Problems and opportunities of applying data-and audio-mining techniques to ethnic music

Abstract

Current research in music information retrieval focuses on Western music. In music from other cultures, both musical structures and thinking about music can be very different. This creates problems for both the analysis of musical features and the construction of databases. On the other hand, a well-documented digitization offers interesting opportunities for the study and spread of 'endangered' music. Here, some general problems regarding the digital indexation of ethnic music are given, illustrated with a method for describing pitch structure, comparing Western standards with African music found in the digitization of the archives of the Royal Museum of Central-Africa in Tervuren (Brussels).

Full text

Page 1

Problems and Opportunities of Applying Data- & Audio-Mining Techniques
to Ethnic Music

Dirk Moelants, Olmo Cornelis,
Marc Leman
IPEM-Dept. of Musicology, Ghent
University
Blandijnberg 2
B-9000 Ghent, Belgium
Dirk.Moelants@UGent.be
Jos Gansemans
Dept. of Cultural Anthropology, Royal
Museum of Central-Africa
Leuvensesteenweg 13
B-3080 Tervuren
jgansemans@africamuseum.be
Rita De Caluwe, Guy De Tré,
Tom Matthé, Axel Hallez
TELIN, Ghent University
Sint-Pietersnieuwstraat 41
B-9000 Ghent, Belgium
Rita.DeCaluwe@UGent.be
Abstract
Current research in music information retrieval focuses on
Western music. In music from other cultures, both musical
structures and thinking about music can be very different.
This creates problems for both the analysis of musical
features and the construction of databases. On the other
hand, a well-documented digitization offers interesting
opportunities for the study and spread of ‘endangered’
music. Here, some general problems regarding the digital
indexation of ethnic music are given, illustrated with a
method for describing pitch structure, comparing Western
standards with African music found in the digitization of
the archives of the Royal Museum of Central-Africa in
Tervuren (Brussels).
Keywords: ethnic music, Africa, pitch, archiving,
databases, cultural heritage
1. Introduction
At the moment most MIR applications are aimed
at popular and classical music in the Western tradition.
With the increasing use of digital search methods, users
are guided to music that is ‘searchable’. Music that
occupies a more marginal position risks to be omitted from
databases, or be much less accessible (e.g. because popular
search fields can not be filled out or similarity measures
with popular styles are very low). Thus the combination of
digitization and commercialization tends to push
‘vulnerable’ music into oblivion. However, if we develop
methods aimed at indexing non-Western music in an
appropriate way, the opposite might be true. The use of
music databases can bring people in contact with music
they would normally never have heard of, and thus give
the community a broader view on music.
Simply integrating more ethnic music, and non-
western music in general, into the existing databases and
indexes, can not entirely solve this problem. Musical
structures can be fundamentally different and also the
relative importance of structural elements in the musical
experiences can be different. In the west, music theory and
research traditionally focus on pitch, but in African music
a fixed tuning does not exist and relative pitch (higher-
lower) is more important than absolute pitch.
Another difficulty of integrating ethnic music in
digital music libraries lies in the construction of the meta-
data. In the description of field recordings, some
information that is ‘compulsory’ in the description of
Western music is lacking, while information that seems
irrelevant in the description of Western music is very
important. Names of composers are usually not known,
performers could be named, but some music is seen as
performed by ‘the community’ and the names of the
participants are not considered very important. On the
other hand, location and date of the recording are
important as documentation for researchers. Not all
recordings are well-documented, in some, even the most
basic information is lacking, this while the knowledge
about traditional music within the cultures is vanishing
under pressure of urbanization and Westernization. This
combination makes it harder and harder to identify and
label the music correctly.
Finally there is also a terminology problem.
Different local names for the same concept can exist, and
then different researchers can use different transcriptions
for these concepts. At this moment the American Folklore
Society and the American Folklife Center at the Library of
Congress are constructing an “Ethnographic Thesaurus”, a
comprehensive, controlled list of subject terms to be used
in describing ethnographic and ethnological research
collections (cf. http://www.etproject.org). But even a
standardized list can not solve all problems. We can give
the example of the ‘thumb piano’ (lamellophone). This
instrument type is referred to with diverse names (mbira,
likembe, sanza, kalimba, etc.). Someone looking for one of
these search terms should also be directed to pieces in
which one of the other denominations is used (Matthé et
al., 2006) To make it even more complicated, one
denomination does not necessarily point at a specific
subtype: size, material, number of pitches and tuning can
widely vary. Therefore the user should be given the means
to refine his search by looking for more specific
instrument characteristics, or for instruments with a similar
tuning.

Page 2

2. Digitization of the audio collection of the
Belgian Royal Museum of Central-Africa
The ideas described in this paper are a result of
work on the digitization of the ethnomusicological
archives of the Royal Museum of Central-Africa (RMCA)
in Tervuren (Brussels). With its 50,000 sound recordings,
with a total of 3,000 hours of music, the audio archive is
one of the biggest archives in world for the region of
Central-Africa, an area where the cultural heritage is
endangered by modernization and political instability
(Cornelis et al., 2005). The goal of the project is to
conserve the unique collection of sound recordings and to
make the archives accessible for a broad group of users.
This implies the digitization of different types of sound
recordings (wax cylinders, sonofil, vinyl recordings and
magnetic tapes) each with their own problems, the
construction of a database structure that is suitable to
describe the music (Matthé et al., 2005) and develop some
tools that allow a further analysis and classification.
Musicological texts describing the music and instruments
of different areas, sound examples and photos from the
archives and the collection of the museum are added as
background information. The progress can be followed on
the website http://music.africamuseum.be/.

An additional challenge is the differentiation
between user groups. We can broadly distinguish three
groups. The first and largest group is that of people who
are just interested in African music, but don’t have a good
knowledge of it. These people want to find e.g
‘drumming’, ‘trance music’ or ‘some song from Rwanda’,
which needs rather vague, general labeling. A second
group consists of people from within the culture. They
might have a very good knowledge of certain repertoires
and functions of the music, and thus ask very specific
questions like. music played by a specific performer,
music from one particular village, lyrics, genres,
instruments (in local terminology). The third group
consists of researchers, using the database in their study.
For them a fully indexed electronic database is an excellent
tool to study elements like the geographical spread of
certain instrument types or the relative importance of
certain musical structures in different regions.
Interesting applications for the first group could
be query by example, or searching by affective parameters.
Latter two groups can add valuable information to the
database. For this reason a forum is set up, when the
database will be fully available, users will be invited to
discuss on certain topics in this forum and hopefully this
will allow us to fill some gaps, come in contact with
researchers how to use the database and might give ideas
for the development of additional search tools and extend
the database with (copies of) recordings kept at other
institutions or by private persons.
3. Pitch structures in African Music
Models for pitch detection generally assume that
there are certain fixed elements like 12 chromatic tones, a
tonic and a dominant and octave relationships. These
concepts allow describing, comparing and classifying the
pitch structure in different pieces. When studying African
Music we can not hold these concepts as basis of our
study. This is a problem with which ethnomusicologists
already have to deal with for many years in making
transcriptions, usually solving it by adding additional
marks for deviations from regular tuning, sometimes
adding precise pitches.
The music of Central-Africa uses many different
tunings and scales. In fact, melody is closely connected to
the tones of Bantu languages. The music thus follows the
melody of speech. Therefore it is more important to have
high and low pitches as such than to have specific
harmonic relations between them. Furthermore in this
region instrumental sounds with a very broad, percussive
spectrum are preferred, which also reduces the importance
of ‘correct’ intervals. Nevertheless there instruments with
a fixed tuning like flutes, citers, (wooden) trumpets or
thumb piano’s allow to study the scales used.
In order to study the scales and tunings, a system
was developed that allows representing scales without
reference to Western notes or scales. Instead of an a priori
thinking in pitch categories, we chose to start from a
continuous representation of pitch. In a first stage the
music is analyzed by a melody extractor (De Mulder et al.,
2004). This system was originally devised for vocal
queries, and is thus optimized for monophonic music and
the normal voice range. Yet, testing the model on different
types of music reveals that it can give a very good image
of the distribution of pitches in music with a more complex
texture. The melody extractor currently used gives a
frequency for every time frame of 10 ms. In order to give
an image of the scale, we transform these values to cent
values (taking the low A, 55Hz as 0 cents) and then make
a frequency table by cent values between 1 and 6000.
The method is illustrated bye the analysis of a
song “Ingendo y'inka” with ikembe (thumb piano)
accompaniment recorded on 14 Jan 1973 by Jos
Gansemans in the village Karengera, Cyangugu province,
Rwanda. The singing is mainly in parlando style,
following the pitches of the instrument. The text describes
the elegance of the local cows. If we extract the most
prominent peaks (table 1) we see more or less equidistant
intervals, with an average size of 236 cents. The intervals
betweens successive tones are somewhat larger than a
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies
are not made or distributed for profit or commercial advantage and that
copies bear this notice and the full citation on the first page.
© 2006 University of Victoria

Page 3

whole tone but the size is rather variable (202-286 cents),
but the octave relationships are generally quite well
preserved, which is also the case for the fifths. However,
we see that the highest octave is much (60 cents) too small,
while the lowest is 25 cents too wide. This causes a
problem when we want to reduce the representation to one
single octave (see Figure 1). Three tones keep a very
narrow peak, indicating near perfect octave relationships,
but the middle peak is clearly widened and in the left peak,
the one corresponding to the extreme pitches we clearly
see a combination of three smaller peaks.

Table 1: analysis of the most prominent peaks in “Ingendo
y'inka”, left the pitch in cents above 55 Hz, next the distance
in cents between successive pitches and in the right column
the size of the octave relationships.
pitch interval
1042
1278
1542
1744
1981
2267
2477
2727
2958
3179
3407
octave
236
264
202
237
286
210
250
231
221
228
1225
1199
1185
1214
1198
1140






Figure 1: pitch analysis of “Ingendo y'inka”, reduced to one
octave, on the x-axis the pitch in cents (0 = a), on the y-axis
the smoothed number of occurrences for every 1 cent.

This example illustrates the difference between
African and Western tone systems very well. The Western
system uses a chromatic scale in which interval sizes are
multipliers of 100 cents and there is a strict adherence to a
1200 cents octave. In African music there is no standard
tuning, in this case there is a more or less equidistant
pentatonic scale, but many different scale types are found.
The octave exists, but it’s role is not absolute. In fact it is
not unusual that players deliberate tune some of the
octaves to wide or to narrow in order to create an extra
rich timbre.
The graphic representation can be the basis of
comparison of tone systems. This can allow to use the
system for musicological applications (e.g. to map the
spread of certain tone systems), but also to let a user of the
database look for similar music.
4. Conclusions
Constructing MIR applications that can deal with
the world’s music is difficult, but necessary to protect the
world’s cultural heritage. It is important to bring
knowledge about the music together and make it accessible
to a broad audience. The fundamentally different use of
pitch in African music compared to Western music
illustrates the difficulty of simply applying existing
methods to other bodies of music in order to index it. In
this phase of the project of digitizing the archives of the
RMCA we started to develop some methods that should be
able to deal with all types of musical content. This was
illustrated by a method of extracting the scale of a musical
piece and representing it in a graphic notation that can
serve as basis of comparison with scales used in other
pieces, regardless of their cultural background and tone
system
References
Cornelis, O., De Caluwe, R., De Tré, G., Hallez, A.,
Leman, M., Matthé, T., Moelants, D., & Gansemans, J.
(2005). Digitisation of the Ethnomusicological Sound
Archive of the Royal Museum for Central Africa
(Belgium). Iasa journal, 26, 35-43.

De Mulder, T., Martens, J.P., Lesaffre, M., Leman, M., De
Baets, B., & De Meyer, H. (2004). Recent improvements
of an auditory model based front-end for the transcription
of vocal queries, in: Proceedings of the IEEE International
Conference on Acoustics, Speech and Signal Processing
(pp. 257-260). Montreal.
Matthé, T., De Tré, G., Hallez, A., De Caluwe, R., Leman,
M., Cornelis, O., Moelants, D., & Gansemans, J. (2005). A
Framework for Flexible Querying and Mining of Musical
Audio Archives. In Proceedings of the 16th International
Conference on Database and Expert Systems Applications
(pp. 1041-1045). Copenhagen.
Matthé, T., De Caluwe, R., De Tré, G., Hallez, A.
Verstraete, J., Leman, M., Cornelis, O. Moelants, D., &.
Gansemans, J. (2006). Similarity Between Multi-valued
Thesaurus Attributes: Theory and Application in
Multimedia Systems. Lecture Notes in Computer Science
4027, 331–342
0
20
40
60
80
100
120
140
0
100 200 300 400 500 600 700 800 900 1000 1100 1200