore read a multiple-page PDF file, and then you must edit the results to correct some of the more obvious errors.  Next, you instruct PhotoScore to pass the file off to Sibelius for further work.

Neuratron PhotoScore Ultimate 6.1
Neuratron PhotoScore Ultimate 6.1

The ready-made PDF score that I had to use as the basis of my Bach project was around 72 dpi. The PhotoScore manual will tell you that it won’t handle scores of such low resolution, but it worked pretty well! A 300 dpi scan would be preferable.

PhotoScore screen shot (in Windows)

Re-creating the masterpiece

I have a love-hate relationship with PhotoScore. It did an amazingly accurate job of reading the grainy PDF of the Triple Harpsichord Concerto. Unfortunately, despite the fact that what it does is near-miraculous, after it reads in the score, it’s necessary for the user to do a great deal of editing of mis-read music within PhotoScore before exporting its output to Sibelius–even when working from pristine printed scores. PhotoScore, I’m afraid to say, provides a painfully difficult editing environment. I could write a lengthy blog about all the improvements that I think PhotoScore could make to the user interface. Many functions seem to take many more clicks, key presses, and menu and dialog box selections than are necessary. To give one example, the OCR function frequently makes errors in identifying clefs on any staves that are connected with braces or brackets. The process of manually correcting these errors is onerous. [If you really know what you are doing, before you run the PDF through PhotoScore, you can open the PDF in a program like PDFPen and “touch up” the notation by masking out certain parts like brackets and braces, or filling in broken lines with a pencil tool. But that feels like something you shouldn’t have to do in the first place.]

Howbeit, PhotoScore is still a life-saver. I would be overjoyed if Avid Sibelius would buy Neuratron and its genius programmers outright and integrate PhotoScore directly into Sibelius, with a good user-interface overhaul. Can I get an “Amen”?

Once the PhotoScore output is passed on to Sibelius, there’s much more editing work to be done. In this context, the normal Sibelius workflow gets turned on its head, and you have to re-think things. For one thing, you are not seated at a blank canvas entering a score a bar at a time from start to finish, watching the layout unfold as Sibelius makes automatic decisions about spacing and pagination. PhotoScore hands off to Sibelius, and Sibelius opens up a 24-page score full of music, much of which is necessarily full of errors that you need to correct. The layout of how many bars to each system, and how many systems per page, has been dictated with hard breaks by PhotoScore in emulation of the original printed score, and this may not be optimal. Old scores were optimized to print on the fewest pages possible, and they are often way too crowded for good legibility. You have to find the hard breaks, delete them, and make your own decisions about note spacing, staff spacing, system breaks, page breaks, and other formatting decisions on the fly as you go through the score.

Editing is all after-the-fact and it may not always make sense to start with Bar One and go all the way to the end in order. As you go through and correct errors, you find yourself selecting small passages and using the “Optimize Staff Spacing” and “Reset Note Spacing” commands over and over again, and using Undo a lot as well.

Any text notations present problems, and PhotoScore is not very good at OCR for text in any language, though it purports to support several. If you are working with Baroque scores, you need to think back to music school and remember all your terms in English, German, French, Italian, and whatever else the printed pages may throw at you. Google Language Tools can help.

A useful tip: Set up two monitors, one for Sibelius, and one for Adobe Reader or another program that displays the PDF scan of the score you are re-creating. Using a PDF editing program like PDFPen, number each measure of the original scanned score in the PDF. Old scores from the 1800s don’t use measure numbers at all. Zoom in for detail and work line-by-line. This works much better than having Sibelius on the screen and a printed score sitting on your desk or on a copy stand. Glancing back and forth between the screen and the paper on your desk for hours is fatiguing.

Hairy passages can’t be read correctly by PhotoScore and have to be re-entered by hand (best done in Sibelius, not PhotoScore), which I do using the QWERTY keyboard exclusively. This particular score presented challenges that I do not blame PhotoScore or Sibelius for having difficulty handling. String parts were quite simple, and were a breeze; PhotoScore nailed everything with little manual editing required. However, as you can imagine, Bach wrote the three harpsichord parts for virtuoso showing-off, and the parts are murderously tricky. Cross-staff beaming is everywhere, and the density of runs and trills are dizzying. It’s a breathless roller-coaster ride from start to finish.

I’ve gotten in arguments with an unbelieving expert, but I know of no way around this problem: Sibelius 6.2 does not provide any collision avoidance for two voices that are input on separate staves but then put close together on one staff using the Cross-Staff Notes commands. I had to spend a lot of time intricately editing around illegible blobs of overlapping noteheads, accidentals, slurs and ties, stem directions, and beams.

1. Music on two staves, needing to be cross-staff beamed to one staff
2. What happens when you tell Sibelius to cross staves. Note the overlapping sharps and noteheads.
3. How you want it to look, after you manage to get all the collisions removed. I had to manually shift colliding accidentals and noteheads horizontally using the Properties palette.

Because I had to add and delete groups of notes interactively from passages that were already in the score, it could be infuriating: I flip the stems of a voice in a measure one way, move something where I think it needs to go, and Sibelius spontaneously decides to flip the stems back the other way at a whim. I have to tell Sibelius to flip them back again. This may happen several times as I struggle towards making the passage look the way it does in the original score. I dearly wish that if I overrode Sibelius’ stem and beam flipping on a certain selected passage once, that Sibelius would keep it that way, and not try to change it back, no matter what other notes are added or deleted from that bar.

One thing to consider is whether or not engraving conventions from the 1600s to the 1800s should be updated to the modern conventions that Sibelius’ algorithms employ. This has to do especially with things like rules for groups of beaming and the placement of slurs. One wants to improve legibility; that’s always the goal. However, specialists in Baroque performance practice are already accustomed to reading old notation, so you don’t want to defy their expectations too much. I opted to stick with the original engraving and manually override the sub-group beaming that Sibelius automatically created on numerous dense melodic runs of mixtures of 16th, 32nd and 64th notes. I filled in extra beams across all the sub-groups, because Daniel Pyle felt that the modern sub-group beaming might subtly change the rhythmic emphasis that the players would tend to perform, making the performance less authentic.

Modern sub-group beaming the way Sibelius does it by default
I changed the beaming to the older method used in the original

Even though Bach’s music moves to different well-defined key centers for extended sections within a movement, the original notation does not use key changes. Rather, each movement uses only one key signature, and modulated sections rely on lots of accidentals added to the notes in each bar. Initially I thought to improve things by notating key changes in a few obvious places, but Daniel Pyle vetoed the idea. It’s hard to justify messing with Bach’s notation, even though in his time the music theory regarding key changes wasn’t formally worked out, and we have the benefit of hindsight.

Another mechanical issue with Sibelius 6.2 is figured bass notation, the Baroque version of jazz chord symbols for the keyboard, cello and bass (collectively known as the basso-continuo, the all-important Baroque rhythm section). In this concerto, all three solo harpsichords have passages with figured bass. Sibelius 6.2 can produce good figured bass, but the user interface is buggy: you can’t actually see a figured bass chord symbol while you are entering it, which can take numerous key commands. You only see the chord symbol after you finish entering it and move on to the next chord, and going back and fixing mistakes is awkward.

Figured bass

There is no audio playback for figured-bass symbols: there used to be a plug-in for that, but it is no longer available, presumably because it never worked well. That is just as well, because Baroque continuo players are expected to improvise their part in the period-correct style, following the figured-bass chord symbols, whenever the part is not written-out long-hand.

As far as audible playback for proofreading purposes, Bach’s fast, lithe runs in the violins just won’t play back with the slow attack of the string instrument samples provided by Sibelius–and there are no sample libraries of authentic Baroque string sections on the market anyway. So I found it useful to assign all the string parts to a Rhodes electric piano sound, which won’t miss a lick–and there are a lot of licks. Fortunately, the stock Sibelius harpsichord sampled instrument is good enough, although an impossibly dense score with three harpsichords playing six hands at once presents quite a challenge to the ear. Hard stereo panning, left, center, and right, is helpful. Sibelius can only provide equal temperament at A-440, but this is not an issue for proofreading.

Printing It Out

For the individual parts, I formatted everything for US Letter (8-1/2 x 11 inches), and optimized system layout and page turns for each part. I output the parts as PDFs so they could be emailed to each performer for home rehearsal. Because early music specialists are so rare, the ABO works by flying some players in from all over the US only three days before each concert, when they have marathon rehearsals leading up to the performance. Therefore it’s important to deliver parts a month in advance that everybody can print out on common inkjet printers. Custom large paper sizes aren’t practical, and US Letter and A4 are becoming the only sizes that matter. By the way, the conductor’s score isn’t used in the concert–there is no conductor in authentic Baroque ensembles! Usually everybody follows the keyboard player’s head-nods–but there was no continuo player per se in this unusual concerto, since the three harpsichordists are all soloists! Frankly, I don’t know how they pulled it off.

Listen to the live concert recording of the final movement.

I would like to give this piece I’ve typeset back to the world music community, so I’m going to make it available free, with this notice:

© 2010 Wheat Williams Editions for the Atlanta Baroque Orchestra. Permission is granted to copy and distributed freely. Please consider making a donation to the Atlanta Baroque Orchestra at if you find this edition useful.

Here are the files for you to examine.

Movement One Conductor’s Score PDF
Movement One in Sibelius 6.2
Movement One separate parts, PDF, .ZIP

Movement Two Conductor’s Score PDF
Movement Two in Sibelius 6.2
Movement Two separate parts, PDF, .ZIP

Movement Three Conductor’s Score PDF
Movement Three in Sibelius 6.2
Movement Three separate parts, PDF, .ZIP

The original scanned score of the Breitkopf & Härtel edition from the late 1800s that I worked from can be found here, at the IMSLP.

At some point I’m going to contribute my score back to the WIMA or the IMSLP.

I hope you’ve enjoyed this tutorial. You can thank me by visiting the Atlanta Baroque Orchestra web site, and making a donation.