Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14.

Berlin, Germany 2014

THE VIOLIN ONTOLOGY

Massimiliano Zanoni, Francesco Setragno, Augusto Sarti1
1 Dipartimento di Elettronica e Informazione - Politecnico di Milano University
Correspondence should be addressed to:[email protected]

Abstract: Bowed musical instruments have been the subject of algorithms for classification/recognition applications; translating
scientific investigations for centuries. Yet, the physical phenomena scientific descriptors/features in more intuitive terms for the com-
that are behind their timbral quality are still far from being fully munitiy of instrument makers and that of musicians; and more.
understood. This is one of the reasons why the art of violin making Several studies from different disciplines are already present in
is still so strongly tied to tradition. This manuscript describes early the literature and the presence of several forums for instrument
results in a study of the relations that exist between timbral and acoustic makers on the Web stresses the needs of sharing information in
characteristics of such instruments and their high-level descriptors. In this particular area. However, to the best of our knowledge,
particular, we propose a suitable ontology for a timbral characterization an expressive representation of the knowledge and an excellent
of violins, where every resource is connected and provided with formally organization of information, able to provide powerful tools for
defined semantics. Semantic web technologies have taught us how searching and sharing, are still needed.
ontologies can become a powerful tool for gathering and managing Numerous data knowledge paradigms have been developed for
knowledge in specific areas of interest, where resources are connected a variety of applications, often based on XML [1, 2]. Such
and described with formally defined semantics. This, in fact, represents paradigms, however, are not expressive enough for defining seman-
a crucial step for building applications that reason over Web data. tic relationships and constraints over information [3]. The lack of
In this paper we present an ontology for knowledge representation of semantics, in fact, makes interoperability difficult, and often hinders
violins, as part of a wider ontology of bowed instruments. With this the automatic reasoning over data [4]. For this reason, in order
ontology we capture timbral and acoustic aspects of violins as well as to capture, formalize and organize the knowledge we need to take
violin making and properties of the materials used for their production. advantage of Semantic Web methodologies [5, 6].
We collected and organized semantic descriptors used by numerous
violin makers (from natural language) to describe sound proprieties The development of tools for Semantic Web has become partic-
of musical instruments. We also developed an initial model of the ularly popular in past few years. Semantic Web supports the
relation between semantic descriptors and low-level audio features. collection and the linkage of structured formalization of knowledge.
The ontology that we present in this study formalizes the semantics In this context, ontologies are a very useful tool for expressing
of the high-level descriptors and investigates the relation with low- knowledge in a specific area of interest. In ontologies, resources
level features. The terminology has been collected through a series are equipped with formally defined semantics. Each resource can
of interviews with violin makers in the city of Cremona (Italy), world be also connected to others in order to express semantic relations
heritage site for the practice of violin makers. Through listening tests in such a way that it becomes possible to build applications that
and a feature extraction we study the correlation between high-level reason over Web data. More recently, Semantic Web has been
descriptors and objective properties of sound. successfully applied to the area of music information management
[7, 8, 9]. One of the main contributions in this area is the OMRAS2
project1 , which aims at building a framework to share information
about music (performances, productions and so on) in the Semantic
1. I NTRODUCTION Web format. For this purpose several ontologies have been created.
The art of violin making is the result of five centuries of evolution, Some examples are the Music Ontology [8], the Audio Effect
which started in Cremona, Italy, with its patriarch Antonio Amati, Ontology [10] and the Audio Feature Ontology2 . For all these
followed by the renowned families of Stradivari and Guarnieri. projects, the use of Semantic Web turned out to be a better choice
Today Cremona is still the heart of this tradition, as confirmed by compared to other knowledge management methods [4].
UNESCO, which in 2012 made Cremona a World Heritage Site In order to investigate, collect, and conceptualize the traditional
for the art of lutherie. Cremona is currently home to over 150 and modern knowledge concerning violin instruments, in this study
violin makers, and thousands more have studied there, all following
techniques that have been passed forward from generations to we propose the Violin Ontology3 . This Ontology is designed to
generations for centuries. As the practice of violin making evolved capture the knowledge-related different aspect of these instruments,
and consolidated, violins have been a constant subject of scientific from the materials to the construction methods, to the description
investigations. Yet, the physical phenomena that are behind their of the sound proprieties. The Violin Ontology is part of the Bowed
timbral quality are still far from being fully understood. Because Instrument Ontology, which is our next goal.
of this inherent difficulty in capturing what physical phenomena The collection of information related to materials and to the con-
make such instruments sound the way they do, violin making has struction methods is a very complex and time-consuming process.
remained an art that that is solidly tied to tradition. In fact, although there exist some general techniques that violin
In the past few decades, however, there has been a renewed frenzy makers usually refer to, they tend to build instruments according
in research, aimed at pushing the boundaries of our physical under- to their own personal styles. Moreover, for historical instruments,
standing of the quality of violin tone. This recently prompted the information is often based on old and often very limited written
city of Cremona to start new research projects with the Politecnico descriptions. Consequently additional investigations based on
di Milano (for aspects of musical acoustics) and the University of advance material analysis (x-ray, micro- and UV imaging, spec-
Pavia (for aspects of material analysis), aimed at exploring new trography, and chemical analysis) are often in order, when it comes
directions in contemporary lutherie. Among the many goals of to tracing the construction history of such instruments. For these
such projects is that of understanding the links that exist between and other reasons, in this manuscript we only present a preliminary
objective and semantic descriptors related to such instruments. The version of this Section of the ontology.
former are geometric, vibro-acoustic, acoustic and timbral features; The characterization of the violin sound should consider timbral and
physical and chemical properties of materials, etc. The latter acoustic proprieties as well as the perceptual aspects of the sound,
are the terms of natural language that are customarily used for where perceptual aspects refer to how the sound is perceived and
describing qualities of the instrument. Identifying such relations
1 http://www.omras2.org/
would allow us to achieve multiple goals: improving the quality
2 http://motools.sourceforge.net/doc/audio
(breadth and span) of information gathering through interviews features.html
and questionnaires; improving the effectiveness of training-based 3 http://purl.org/net/bowedinsturmentsontology
Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14. Berlin, Germany 2014

how it is described and what the relation between sound perception

and description is. Bow: Instrument
Bow: Sound
Descriptors
As far as sound description is concerned, people use terms from Maker
the natural language, such as Sweet or Bright. These terms are not
directly related to mathematical formulas and they are characterized Bow:Sound_description Bow:Correlation
Bow:Made_by
by a high-level of abstraction (Semantic Descriptors or High-
Level Features). Try to formalize the semantics of these terms Bow: Bow:Model Bow:
(conceptualization) is a tough task because of the subjectivity in Instruments Materials
their use and the lack of a precise and shared meaning. Because of
that, in the past decades, several studies have been presented in the Bow:Composed_by Bow:Made_of
literature [11, 12, 13, 14]. The main purpose of these studies is to
collect and arrange descriptors in a multi-dimensional perceptual Bow:
space where distance between concepts can be defined. Similar Instrument Making Instrument
Ontology Section parts
approaches have been adopted also for the semantic description of
the violin timbre [15, 16, 17]). Nevertheless, this approach showed
some limitations: the positions of the terms in the space tend to have
a large variance because their use is subjective and depends on the Figure 1: Overall scheme of the Violin Ontology. Different colors
context. Moreover, these studies mainly focus on the organization represent the different sections of the ontology.
of terms in spaces. They do not provide any formal definition of
their semantics in the context of application, which is mandatory in
the characterization of the knowledge. In [18] the authors highlight
the potential of semantic audio description. However, they present 2.1. Ontologies
the limitation of the used description paradigms. They stressed the An ontology is a semantic formalization of a domain of interest.
importance of a superior representation able to express the semantic In other words, it is a representation for expressing knowledge in
of the descriptors and the complex relation among. We assert that a certain area in a formal (machine-readable) way. In particular, it
Semantic Web technologies represent a superior choice. provides and shares conceptualization of the world of interest by
Though our way of describing sounds is based on subjective characterizing entities in terms of their meaning and relationships.
Semantic Descriptors, there exist a strong connection between Ontologies are the core of the world of Semantic Web and their
sound description, sound perception and physics. Our brain, main components are:
in fact, processes stimuli from the auditory system in order to • Classes: they represent categories or concepts. A class can
formulate a proper description. It is very hard to understand what have subclasses that have a more specific meaning.
aspects of the sound influence our perception [19]. However, • Individuals: individuals are instances of classes and represent
several studies have been to address such issues. In particular, in concrete objects. An individual can belong to more than one
Music Information Retrieval, an interdisciplinary research area that class.
studies the extraction of useful information from musical content, • Properties: a property is a relation defined between two
the relation between objective timbral and acoustic proprieties classes and applied on their individuals.
(Low-Level Features - LLFs) and subjective semantic descriptors
(High-Level Features - HLFs) is studied by means of feature Classes and properties can be organized in hierarchies with some
analysis methodologies [19, 20, 21]. Also in the context of bowed constraints over their use. Several languages for ontologies exist.
instruments, some work has been presented in the literature [22, 23]. In this study, we adopt the well-known Web Ontology Language
(OWL) [25].
The conceptualization of LLFs and HLFs are included in our
ontology in order to provide the violin sound characterization. 2.2. OWL
As far as HLFs is concerned, we collected a set of Semantic Web Ontology Language is a very popular language endorsed by the
Descriptors through several interviews to violin makers. In order
to investigate the relation between LLFs and HLFs, we recorded World Wide Web Consortium (W3C). Its latest version is OWL24 .
musical performance from eleven violins. We then extracted a set OWL2 is based on a Description Logic, a family of formalisms used
of Low-Level Features from the recordings in order to provide a to represent knowledge in a domain. Description Logic is what
low-level audio characterization for each violin. The recordings makes ontologies differ from taxonomies. Taxonomies are, in fact, a
were then annotated by five expert violin makers using the collected list of terms and definitions. Whereas, Description Logic allows the
terms. The correlation between Low-Level Features and semantic construction of intelligent applications (knowledge based systems)
annotations were then explored through two statistical correlation able to infer implicit consequences from the explicit information
indexes: Distance Correlation Index and RreliefF algorithm. that has been represented. This is done by means of the reasoner.
For better understand the reasoner we provide here an example.
2. A B RIEF OVERVIEW OF S EMANTIC W EB Given the class ItalianViolin that collects violins built in Italy and an
T ECHNOLOGIES its individual Stradivari Soil built in the year 1714; given the class
ViolinMaker and an its individual Antonio Stradivari. Through the
Semantic Web [24] aims at bringing intelligence to the Web by reasoner we can in infer that Antonio Stradivari lived Italy at that
assigning the Web resources a formally defined semantics and time. We could not infer such an assumptions without building an
forming a very large web of data that can be interpreted by ontology.
the machine. In order to represent information in a machine
interpretable format, a proper language is needed. The Resource
Description Framework (RDF) is a data model for describing
3. T HE V IOLIN O NTOLOGY
statements about Web resources using subject, predicate, object The overall ontology is composed of three sections: Instrument
triples. When combined, object triples form a RDF graph. If the Maker Ontology Section, Instrument Making Ontology Section,
elements in the RDF graph are identified by a uniform resource Sound Description Ontology Section. Figure 1 shows the overall
identifiers (URI) they build the Web of Linked Data, which is scheme of the Violin Ontology.
interlinked and globally distributed. In order to represent the Currently, the Instrument Makers Ontology Section is composed
knowledge on the entities described in the graph and, hence, of a unique class Bow:Instrument Maker that collects information
to enable reasoning over linked data, a further, more complex about the makers of specific instruments, such as historical period
representation is needed. The one adopted in this study is the and used techniques. The Instrument Making Ontology Section
ontology. aims to capture the knowledge related to the violin making process

4 http://www.w3.org/TR/owl2-syntax/
Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14. Berlin, Germany 2014

Brilliat
Bow: Paint

Bow: Wood Bow: Glue

rdfs:subClassOf
Bow:Similarity
Bow:Belong Timbre
rdfs:subClassOf rdfs:subClassOf
Bright
Bow:Belong
Bow:
Definition
Materials
Bow:Opposition

Dark
Figure 2: Ontology scheme for the Violin Making Ontology
section.

Figure 4: Example of ontology usage for the description of the term

Bow:
Bright
Instruments

rdfs:subClassOf rdfs:subClassOf

rdfs:subClassOf
ety. As previously mentioned, the quality of sound is strictly related
Bow: Bow: to the intrinsic composition of the materials and to the characteris-
ItalianViolin Bow: FrenchViolin tics of the violin parts. For this reason, we link Bow: Sound De-
GermanViolin scriptors to the class Bow: Materials through the Bow:Correlation
propriety and to Bow: Instrument parts through the Bow:Model
propriety.
Figure 3: Ontology Scheme for the instrument concept. In order to represent the knowledge related to both Low- and High-
Level Features, Bow: Sound Descriptors class has the Audio: Low-
Level Features and Bow: High-Level Features subclasses. The
overall scheme of this section is depicted in figure 5.
and includes information on instrument parts and material. Finally, As far as HLFs are concerned, there is a common vocabulary that
the Sound Description Ontology Section conceptualize the timbral violin makers use to describe the sound proprieties of instruments.
and acoustic proprieties of the instrument sound. Terms in the vocabulary express different shades of sounds. Our
3.1. Instrument Making Ontology Section intent is to find the most used and relevant terms that are the terms
whose meaning (and self-confidence) reached a consensus among
The Instrument Making Ontology Section is composed of three makers. We populate the vocabulary of the semantic descriptors
classes (figure 1): Bow:Instruments, Bow:Instrument parts and through a series of interviews with 20 violin makers in Cremona and
Bow: Materials. Bow:Instruments class provides the character- Milano, Italy. Starting from a seminal vocabulary populated with
ization of the violins instrument (figure 3). Along the history, terms retrieved by other studies in the literature, in each interview
different making process procedures has been developed. It is we asked to propose further terms to add. After that, the subjects
possible to narrow three main schools: Italian, German and French. were asked to outline the meaning of each term, their usage, the
For this reason, the Bow:Instruments concept has three subclasses: relations between them in terms of similarity and their nuance
Bow:ItalianViolin, Bow:GermanViolin, Bow:FrenchViolin. (positive or negative, when possible).
The Bow:Instrument Maker class is linked through the propriety Trying to generalize the human description of sound is a very hard
Bow:Made by to the Bow:Instruments class, since a violin maker task. This is due to different reasons: different reactions to physical
produces and repair violins. Instruments are composed of numerous stimuli, different musical backgrounds, different tastes and different
components, each one with specific characteristics that has been way to describe the sounds. For this reason, terms which semantic
refined along the centuries. For this reason, the Bow:Instruments did not converge to an agreed definition, were discarded. The
class is linked to the Bow:Instrument parts class through the residuals compose the set of individuals of the Bow: High-Level
propriety Bow:Composed by. The Bow:Instrument parts class Features concept and the nuances (positive, negative, neutral) are
defines a conceptualization of the description of single component attributes.
that compose the instruments. Each part of the instruments is made
of specific materials, whose choice has been a process refined along The terms might be related to each others in terms of similarity or
the centuries. The Bow:Made of propriety describes the relation opposition through the two proprieties: Bow: Similarity and Bow:
between the Bow:Instrument parts class and the Bow: Materials. Opposition. This two proprieties have been declared symmetric,
so that if a term A is linked to the term B, then B will be linked
Bow:Materials (figure 2) represents the generic concept of material to A with the same property. The vocabulary is listed in Table 2.
and has three subclasses: Bow:Wood, Bow:Paint, Bow:Glue, which Terms in the same row and different columns have been annotated
are the predominant materials used in violin liuthery. The class as opposite. Whereas terms in same row and same column are
Bow:Wood, which individuals are the types of wood, such as maple annotated as similar.
and spruce, represents the description of the typically used type of
wood. Internal and external parts of the instruments are generally Figure 4 shows an example of the formalization of the term Bright
painted using a mixture of pigments, copals, white coat and other using the Violin Ontology. In the example, Dark and Bright are
essences. The class Bow:Paint conceptualize the paint mixtures. opposite, whereas Bright and Brilliant are similar.
Various components of the instruments are typically joint using Terms in the vocabulary tend to capture specific aspects of the sound
different types of glues. For this reason, we also added the class (timbre, propagation, instrument response, overall sensations). This
Bow:Glue. effect is formalized in the Ontology by introducing the class Bow:
SemanticContext and the propriety Bow: Belong. Each term can
3.2. Sound Description Ontology Section belong to one or more areas. Terms Timbre and Definition are two
In the Violin Ontology sound proprieties are described through semantic context (figure 4). The semantic contexts included in our
the Bow: Sound Descriptors concept, which is linked to the Ontology are:
Bow:Instruments class through the Bow:Sound description propri- • Timbre: this class includes terms that are mainly related to the
Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14. Berlin, Germany 2014

Audio: Low-Level Terms (HLF) Bowed Instrument

Features Interviews
Ontology
Terms (HLF)
rdfs:subClassOf
LLF-HLF
Linstening correlation
test Annotations
Bow: Semantic
Bow: Sound
Bow:hasCorrelation Context Recordings
Correlation
Descriptors Instrument
study
Recordings Recordings
rdfs:subClassOf Bow:Belong Low-level feature LLFs
extraction

Bow: High-Level
Features
Bow:Similarity Bow:Opposition Figure 6: Overall procedure for the Low- and High-Level Features
correlation study.

Figure 5: Ontology scheme for the sound description part.

Dark Bright, Brilliant
Audio:Low-Level Features class refers to the Audio Features Strident, Metallic Warm
Ontology developed in the OMRAS project. Harsh Sweet
Unfocused Focused
Hard Soft
1: Root Mean Sqaure energy 21: Zero Crossing Rate
Hoarse Clean
2: Spectral Centroid 22: Spectral Centroid Balanced, Even Unbalanced, Uneven
3: Spectral Skewness 23: Spectral Kurtosis Weak Strong
4: Spectral Roll-off 24: Spectral Flux Thin Full
5: Spectral Irregularity 25: Harmonic Ratio Deep
6: Attack Slope 26: Attack Time Nasal
7-20: MFCC Rounded
Big, Loud Quiet
Responsive, Sensitive Inert
Table 1: List of Low-Level Features used for the correlation study Open Closed, Covered, Boxy
Projecting
Edgy
Sparkly
spectral shape of the sound.
Colourful
• Definition: this area groups terms that describe how the sound
is defined and focused (e.g. blurred, sharp).
• Propagation: it is related to the irradiation of the sound in the Table 2: This table shows the relations among descriptors. Terms
space, including information about volume and directionality. in the same row but in different columns are considered as opposite.
• Ease of play: this area represents the sensations felts in Terms in the same row and in the same columns are considered as
playing the instrument. similar.
As far as the LLFs are concerned, we exploit the Audio Feature
Ontology 5 that is one of the ontologies developed within the OM-
RAS2 project6 . It is a representation of the knowledge concerning testers had a disposal all the recordings for each instrument. Each
the audio feature characterization of sounds. We include the Audio descriptors were presented along with its opposite. For those that
Feature Ontology as part of the Violin Ontology. no opposite term exists in the ontology, they were presented with
As introduced so far, there is a relation between Low-Level and their nominal opposites (e.g. nasal-not nasal). The testers were
High-Level descriptors. In order to conceptualize this relation the asked to assign a graded annotation ranging from −1 to 1. −1
classes (Audio:Low-Level Features) and Bow:High-Level Features assigns a strong preference to the first term, whereas 1 assigns a
are linked by the propriety Bow:hasCorrelation. This relation will strong preference to the second term.
be explored in Section 4. As far as the LLFs are concerned we extracted a set of audio
cues able to capture timbral, harmonic and temporal information.
4. AUTOMATIC LLF-HLF RELATION INVESTIGATION Feature extraction has been performed on the music pieces and on
the major scale. The used set of features are listed in Table 1.
In order to investigate the relation between LLFs and HLFs we ex-
ploit feature analysis and semantic description modeling techniques Exploiting the low- and high-level sound characterization for each
[19]. The overall procedure is shown in figure 6. It is first needed violin, we investigate the correlation between LLFs and HLFs
to collect both low- and high-level audio descriptors for a set of through two well-known correlation methods. The first method
instruments. We collected recordings for eleven modern violins is the Distance Correlation Index [26]. Given Xi the vector that
played by a professional musician. The recordings were realized collects all the values of the i − th LLF within the corpus and Y j the
with a high-quality recording system in a semi-anechoic room. In vector that collects all the values of the j − th HLF, the Distance
order to capture different nuances of the instrument sound and in Correlation Index is defined as:
order to attenuate the impact of possible noises, interfering signals,
and performance-dependent features, the musician was asked to dCov(Xi ,Y j )
play single notes along the neck for each string, a major scale and D= p , (1)
dVar(Xi )dVar(Y j )
two short compositions.
As far as the HLFs are concerned, we collected the annotations by where dCov(Xi ,Y j ) is the distance covariance that is defined as
asking to five professional violin makers to provide a description
for each violin using the semantic descriptors in the ontology. The v
L L
u
5 http://motools.sourceforge.net/doc/audio
u1
features.html dCov(Xi ,Y j ) = t 2 ∑ ∑ ||Xl1 − Xl2 ||||Yl1 −Yl2 ||, (2)
6 http://www.omras2.org/ L l1=1 l2=1
Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14. Berlin, Germany 2014

Distance Correlation Index

rms
zcr 0.7
spectral centroid
spectral spread
spectral skewness
spectral kurtosis 0.6
rolloff
spectral flux
harmonic ratio
spectral Irregularity 0.5
mfcc1
mfcc2
mfcc3 0.4
mfcc4
mfcc5
mfcc6
mfcc7 0.3
mfcc8
mfcc9
mfcc10
mfcc11 0.2
mfcc12
mfcc13
attack time
attack slope 0.1
dark strident harsh unfocused hard hoarse unbalanced weak thin not deep nasal rounded
bright warm sweet focused soft clean balanaced strong full deep not nasal not rounded

Figure 7: Distance Correlation Index computed on Low-Level Features extracted from 11 violin recordings and semantic descriptors used
for the annotations.

L is the number of observations and dVar is the distance variance RreliefF ranking : dark/bright,brilliant
which is given by
0.25
dVar(Xi ) = dCov(Xi , Xi ). (3)
spectralKurtosis

spectralSpread

0.2
The Distance Correlation Index ranges from 0 to 1, where 0 means
spectralSkewness

no correlation (independence).
spectralCentroid

0.15

spectralIrregularity
mfcc13

The second method is a regression-based analysis using the RreliefF

mfcc3

harmonicRatio
attackTime

algorithm [27]. Given X the observations-by-features matrix and Y j

Weight

0.1

spectralFlux

attackSlope
the collection of the j − th HLF associated to each observation, the
rolloff

mfcc10
mfcc7

algorithm performs a regression and returns a weighted ranking of

mfcc12

mfcc11
0.05
mfcc5

mfcc9
mfcc4

mfcc8

mfcc1
mfcc6
mfcc2
the Low-Level Features. The higher is the weight for Xi , the higher

rms
zcr

is the correlation between Xi and the semantic descriptor Y j . 0

The limited space of the paper do not allow us to show all the
results of the study. For this reason, we present here only some −0.05

representative examples. Figure 7 shows the Distance Correlation

Index for the LLFs and a subset of the semantic descriptors. It can −0.1

be noticed that each semantic term has a strong correlation with

the only a subset of audio cues. For example, Dark and Bright 0 5 10 15 20 25
are intuitively mainly related to the predominance of low and high
frequencies. In fact, their definition collected in our ontology are: Figure 8: Output of the RreliefF algorithm for the pair
• Dark: A dark sound has a predominance of low frequencies Dark/Bright,Brilliant. Feature are ranked based on the value of
and it is slightly blurred. It is opposed to the concept of bright the correlation. The higher is the histogram, the higher is the
and sharp. The exaggeration is Unfocused, Covered. Full of correlation.
harmonics. This feature is related to the wood. Low notes are
enhanced.
• Bright: This term is often used both in timbre and definition
contexts. It indicates a clear, direct and defined sound, with figure 7 shows how the terms Thin/Full and Rounded/Non Rounded
a predominance of high frequencies. It is a quality that presents a similar correlation with Spectral Centroid, Spectral
characterizes violin from other instruments (viola, cello). Spread, Spectral Skewness, Spectral Kurtosis, Spectral Rolloff,
This effect is confirmed by the Distance Correlation Index analysis. mfcc10 and mfcc13 features.
The figure 7, indeed, shows that Dark/Bright are mainly correlated
with features that capture the spectral shape, such as Spectral 5. C ONCLUSION AND F UTURE W ORKS
Spread, Spectral Skewness and Spectral Kurtosis. Particularly In this study, we presented a Semantic Web Ontology covering the
important is also the high correlation with the mfcc3 and the mfcc13 domain of the knowledge about violin instruments. The Violin
features that are respectively related to low and high frequencies. Ontology aims at collecting and sharing knowledge concerning
The results for the Dark/Bright terms are confirmed by the RreliefF materials, violin parts, as well as Low-Level and High-Level audio
analysis. In figure 8 it can be noticed that mfcc13, mfcc3, Spectral features. In particular, the ontology allows to conceptualize the
Spread, Spectral Skewness and Spectra Kurtosis are at the top of relation between these elements, which are naturally correlated.
ranking. At this purpose in this study we collected the vocabulary and the
Intuitively, semantic descriptors that share a semantic similarity, semantic definition of descriptors used to describe instrument sound
should also presents some similarity in the distribution of the proprieties through a set of surveys. We then exploited the recording
correlation values with the LLFs. With the Distance Correlation of eleven violins to study the relation with Low-Level features.
Index analysis it is possible to capture this effect. For example, Feature works provide for extensively examine the dependencies
Proceedings of the 9th Conference on Interdisciplinary Musicology – CIM14. Berlin, Germany 2014

between material proprieties, violin parts proprieties and the sounds International Computer Music Conference (ICMC 1999).
proprieties. Beijing, China, 1999.
[19] M. Zanoni, D. Ciminieri, A. Sarti, and S. Tubaro: Searching
R EFERENCES for dominant high-level features for music information
[1] D. Beenham, P. Schmidt, , and G. Sylvester-Bradley: XML retrieval. In Proceedings of the 20th European Signal
based dictionaries for MXF/AAF applications. Technical re- Processing Conference (EUSIPCO). 2012.
port, Sony Broadcast and Professional Research Laboratories, [20] Y. E. Kim, E. M. Schmidt, R. Migneco, O. G. Morton,
UK, 2000. P. Richardson, J. Scott, J. A. Speck, and D. Turnbull: Emotion
[2] J. Smith and P. Schirling: Metadata standards roundup. In Recognition: a State of the Art Review. In In Proceedings
MultiMedia, IEEE, volume 13(2):84–88, 2006. of the 11th International Society for Music Information and
Retrieval Conference (ISMIR2010). 2010.
[3] M. Klein, D. Fensel, F. van Harmelen, and I. Horrocks:
The relation between ontologies and XML schemas. In [21] D. Turnbull, L. Barrington, D. Torres, and G. Lanckriet:
LINKÖPING ELECTRONIC ARTICLES IN COMPUTER Towards Musical Query-by-semantic-description Using the
AND INFORMATION SCIENCE, page 2001. 2001. CAL500 Data Set. In Proceedings of the 30th Annual
International ACM SIGIR Conference on Research and
[4] G. Fazekas and M. Sandler: Novel methods in information Development in Information Retrieval, SIGIR ’07, pages 439–
management for advanced audio workflows. In In proceedings 446. ACM, New York, NY, USA, 2007.
of 12th International Conference on Digital Audio Effects.
Como, Italy, 2009. [22] J. Štěpánek and Z. Otčenášek: Acoustical correlates of the
main features of violin timbre perception. In Actes du
[5] G. Fazekas, Y. Raimond, K. Jacobson, and M. Sandler: An Colloque interdisciplinaire de musicologie (CIM05), pages
Overview of Semantic Web Activities in the OMRAS2 Project. 10–12. Québec, Canada, 2005.
In Journal of New Music Research, volume 39(4):295–311,
2010. [23] E. Lukasik: Towards timbre-driven semantic retrieval of
violins. In Proceedings of the Fifth International Conference
[6] P. Borst, H. Akkermans, and J. Top: Engineering Ontolo- on Intelligent Systems Design and Applications (ISDA 2005).
gies. In International Journal of Human-Computer Studies, 2005.
volume 46(2-3):365–406, 1997.
[24] T. Berners-Lee, J. Hendler, and O. Lassila: The Semantic Web.
[7] G. Fazekas and M. Sandler: The Studio Ontology Framework. In Scientific American, 2001.
In Proceedings of the International Society for Music
Information Retrieval Conference (ISMIR2011). 2011. [25] G. Antoniou and F. van Harmelen: Web Ontology Language:
OWL. In S. Staab and R. Studer (eds.), Handbook on
[8] Y. Raimond, S. Abdallah, M. Sandler, and F. Giasson: Ontologies, International Handbooks on Information Systems,
The Music Ontology. In Proceedings of the International pages 67–92. Springer Berlin Heidelberg, 2004.
Conference on Music Information Retrieval (ISMIR 2007).
2007. [26] G. Székely, M. Rizzo, and N. Bakirov: Measuring and
testing independence by correlation of distances. In Annals
[9] K. Barkati, A. Bonardi, A. Vincent, and F. Rousseaux: of Statistics, 2007.
GAMELAN: A Knowledge Management Approach for Digital
Audio Production Workflows. In Proceedings of Artificial [27] M. Robnik-Ŝikonja and I. Kononeno: An adaptation of relief
Intelligence for Knowledge Management Workshop (A14KM) for attribute estimation in regression. In Proceedings of
of ECAI. 2012. the fourteenth International Conference on Machine Learning
(ICML1997). 1997.
[10] T. Wilmering, G. Fazekas, and M. B. Sandler: The Audio
Effects Ontology. In Proceedings of the 14th International
Society for Music Information Retrieval Conference. 2013.
[11] A. Zacharakis, K. Pastiadis, G. Papadelis, and J. D. Reiss:
An Investigation of Musical Timbre: Uncovering Salient
Semantic Descriptors and Perceptual Dimensions. In
Proceedings of the 12th International Society for Music
Information Retrieval Conference (ISMIR2011). 2011.
[12] J.M.Grey: Multidimensional perceptual scaling of musical
timbres. In Journal of the Acoustical Society of America,
1977.
[13] G. Bismarck: Timbre of steady tones: A factorial investigation
of its verbal attributes. In Acustica, 1974.
[14] A. C. Disley, D. M. Howard, , and A. D. Hunt: Timbral
description of musical instruments. In Proceedings of the 9th
International Conference of Music Perception and Cognition.
2006.
[15] J. Štěpánek: Musical Sound Timbre: Verbal Description and
Dimensions. In Proceedings of the 9th Int. Conference on
Digital Audio Effects (DAFx-06), Montreal, Canada. 2006.
[16] C. Fritz, A. F. Blackwell, I. Cross, B. C. J. Moore,
and J. Woodhouse: Investigating English Violin Timbre
Descriptors. In Proceedings of the 10th International
Conference on Music Perception & Cognition (ICMPC 10).
2008.
[17] C. Saitis, B. L. Giordano, C. Fritz, and G. P. Scavone:
Perceptual evaluation of violins: a quantitative analysis of
preference judgments by experienced players. In The Journal
of the Acoustical Society of America, volume 132(6):4002–12,
2012.
[18] P. Herrera and X. Serra: Audio descriptors and descriptor
schemes in the context of mpeg-7. In Proceedings of the