Magazine Archive

Home -> Magazines -> Issues -> Articles in this issue -> View

Article Group:
Quality Control

Digital Music MIDIScan

OCR music software for the PC

Imagine what it would be like to scan a piece of sheet music and have it turned into a MIDI file. Imagine the MIDIScan.


Optical character recognition systems have been with us for some time, but now the technology has been adapted to read musical scores. Ian Waugh judges the success of the system...


This is so obviously an ideal computer application, you might wonder why no one has thought of it before. The purpose of MIDIScan is to convert printed music into a MIDI file. It does this by looking at a scanned image of a page, searching for staves and then analysing the notes it finds there.

The program runs under Windows and as well as the software, you'll need a scanner. This could be of the hand-held type which would cover a page in two runs (with the pieces spliced together), but a flat-bed scanner would provide much more reliable results. The software works with TIFF files (one per page) through a conversion process which involves a series of individual stages. The first of these is the editing of the scanned TIFF file. Although the program usually ignores non-note data, sometimes it doesn't, so you may have to remove guitar chords, lyrics and music instructions etc, in case it tries to interpret them as note data. What to take out may require a little trial and error.

The next stage is the music recognition process. The program analyses a page at a time and you can construct a complete score by selecting several TIFF files to be analysed in order. The program will analyse both ensemble scores and single, 1-stave scores, but you have to set a recognition region to determine how far above and below each stave the program looks for notes.

This is actually one of the weakest aspects of the program. If the notes stay within their stave boundary all will be well, but quite often a note in the bass clef (in a piano part, for example), will stray into treble clef territory and MIDIScan will, like as not, assign these to the upper stave. Similarly, in many older scores the staves are quite close together and this can result in incorrect note assignments. As it proceeds through the recognition process the program outlines the recognition area so you can see which notes it hits and which it misses. With problem scores, what is really needed is the ability to draw a line with the mouse between the upper and lower parts.

Each page is handled separately. The program reports the number of staves found and if gets it wrong, you can correct it. MIDIScan can process up to 16 staves per system and you can make it ignore certain staves if you wish.


1. An original TIFF scan of a song by our Tina.


2: The file after conversion to MNOD format. You can compare the original scan to the MNOD file.


After recognition, the program produces a half-way file called MNOD (Music Notation Object Description). The screen splits in two and you can see the scanned image in the upper half and the MNOD version in the lower half. This looks like a very neat DTP score. But - and it's a big but - it may not be exactly the same as the original score. Some note durations may not have been interpreted correctly, some ties may be missing and some notes may even be missing, too.

To put things right, you can edit the MNOD file using the MNOD toolbox which contains all the tools you'd expect to find in a scorewriting program. You can edit notes, insert and delete objects, change clef, insert a time signature and so on. In fact, you must insert a time signature otherwise the resulting MIDI file will cause a "Divide by Zero" error when you try to load it into a sequencer - something you learn the hard way or by referring to the troubleshooting guide.

After editing the MNOD file you begin MIDI conversion. You can assign patch numbers and MIDI channels to each of the tracks (parts) and set a tempo. Then the program saves its work as a MIDI file.

The information in the manual is comprehensive enough, but it's not very well ordered. Having said that, once you understand the basic conversion processes - and don't forget to set a time signature - the program is fairly straightforward to use.


3: The MNOD toolbox contains scorewriting edit tools for correcting the score.


4: And this is how the score appears in Cubase Lite without any editing. It's pretty close.


So how well does MIDIScan work? This depends entirely on the complexity and the layout of the score. It doesn't appear to like the pyrotechnical displays of Liszt, generally preferring scores which are proportionally spaced out. Give it a nice clean score and it will do a creditable job.

OCR (Optical Character Recognition) for text is usually said to have an accuracy of so-many percent - the best comes close to 99%. It's difficult to say how accurate MIDIScan is because the results depend so much on the original score. With standard piano/vocal sheet music, however, I'd suggest it has an accuracy of around 90%. With more complex scores this reduces dramatically.

If you have a degree of playing ability it could well be quicker to record a score in a standard sequencer. If, however, you're skills aren't up to scratch, MIDIScan could save you a significant amount of time. It does help if you can read music, if only to check and correct the MNOD score.

Who would use MIDIScan? That intrepid band of people who convert your favourite songs into MIDI files seem the most likely target, although quite a few of them actually create their files from the record without referring to a score at all. Giggers could use it to create their own backing tracks, though arrangements would be somewhat minimal unless they work from scores (these are available in books called Rock Scores by Music Sales who, interestingly, are already busy creating MIDI file versions of their books). The program could well appeal to the non-pro, too, although the price may be a sticking point for them.

MIDIScan is quite clever in what it does - it can even read staves which have been scanned at a slight angle - but it needs a little refinement before it becomes a serious production tool. This, however, is the first commercial incarnation of the program and updates shouldn't be too difficult to implement. It's at the same stage of development that text OCR systems were a few years ago. If its development follows a similar pattern, the future for music OCR looks good.

THE LAST WORD

Ease of use Not difficult, but a little involved and can be time-consuming
Originality Never seen anything like it in me life, guv
Value for money A bit pricey - but then so were the first text OCR programs
Star Quality In the making
Price £334.87 inc. VAT
More from Digital Music, (Contact Details)



Previous Article in this issue

Korg Audio Gallery AG-101 & AG-102

Next article in this issue

Analogue Systems FB3


Music Technology - Copyright: Music Maker Publications (UK), Future Publishing.

 

Music Technology - Feb 1994

Quality Control

Gear in this article:

Software: Misc > Digital Music > MIDIScan


Gear Tags:

PC Platform

Review by Ian Waugh

Previous article in this issue:

> Korg Audio Gallery AG-101 & ...

Next article in this issue:

> Analogue Systems FB3


Help Support The Things You Love

mu:zines is the result of thousands of hours of effort, and will require many thousands more going forward to reach our goals of getting all this content online.

If you value this resource, you can support this project - it really helps!

Donations for September 2020
Issues donated this month: 0

New issues that have been donated or scanned for us this month.

Funds donated this month: £34.00

All donations and support are gratefully appreciated - thank you.

If you're enjoying the site, please consider supporting me to help build this archive...

...with a one time Donation, or a recurring Donation of just £2 a month. It really helps - thank you!
muzines_logo_02

Small Print

Terms of usePrivacy