The main window''s top pane shows a scanned portion of my handwritten score. The area below is PhotoScore Ultimate''s attempt to interpret it. The two bars contain six or seven mistakes.
When I heard that PhotoScore Ultimate ($249) could read handwritten manuscript, I had to try it. Like many old-school musicians, I have piles of music written in pencil. When I perused the manual, though, I learned that PhotoScore's handwriting algorithm is intended to read freshly copied manuscript, not existing material. And even that process has some severe limitations.
If you need a program for simply scanning published sheet music into Sibelius — for transposing it to an easier key, for instance — PhotoScore is an excellent choice. Sibelius comes bundled with PhotoScore Lite, but PhotoScore Ultimate has many more features, such as the ability to read slurs, triplets, and grace notes.
Easy Does It
Setup was hassle-free. PhotoScore had no trouble communicating with my HP Officejet J5780 scanner. I chose a high-quality page for my first experiment: a cello part originally done in Finale.
Scanning took a couple of minutes, and PhotoScore needed another minute or two to interpret the data. The results were close to perfect — PhotoScore captured all the notes, accidentals, slurs, dynamics, staccato dots, and accent marks, though it did miss one down-bow mark and doesn't read bar numbers. If the spacing of the accents hadn't changed, I would have thought I was looking at the scan itself, not the resulting notation file.
The next page I tried, though still publication quality, threw a few curves at PhotoScore. It had penciled-in fingerings and bowings, bars that ran off the end of one staff and continued on the next, 3/4 time with triplets that weren't consistently marked with little 3s, and 16th-note pickups at two spots. The pencil marks were mostly ignored, though some were turned into random bits of clutter.
The manual advises fixing any rhythmic problems using PhotoScore's editing tools before exporting the notation file to Sibelius. When I tried exporting to Sibelius without doing that, the result was a train wreck, so I went back and tried again. After I spent 10 or 15 minutes editing in PhotoScore, Sibelius read the file with reasonable fidelity, though one entire bar (visible in PhotoScore) near the end of the page mysteriously vanished, along with two notes in an earlier bar. The transcription needed another half hour of editing in Sibelius to deal with incorrect accidentals and numerous slurs that crossed over notes and stems. Being reasonably proficient with Sibelius, I could have created the same page from scratch just as fast, without the fussy process of hunting for glitches.
The Acid Test
Neuratron provides guidelines for writing a score that PhotoScore will read more accurately, but they're quite restrictive. The company recommends using a pen rather than a pencil. Because mistakes are inevitable with a pen, I was relieved to find that PhotoScore ignores notes that have been crossed out.
The program recognizes handwritten notes and rests, including dotted and double-dotted rhythms. It also understands accidentals and slurs but not clefs, time signatures, or dynamics. Neuratron recommends using score paper of precise dimensions and supplies blank templates as PDF files you can print. PhotoScore had no trouble finding the staves on pages printed from templates.
In one test, PhotoScore correctly guessed that the music was in tenor clef (based on the sharp's position in the key signature). But when I tried some piano music in the key of B-flat, it missed a flat and consequently guessed that the lower staff was in tenor clef rather than bass. In both tests, it assumed 4/4 time and arbitrarily dropped notes to make the bars come out right. Other rhythms were misinterpreted, and some chords were turned into single notes.
The whole point of transcribing by hand rather than using Sibelius's very fine input methods is that you should be able to sit with your instrument, relax, and just write. If you have to conform to restrictions such as using a pen, ensuring that ledger lines are straight, leaving out symbols that won't be scanned and then adding them afterward, avoiding stems on the wrong side of note heads, ensuring that stems on eighth rests are long enough and beams aren't too curvy, and so on — then even if the process of scanning the manuscript worked far better than it does, I would never use it.
I hope Neuratron keeps working on its handwriting-recognition technology, because it could be a terrific boon to musicians someday. Unfortunately, however, I would be very surprised to see usable results anytime soon.
Value (1 through 5): 2
Neuratron (distributed by Sibelius)