End of August the German computer magazine c’t published a thorough review of programs that digitize sheet music (18/2019, page 122-126). The three programs that achieved ‘good recognition’ were PhotoScore, SharpEye, and SmartScore with a reported 98% accuracy – the same programs that did well in the review that c’t published in 2001!

PlayScore was not included in the review. That is understandable, since the focus of the program is to play scores, but it is unfortunate nonetheless: it does support MusicXML export. So I decided to use the ground truth that I have created to assess PlayScore, and compare it to SmartScore and PhotoScore in terms of notes recognition accuracy.

To run the assessments, 14 scores were used from 1 page to 4 pages, with 318 staves in total. After thinking for a while how to best summarize the results, I grouped the scores as follows:

• Single stave/solo scores written for violin, cello and flute, with some use of chords and voices (not uncommon for string music) [88 staves]
• Piano scores – this includes pages from sonatas by Scarlatti and Beethoven, a Ravel sonatina, and a Scott Joplin rag [110 staves]
• Ensemble scores  – a Haydn piano trio, a Mozart string quartet, and a Brahms violin sonata [120 staves]. Here only the piano scores were used, which means that the violin/cello scores were reduced in size (to about 60% of the piano score).

	PhotoScore 8.8.7	SmartScore X2	PlayScore 2
Solo	92.8	89.3	94.2
Piano	83.1	85.6	91.8
Ensemble	96.5	94.2	97.5

PlayScore is doing very well! When ranking the programs on each of the 14 scores, PlayScore finished first 10 times, PhotoScore 3 times, and SmartScore 2 time (there was one tie).

Accuracy is substantially lower than reported by c’t. Possible reasons are :

• More challenging scores: the solo scores included chords, voices, and tuplets.

• In the piano scores the accuracy for the Scarlatti sonata was about 98%, but the scores for the much more difficult Ravel and Scott Joplin pieces were substantially lower.

I was surprised that the smaller size of the violin/cello parts in ‘Ensemble’ did not negatively impact accuracy.

Some more information on PlayScore (www.playscore.co):

• It is a relatively new product.
• It is primary purpose is to play scores based on photos taking with a mobile device.
• It is a multi-platform product running on both Android and IOS device. (The IOS version is substantially better, though!)
• The free version supports single pages. The other versions are available for subscription (‘Productivity’ handles multiple pages, and costs $3.99/month or $9.99 per year, ‘Professional’ adds support for PDF import and MusicXML export for $5.99/month and $22.99/year)

It is great to see such a capable new product entering the OMR market!