[1]Several pieces have been written for performers accompanied by a tape (e.g. by Stockhausen and Boulez), but a tape is a very inflexible medium: it can't react to the performer(s). A computer, listening to the performer(s) via a microphone and operating a synthesizer (or another instrument that can be driven by a computer), can.

[2]According to the Oxford combined dictionary of current English & modern English usage.

[3]Musically speaking, there is no such thing as a perfect performance, because there is no `best' interpretation of a musical piece; every interpretation has its merit. In this case, I mean a performance without errors: every note in the score has been played, and nothing but the notes in the score have been played.

[4]In a score notated as p for piano, and ff for fortissimo, respectively.

[5]Rests do appear in a performance, but it is impossible to derive the presence of a rest in a polyphonic performance. The presence of a rest in a performance can only be shown by matching it with the score.

[6]It is not very easy, nor readable, to express such a complex type hierarchy in CIP-L (see section 3.6.1). Were the data type `musical object' be implemented in some kind of object oriented language (e.g. CLOS, see Steel [1990] or Keene [1989]), the complex hierarchy would present no problems at all.

[7]The representation used in this report doesn't take things like timbre, or pedal information in the case of piano music, into account.

[8]It has been said already that the information lost by representing the score in another way can be vital to the matching process (Balaban [1989] and Honing [1992]).

[9]Actually, they consist of two sub-phrases; the length of the phrase of which they are sub-phrases is four measures, but since only two are displayed, we will consider the sub-phrases to be phrases.

[10]A leaf in a score tree is a simple object, but since we chose to ignore the existence of rests (see section 2.1), a leaf is a note.

[11]The timing at the beginning of a phrase is very different from the timing in the middle or at the end of a phrase. Though very subtle, this information could be used.

[12]This means we also have limited the music allowed to music played on instruments that are discrete, i.e. that have nothing between e.g., the notes numbered 60 and 61, like a piano. Music played on a violin could not be represented in this way.

[13]Note that from a musical point of view, they are by no means erratic. Moreover, a trained musician can usually make a very good prediction of them.

[14]Music isn't about playing every note beautifully, it is about telling a story; and if you can capture and hold the attention of your audience with the story you're telling, no one will care if you mispronounce a word. Being able to play any piece of music without errors does not make one a musician. Therefore, one cannot expect a perfect performance, and hence, a matcher should be prepared to deal with performance errors.

[15]There are lots of examples of this. A note could be played slightly later than a matcher would expect, to obtain a more dramatic effect. There are even cases where a performance note is deliberately (and quite correctly) played at another pitch than the corresponding score note. An example is the second recitative of Bach's Saint Matthew Passion, where there are at least four occasions where a performer would sing a different note than written.

[16]The original scheme has an extra level, not needed for our purpose. Refer to Palmer & Van de Sande, 1993 for an explanation of that level.

[17]Large calls it a score, and the table a score table, but as I am using the notion of a score in a quite different way, I have chosen a different term.

[18]It is a transcription of the fugue in g-moll for lute. There is also a version for violin: the second movement of the first sonata, in g-moll, and a version for organ, in d-moll.