alan little’s weblog

itunes and classical music

14th March 2004 permanent link

I was tidying up my iTunes music collection yesterday evening, and thinking about what an appalling useless mess the id3 tag system and the Gracenote CDDB database are for classical music. I’m not the first to think this – see this piece by Brian Tiemann, and this one by Kimbro Staken.

(Yes, tidying up my music collection in iTunes is my idea of a good time on a Saturday night. I have a ten month old baby, what else do you expect? Maria wanted to watch one of those appalling “Pop Idol” style karaoke shows; I couldn’t bear it, but didn’t have the energy to write anything.)

Brian thinks:

it'd be a matter of adding an extra field or two to hold "Work Title" (e.g. Konzert für Klarinette u. Orchester A-dur ) and maybe a movement number, and then it'd involve simply telling the software to handle organizing songs differently that have these fields set. If the "Classical" genre is selected, organize first by Composer; and then, instead of the "Album" field, show the "Work Title" field in the browser. Albums and artists (performing groups) can then become secondary meta-data, not used for organizing the tracks once they're in the database.

If only it were that simple. I’ll talk about what would be needed to really do the job properly, and how to use the fields in iTunes to approximate doing the job properly. The fields available in iTunes, and the ones Brian proposes, are:

artist

Not straightforward. For classical orchestral music I always want the conductor and the orchestra, plus for concertos the soloist. (I never want the composer, which some idiots put in the “artist” field, unless the recording actually is the composer playing or conducting their own work). For chamber groups the name of the group is normally good enough, but not always. Take the Borodin and Juilliard string quartets – two very famous ensembles that have existed for decades (longer even than the Rolling stones) but with major changes of personnel. I believe the Juilliard Quartet now has none of its original members and the Borodin Quartet has one. In the case of the Borodin, the crucial watershed is before and after the original leader, Rostislav Dubinsky, defected to the west in 1975. The post-Dubinsky quartet is also good but it’s different; and if I’m looking for recordings by Dubinsky then I would want to see pre-1975 but not post-1975 Borodin Quartet, plus the Borodin Trio that he formed in America, plus anything he did with other chamber groups or as a soloist.

Artist name search needs to be multilingual. A search for “Vienna Philharmonic” that doesn’t return “Wiener Philharmoniker” is useless; “Furtwängler” should also match “Furtwangler” and “Furtwaengler”.

For orchestras I’m personally only interested in the name of the orchestra, not individual musicians. (My principal classical music adviser points out an important exception to this in the case of anything featuring one O. Little anywhere in the violas – the Royal Liverpool Philharmonic Orchestra’s Beethoven symphony recordings with Charles Mackerras, for example.)

composer

iTunes’ search is clever enough to do partial matches, so if we have anything at all in the composer field, then “Beethoven, Ludwig van” and “L v Beethoven” both match “Beethoven” without any problem. A bigger problem is the classical CDs in the gracenote database that either don’t have composer at all, or where some cretin has put composer in the “artist” field and (if you’re lucky) stuffed the artist into the “album” title.

But once again we have the not-small question of different spellings of composers’ names in different languages. A simple example: Händel (correct German spelling) = “Haendel” (alternative German spelling, acceptable for example in primitive ASCII-only computer systems) = “Handel” (normal English spelling). I would expect a search for any of these to match all of them. Another example: Дмитрий Дмитриевич Шостакович (correct Russian spelling) = Dmitriy Shostakovich (English transliteration) = Dmitrij Schostakowitsch (German transliteration) = Dimitri Shostakovich (alternative English spelling of the first name, as used last week by Brian Micklethwait). In cases like this, where there are several possible acceptable English spellings of “Dmitriy”, I would expect a decent search system to be able to cope with near-misses rather than having to specify all the possibilities in advance.

In all these cases, there is arguably a canonically correct name – the full version most commonly used in the composer’s native language – and there is a version I personally would be mostly likely to use to search, which would normally be the most common English-language spelling of the composer’s surname. But it wouldn’t be necessary for any version of the name to have any special status in the system in order for search to work correctly, a big pile of equal-status synonyms would do just fine. (What about composers with the same surname? I would expect a search for “Bach” to turn up works by Johann Sebastian Bach. I would not want to have to specify “JS Bach” to avoid seeing things by his lesser relatives. This would be less clear-cut for “Strauss” though. But in any case, it could be easily dealt with by using some combination of rating / most recently played / most frequently played to weight search results.)

album / work title

What does the “work title” of a piece of classical music actually consist of? I’ll take as an example Beethoven’s String Quartet no. 9 in C op.59 “Rasumovsky” no.3. Let’s parse that. It contains at least six pieces of information:

String Quartet. It’s a string quartet. This is a conventional description of a particular type of music. Note that I said it’s a description, not a name, therefore I again expect searches on it to be multilingual. I have lots of German editions of classical CDs and a few French ones, so anything that doesn't recognise “Streichquartett” and “quatuor à cordes” when I search for "string quartet" is useless. This is not the last we will hear of this theme. Also, I would normally take “quartet” on its own to be a synonym for “string quartet” (two violins, viola, cello) – any other group of four instruments would always be a [something else] quartet.
no. 9. It’s the ninth string quartet Beethoven wrote. I would expect a search for “Beethoven’s ninth quartet” to find this work. For some reason, the work-of-this-type-number is most commonly used for symphonies, and normally means symphony if nothing else is specified – “Beethoven’s Fifth” definitely means his fifth symphony and not his fifth piano sonata or whatever. (This seems only to be true for composers whose symphonies are particularly well known. Not always even then - I would not say “Shostakovich’s Seventh”, because people generally don’t seem to refer to Shostakovich’s works in that way and it would not be completely unambiguous – did I mean seventh symphony or seventh string quartet, or something else Shostakovich wrote seven or more of? Does it, in fact, only apply to German symphonists whose names begin with B or are “Mahler”?)
in C . It’s in the key of C. This is normally referred to as part of the identity of a work.
op. 59 . It’s part of Beethoven’s 59th opus, or published work. This is the normal way of giving some kind of id number to a composer’s pieces of music (but see below). Note that “part of”, however. Music was commonly published in sets, so an opus number can refer to one or more pieces of music. A good classical search would be smart about opus numbers. I would want a search for “Beethoven opus 18” to return a set of six string quartets and not Beethoven’s 18th piano sonata.
“Rasumovsky” is the commonly used name of this set of string quartets – in this case, because they were dedicated to Count Rasumovsky, a Russian diplomat who [commissioned them? Was generally a patron and admirer of Beethoven? I’m not sure]. Anyway, it’s quite common for particularly famous and popular pieces of music to have names – other examples are Mozart’s confusingly-named “Haydn” quartets, Beethoven’s Third Symphony universally referred to as “The Eroica”, or Schubert’s famous string quartet “Der Tod und das Mädchen”. Which brings us back to … translation and internationalisation. Unlike titles of pop songs, names of pieces of classical music are, if they aren’t names of people, often translated. So “Der Tod und das Mädchen” is always referred to as “Death and the Maiden” in English. But not always. Beethoven’s “Pathetique” sonata generally isn’t translated from French into English or German. Note also that sometimes the name refers to a single piece of music, sometimes to an opus containing multiple pieces.
But if a name refers to more than one piece of music, then they’re always parts of a set by the same composer, right? No. There are two famous pieces of music referred to by the name “Kreutzer Sonata”. One is Beethoven’s violin sonata no.9 in A major, opus 47. The other is the second string quartet by Janacek – which is named not directly after the Beethoven violin sonata, but after a story by Tolstoy. The story by Tolstoy is named after the Beethoven violin sonata. (Tolstoy is, bizarrely, ahead of Beethoven in a google search)
no. 3. This one the third (and last) of the opus 59 “Rasumovsky” quartets.
A few particularly prolific and famous composers - Johann Sebastian Bach, Mozart, Schubert (though not, for some reason, Haydn or Beethoven), have actually had their works catalogued with authoritative work numbers. So “BWV 988” is the standard, correct, unambiguous id for Bach’s Goldberg Variations. These are clearly much better than opus numbers because they always refer to single pieces of music, and where they exist they seem to be used in place of opus numbers. But is this always reliably the case? Twenty years of data modelling has taught me to assume that there’s always an exception – so would a good classical search system need to allow for somebody coming along and searching for a Bach opus number instead of a BWV-number? I don’t know.

People who really know about this stuff could doubtless point out lots of errors and omissions in the above. But it’s broadly correct, and hopefully enough to get the message across that an apparently simple matter of an extra field called “work title” barely begins to scratch the surface of what’s actually there.

I picked Rasumovsky Quartet no.3 as an example because it is the subject of one of my all-time favourite recordings of anything by anybody, in the form of a stunning 1960s performance by the Smetana Quartet that was originally recorded by the excellent Czech label Supraphon, and is now available as a cheap reissue on Cantus, catalogue number 500152. HIGHLY RECOMMENDED

Brian wants to stuff all this into one field and call it “Work Title”, rather than using “Album”. Personally I was inclined to use “Album” – it’s a matter of complete indifference to me which CD something originally happened to be on. So I tried putting standardised work titles in the Album field. Beethoven’s Symphony No. 3 In E-Flat Major, Op. 55 “Eroica”, to take one example of a work I have about half a dozen different recordings of (out of the 200 or so recordings of it that exist). Ah. Problem. iTunes now sorts all the first movements together, then all the second movements etc. Not useful. What I want, of course, is all four movements of one recording, then all four movements of the next. In order to achieve which I have to take my nice standardised work titles and append some kind of “recording id” to them, duplicating information I already have in Artist and Date. Blah.

Oh, and “Date” in the rare cases when it’s set in the gracenote database is normally the year the CD was released, about which I don’t care a fart. I want the year the recording was made, please.

I don’t usually care if I have two different versions of the same recording. The exception to this is if one of them is remastered or otherwise of noticeably better quality. Take Wilhelm Furtwängler’s 1944 recording of the Eroica with the Wiener Philharmoniker, a.k.a. Furtwangler/VPO Beethoven op. 55, or any of a thousand other possible descriptions of the same thing. I originally downloaded this as an mp3 from emusic, and was so impressed I decided to buy a “proper” CD version to see if I could hear a difference. I could. (Also HIGHLY RECOMMENDED)

song title

I would prefer to just have the names of movements in here. It isn’t very important for classical music anyway – I rarely want to listen to single movements of things and would almost just as soon have an entire work as one track, but I can’t be bothered to do all the clicking about to glue them together in iTunes.

track number

The track number that something was on the original CD is completely irrelevant. As Brian says, you either think “So Track 14 is really the fourth movement of the third work on the disc …”, or you go through and edit everything laboriously by hand to say this is movement 2 of 3 in so-and-so concerto, rather than track 7 of 13 on some irrelevant CD. It would be nice if iTunes provided a quick and easy way to do this, but it doesn’t.

so what do i propose to do about all this?

The information in gracenote is such complete garbage it’s almost a write-off. I just looked up a CD of Janacek’s string quartets by the Talich quartet. What I find is typical:

The titles of the actual works are in the “song name”, so for the four movement string quartets we have the same song name four times. This is actually not a blatantly stupid choice given that the whole gracenote database is built around the CD as its fundamental unit, which means you can’t put anything useful for classical music in the Album field
The actual “song names”, the names of the movements, are in the “artist” field. Great. I’ll have to cut and paste these out one by one into the Song Name before I can put the actual artist in. And where (if anywhere) have they put the artist?
… nowhere. Composer is blank too

What I intended to write about is:

Given that the gracenote data is a write-off and I’m resigned to having to input all the metadata myself if I want anything half usable, how am I actually going to use the fields available in iTunes?
How would I actually do it myself in an ideal world, given that what I’m trying to achieve is some kind of super-iTunes cataloguing / sorting / jukebox system, not a research database for musicologists? This, if I ever have time to write it, will get into all sorts of interesting stuff about how to keep an ultra-simple user interface to a search system that, behind the scenes, does all kinds of smart language-independent, user-adaptive stuff.

But this is already too long and it’s getting late. Another day.

related entries: Music Mac