Tune Comparisons

Introduction

"abc-compare" is a program that reads a list of tunes (in ABC format) and produces numbers representing how "similar" each pair in that list is (as a percentage), along with some identifying details on each tune.

Read the Readme to learn more.
dowload the source
download abccompare.exe, a pre-compiled Windows (console) executable

Sample Results

I started wondering about all this when I was getting the Aird Collection ready for inclusion in the Tunebook, and noticing tunes I was already familiar with from the Winder collection. And I wondered even more when I saw Johnny Adams' Village Music Project :- how to spot instances of similar tunes that turn up in different versions in different collections ? To some extent, of course, just by playing through the collections (!) and seeing what you recognise. But, maybe some software would be nice ... I had a look around, and couldn't seem to find any ... so, in the end, I've had a try at writing it. I fed abc-compare with (some of) the substantial body of old English & Scots publications and manuscripts that can be found on the web - the abc files available from the Village Music Project, and various others taken from my own Richard Robinson's TuneBook. (There is a complete list of the files I used at the bottom of the results page).

Given a set of lines, each "scoring" the "similarity" between a pair of tunes, what can you do with it ? The obvious thing seemed to be to build groups out of the pairs. ie, if tune 1 is "similar" to tune 2, and tune 2 is "similar" to tune 71, and tune 71 is "similar" to tune 187, then we have a little set of tunes that are all "similar" to each other. For, as they say, some values of "similar". But, what value ? I built a little perl script to build these groups and format them as a webpage, giving links to enable viewing the tunes, and I played around with it. The lower the threshold value, the bigger the groups, obviously, and the greater the likelihood of spurious matches. Having spent a few hours playing with this, I thought I might as well put it up here for people to play around with for themselves ...

For the results that I have, on this set of tunes, 66% seems like the best compromise. This script being rather slow, here is a set of results that I made earlier.

Try it for yourself

Nearly all of this material can be found on the web, but a couple of things seem to have crept in for which I can't give links to the tunes - "Playford" is a local conglomeration of stuff I found on the web and failed to keep a note of where it came from; "the Fiddler of Helperby" is only partly available on the web - the publishers (Dragonfly Music, to whom be thanks) have allowed us to publish 10 tunes out of it, which unfortunately are not necessarily the ones exposed here. But I left the conections in anyway, in spite of not being able to show the tunes, in order to make the point that there are advantages to having tunes available in digital formats. None of this could have been done with tunes represented on paper, or in sound. "You can't grep dead trees".

And, by a curious coincidence, one other match that doesn't show up in the table :- I recently bought a CD of (mostly) old Danish tunes ("Spring", by ULC, GO0100), on which they play a tune called "Fredricks Contra", taken from the Bast brothers' manuscript (Lolland, Denmark, 1760) ... which turns out to be another name for "Lads of Dunse", "the Lads Dance", "the Land We Live In", etc etc.

The links to tunes that you find here use John Chambers' very excellent Tune Finder to build the images, from the original ABC files wherever they live on the 'net. This is to avoid the necessity of keeping copies up-to-date, in the event of changes to the originals.

I have rejected cases where 2 or more tunes within the same collection are found to be similar - you won't see any of these unless they also resemble a tune in a different collection. If you have any comment on any of these tunes, remember to check which collection it comes out of, to find out whether the VMP or myself are the people to contact.