Motivic Pattern Extraction in Symbolic Domain - Intelligent Music Information Systems: Tools and Methodologies - page 253

therefore be added to the representation of the
pattern, as shown in the Figure .
A mechanism has been added that unifies all the
possible rotations of the periodic pattern ( b'c'd' ,
c'd'b' , d'b'c' ) into one single cyclic pattern. For
instance, in Figure 15, the periodic sequence be-
ginning in a different phase than previously (on
an upbeat instead of a downbeat) is still identified
with the same cyclic pattern.
By construction of the cyclic pattern, no
segmentation is explicitly represented between
successive repetitions. Indeed, the listener may
be inclined to segment at any phase of the cyclic
chain (or to not segment at all). However it would
be interesting to estimate the positions in the cycle
where listeners would tend to segment. Several
factors need to be taken into consideration, such
as primacy, local segmentation or global context.
For instance, a primacy-based segmentation will
favor the period that appears first in the sequence,
depending on the phase at which the cyclic se-
quence begins. Global context corresponds to the
general segmentation of the piece, based on the
major motives and the metrical structure. This
might be considered in future work.

minimum number of pieces a pattern must occur
in) was set to 10. For the soprano line, 33 longest
significant melodic interval patterns were found.
For the alto line, 29 were found. One single signifi-
cant melodic contour pattern has been detected. A
total of 57 longest significant pitch class interval
patterns were found in the bass lines. Chromatic
stepwise movements, a well-known common
pattern in the bass line of the chorales, have been
detected too. The combination of melodic intervals
and duration leads to the discovery of 23 longest
significant patterns. Hence this method offers a
synthetic description of most common patterns
in a given corpus of music.
Cambouropoulos's model (2006) is aimed at
studying some particular problems in melody seg-
mentation. It suggests segmentation boundaries on
a given selection of short melodic examples ( Frère
Jacques , finale theme from Beethoven's Ninth
Symphony , opening melody of Mozart's A-Major
Sonata , K. 331, etc.) but is not supposed to work
on more complex repertoire, neither to highlight
the motivic patterns that have an impact on the
inference of these global segmentations.
Rolland's approach (1999) was primarily dedi-
cated to the analysis of Charlie Parker's improvised
playing, and to the characterization of recurrent
melodic patterns (“formulae”) in a corpus of 250
solos transcribed by the musicologist T. Owens
(1974). A restricted corpus made of 10 Parker
solos in the “C major—Blues” category of Owens'
corpus (3 takes on “Cool Blues”, 3 on “Relaxin'
at Camarillo” and 4 on “Perhaps”) is used for
the computational experiment. The cumulative
length is about 2000 notes and rests, each solo
(except one) comprising between 25 and 40 bars
and between 130 and 250 notes. The similarity
graph obtained from the analysis contains about
12000 vertices and 40000 edges; the largest star
has a size of 35 edges (plus its own center). The
longest pattern has a prototype of length 27 notes.
58 patterns have been found whose occurrences
are spread out in all 10 of the corpus' sequences.

results

review of other approaches

Conklin and Anagnostopoulou's study (2001)
analyzes 185 Bach chorales, comprising a total
of about 40000 notes. About 20 viewpoints were
encoded; most of them pertain to melodic and
rhythmic aspects of the chorales. Several view-
points are test viewpoints used for combination
(“linking”) with other viewpoints, or in order to
restrict certain viewpoints to precise temporal or
metrical locations within the piece (“threading”).
Additional viewpoints model harmonic or verti-
cal structures. The adjusted p-value cut-off for
patterns was set to 0.01 and the parameter k (the