Motivic Pattern Extraction in Symbolic Domain - Intelligent Music Information Systems: Tools and Methodologies - page 244

pattern description has been proposed by Rolland
(1999): the Star-Center methodology, influenced
by works in molecular biology, consists in first
constructing a graph whose vertices represent the
different substrings of the score, and whose val-
ued edges represent the corresponding similarity
distances between each couple of substrings. For
each vertex is computed the sum of the similarity
values associated with the connected edges. The
vertices with maximum sum are considered as
“pattern paradigms”: they form the description of
the corresponding patterns, and the occurrences
of the patterns are the set of vertices around these
centers. The cognitive and musical relevance of
this algorithm has however not been studied.

of pattern b is included in the description of pat-
tern a (or a ').
The diatonic description can be directly
obtained from the score when the tonality of a
piece strictly follows the indication given by the
key signature. But in more general cases, local
modulations need to be taken into account through
a proper harmonic analysis. In particular, when
analyzing MIDI files where no tonality is specified
explicitly, diatonic pitch representations need to
be reconstructed using pitch-spelling algorithms
(Cambouropoulos, 2003; Chew & Chen, 2005).
If the compared motives cannot be exactly
matched along the diatonic-pitch-interval dimen-
sion, then they are compared along the gross
contour dimension, which simply indicates the
sense of variation—ascending (“+”), descend-
ing (“-”), constant (“0”)—of the pitch intervals.
For instance, in Figure 5, all the occurrences of
patterns a , a' and a'' , and two more motives (bar
10 and 59) are occurrences of the contour pattern
c . This strategy may sound at first sight rather
simplistic, since the actual interval distortion
is not taken here into account in the matching
process. Nevertheless, this strategy is grounded
by cognitive studies (cf., for instance, Deutsch,
1972; DeWitt & Crowder, 1986; Dowling, 1978;
Dowling & Fujitani, 1971; Dowling & Harwood,
1986; Edworthy, 1985). Studies have shown the
perceptual importance of gross contour dimen-
sions (Dowling & Harwood, 1986; White, 1960):
distorted repetitions of the same motive can be
recognized even if the interval values have been
significantly changed, as long as the gross contour
remains constant.
Similarly, motives can be compared through
a matching of the rhythmic parameters. Two
dimensions are considered in particular: the
rhythmic values indicating the temporal distance
between successive notes, and the position of
each note with respect to the metrical pulsation.
These parameters are directly available in score,
and can be expressed with respect to the implicit
metrical unit. For instance, in Figure 5, with

multiple-viewpoint approach

Instead of using numerical distances, melodic
comparison can be carried out through an exact
matching along a multiple set of musical dimen-
sions (Conklin & Anagnostopoulou, 2001). For
instance, instead of computing numerical distance
between chromatic-pitch-interval descriptions,
motives are compared along chromatic-pitch-in-
terval and diatonic pitch-interval descriptions in
parallel. Diatonic pitches represent positions of
the notes on a given tonal scale, and diatonic pitch
intervals indicate the successive scale transitions
on this tonal scale. If two motives have exactly the
same chromatic-pitch-interval descriptions, then
they are considered as chromatically matched. For
instance, in Figure 5, pattern a contains occur-
rences that present exactly the same chromatic
description. If on the contrary, there is a major-
minor difference between the two motives, then
the two motives are not chromatically matched,
but they are still diatonically matched. In Figure
5, the “major-third” ( a ) and “minor-third” ( a' )
patterns can be unified into a “third” pattern
b that represents the identification along the
diatonic representation. Pattern b is considered
as less specific than patterns a (or a '), which is
notated “ a > b ” in Figure 5, since the description