Identifying Saxophonists from Their Playing Styles - Intelligent Music Information Systems: Tools and Methodologies - page 106

has focused on classical piano music where the
tempo of the performed pieces is not constant.
Thus, these works focus on global tempo and
energy transformations while we are interested
in notelevel tempo and energy transformations
(i.e., note onset and duration).
Widmer (2001; 2002) reported on the task of
discovering general rules of expressive classical
piano performance from real performance data
via inductive machine learning. The performance
data used for the study are MIDI recordings of
13 piano sonatas by W.A. Mozart performed by a
skilled pianist. In addition to these data, the music
score was also coded. The resulting substantial
data consists of information about the nominal
note onsets, duration, metrical information and
annotations. When trained on the data an induc-
tive rule learning algorithm discovered a small
set of quite simple classification rules (Widmer,
2001) that predict a large number of the note-level
choices of the pianist.
Tobudic and Widmer (2003) describe a rela-
tional instance-based approach to the problem of
learning to apply expressive tempo and dynamics
variations to a piece of classical music, at differ-
ent levels of the phrase hierarchy. The different
phrases of a piece and the relations among them
are represented in first-order logic. The descrip-
tion of the musical scores through predicates
(e.g., contains(ph1,ph2) ) provides the background
knowledge. The training examples are encoded
by another predicate whose arguments encode in-
formation about the way the phrase was played by
the musician. Their learning algorithm recognizes
similar phrases from the training set and applies
their expressive patterns to a new piece.
Other inductive approaches to rule learning
in music and musical analysis include (Dovey,
1995), (Van Baelen & De Raedt, 1996). In Dovey
(1995), Dovey analyzes piano performances of
Rachmaniloff pieces using inductive logic pro-
gramming and extracts rules underlying them.
In Van Baelen & De Raedt (1996), Van Baelen
extended Dovey's work and attempted to discover

regularities that could be used to generate MIDI
information derived from the musical analysis
of the piece.
There are a number of approaches which
address expressive performance without using
machine-learning techniques. One of the first at-
tempts to provide a computer system with musical
expressiveness is that of Johnson (1992). Johnson
developed a rule-based expert system to deter-
mine expressive tempo and articulation for Bach's
fugues from the well-tempered clavier . The rules
were obtained from two expert performers.
A long-term effort in expressive performance
modeling is the work of the KTH group (Bresin,
2002; Friberg, 1998; Friberg, Bresin, & Fryden,
2000). Their Director Musices system incorpo-
rates rules for tempo, dynamic and articulation
transformations. The rules are obtained from both
theoretical musical knowledge, and experimen-
tally from training using an analysis-by-synthesis
approach. The rules are divided into differentia-
tion rules which enhance the differences between
scale tones, grouping rules which specify what
tones belong together, and ensemble rules which
synchronize the voices in an ensemble.
Canazza, De Poli, G., Roda, A., and Vido-
lin (1997) developed a system to analyze the
relationship between the musician's expressive
intentions and her performance. The analysis
reveals two expressive dimensions, one related
to energy (dynamics), and another one related to
kinetics (rubato).
Dannenberg et al. (1998) investigated the
trumpet articulation transformations using
(manually generated) rules. They developed a
trumpet synthesizer which combines a physical
model with an expressive performance model. The
performance model generates control information
for the physical model using a set of rules manu-
ally extracted from the analysis of a collection of
performance recordings.
Nevertheless, the use of expressive perfor-
mance models for identifying musicians, either
automatically induced or manually generated, has