Robotics Reference
In-Depth Information
availability of music in computer readable form, such as MIDI 28 files,
it seems likely that research in music attribution systems will soon be
making faster progress.
If a computer can recognize the style of a particular composer, might
it also be able to distinguish between the playing styles of different mu-
sical performers? This is a far more difficult task because it means dis-
tinguishing between performances in which, in theory, all the notes and
their durations are the same. 29 Some highly experienced lovers of classi-
cal music are able to distinguish the playing of one world-class violinist
from another, similarly with pianists and singers. But these are the ex-
ceptions. The vast majority of concert-goers do not even notice most
of the time when a soloist plays a wrong note, and they are unable to
distinguish between the tones and playing styles of the leading soloists.
Research in Austria has shown that it is indeed possible for a machine
to learn to distinguish and recognize famous soloists based on their style
of playing. Gerhard Widmer and Patrick Zanon have identified several
features of expressive timing and musical dynamics and used these fea-
tures to train six different learning mechanisms. The training data com-
prised 12 selected movements from Mozart piano sonatas, each played
by five world class pianists: Daniel Barenboim, Glenn Gould, Maria Joao
Pires, Andras Schiff and Mitsuko Uchida, and Roland Batik, a Viennese
pianist who is highly skilled though not in the same class as the other
five. The recognition rates of the best of the six learning algorithms
ranged from 66 percent for Uchida up to 75 percent for Batik. Even
more remarkable was the system's performance when tested on the music
of Chopin. Although it had been trained on CDs of the various pianists
playing Mozart, the software was able to score as high as 68 percent when
the performers were playing music by Chopin, whose composing style is
quite different.
28 Musical Instrument Digital Interface—A standard for representing musical information in a
digital format.
29 After all, the performers are playing the music as it was written, in theory.
Search WWH ::




Custom Search