Re: Scanner and software recommendations


[ Follow Ups ] [ Post Followup ] [ TubeNet BBS ] [ FAQ ]

Posted by Jim Andrada on September 14, 2003 at 15:40:26:

In Reply to: Scanner and software recommendations posted by M.M. Carrasco on September 14, 2003 at 08:38:53:

Having been in the computer "biz" for 45 years or so and having done the computer show demo thing a few times, I've learned that the key to an impressive functioning demo is to make sure you find a happy combination of things that works, and practice, practice, practice not doing anything EXCEPT what works. Follow this rule and you too can have an impressive demo!

Not too different from all the salesmen who could wave the statically charged wand and make all the little foil airplanes fly gracefully through the air. Not at all what happened when the average kid parted with some allowance money and brought the thing home.

Having said all that, the promise of scanning and recognizing music notation is incredibly appealing, as is the idea of scanning and recognizing text. And the latest generation of text recognition (Optical Character recognition or OCR) software has actually gotten pretty good. I even have a package that does OCR on Japanese characters and works - sort of OK!

The problem with OCR, whether for text or music is what to do when it makes a mistake. I once scanned a 300 page text document and passed it through an OCR package and it did OK (it was 12 years ago - things are much better now) - maybe it had 99% accuracy. Which meant that I had to hunt down and fix maybe 10 or 15 errors on each of the 300 pages. It took over a week to fix it up. It might have been easier to just re-type the damned thing!

OK, enough rambling.

I have an Epson "Perfection" scanner which is very nice, but definitely more than you need for OCR input (text or music.) The recent generations of scanners all have resolutions (dots per inch) far higher than what you want for OCR, which tends to work best on material scanned at 200 to 400 dots per inch (DPI.) I don't think you can buy a flat-bed scanner that isn't more than adequate for the job you have in mind.

Something to watch out for though is that the scanner elements, even on fairly pricey scanners are really stingy on the width. I've noticed that they don't generally even cover the full width of the glass, probably on the assumption that anything people want to scan has some white space margin around it. They may give more length, but that's just a mateer of running the scanning bar a little further, whereas wider requires a more expensive (wider) scanning element.

Since a lot of music is actually printed on 9 by 12 paper instead of 8 1/2 by 11, getting the paper positioned on the scanner can be quite tricky, and if it is a somewhat thick piece of music it gets trickier and trickier as you get towards the inner pages. Generally perseverance will pay off, but you might have to make a few passes at each page to get a good, well centered scan. Parts are somewhat easier as they tend to only be a few pages.

I've sometines fastened a little guide along the edge of the scanner a little outside the glass area and this helps a bit. If you can unstaple the music so you're only working with a page at a time, it's usually easier. The good news is that I've never run into a page where the actual content exceeded the 8 1/2 by 11 or 12 limit.

When scanning, it's essential to get a clean, high contrast, straight scan. Originals have to be really clean and sharp, not copies of copies of copies with a lot of fly specks and pencil marks on them. The OCR software won't know the difference between a flyspeck and a cue note!

Software differs in regard to how it handles the scanning job.

The package that comes with Sibelius (Neuratron PhotoScore) drives the scanner itself through a standard interface supported by almost all modern scanner drivers. It also has some capability of filing the scans and recognizing them at a later time as I recall As it performs the recognition it gives you an opportunity to fix up problems with the scan before it hands the complete scan off to Sibelius. It is really essential that you do this and fix everything you find, and you'd better find everything. Sibelius tends to be really snotty about measures and if the scan makes a mistake and inserts an extra note, Sibelius will either throw away the correct note that now won't fit in the measure, or happily screw up every measure that follows. You won't ever recover from this mess in my experience - you HAVE TO clean up the scan before exiting PhotoScore.

And the clean-up isn't always easy. Earlier versions didn't recognize triplets etc for example, so they created an apparent situation of "too many notes in the measure". The latest versions do recognize tuplets, but if it misses one because the little "3" or something wasn't clear, you will have a mess on your hands. Ditto if it mis-recognizes time signatures. Grace notes, cue notes, etc can also give it heartburn - it will often misread them as additional "voices" and hopelessly ball things up if it thinks one of the notes is part of a different voice.

I've tried a different scanning package called Sharp Eye (Braeburn Software) and I was more favorably impressed with it's accuracy, but it outputted the scans in a "NIFF" file and I couldn't figure out how to get this handed off to Sibelius. I believe the latest version of Sibelius/PhotoScore can accept these files, but I haven't tried it myself and haven't really confirmed this.

I just got hold of a copy of Finale They were selling the older version for less just before releasing the latest version., and I did try the "Lite" version of the scanning package that they include.

This package differs in that it accepts a TIFF file that you've scanned separately. I tried it on a couple of really simple things and I was favorably impressed - it did a pretty good job. I haven't tried the full version yet so can't comment on it.

I also think ( THINK is the operative word because I have no real Finale experience as of yet) that you can set it to not be at all strict about measure timing so a few extra notes might not cause a problem. I think it would be much better to have the music notation software let you get away with a lot so you could clean things up in the notoation software which after all is a fine music editing package rather than in the scanning software ( such as PhotoScore) which doesn't have as much editing capability as you'd really want

I think it's an advantage to separate the scanning and recognition as it lets you use scanning software that has pretty good straigtening logic to align the music. And, depending on the software, you can also edit the graphic files to clean up obvious blotches and fly specks etc before trying to recognize it.

I use Paper Port and it lets you do some clean up and store the file as a PDF or its own format, which you can then export to TIFF and input to the Finale package. You do have to be careful in that some packages may not let you choose UNCOMPRESSED TIFF as an output format and this is what the Finale recognition package requires.

Oh, and by the way, when I want to transcribe/transpose a part, I usually just spend the hour or two required and type it directly into Sibelius rather than playing with the scanning packages and spending four hours cleaning things up.

And if I'd learn to use mu MIDI keyboard better and play the music in (in step time where you also set the duration of notes) this would be a real option. I don't think though that you'd want to just "play the music in" as I've noticed that Sibelius will sometimes not follow minor tempo variations and you'll get lots of 64th rests and the whole thing will be thrown off so far you'll never be able to fix it.

Sorry for the long rant, and this is all based on my less than expert (at least about music software) experiences so take it all with a grain of salt or two.


Follow Ups: