Comparative Analysis of A Cappella Flamenco Cantespersonal.us.es/dbanez/papers/cofla-journal.pdf ·...

Mora et al. Comparative analysis of a cappella flamenco cantes

1/28

Comparative Analysis of A Cappella Flamenco Cantes

Joaquín Mora1, Francisco Gómez

2, Emilia Gómez

3,4, Francisco Escobar-Borrego

5,

José Miguel Díaz-Báñez6, Juan Jesús Cabrera

6, Jordi Bonada

3

1Department of Evolutive and Educational Psychology, University of Seville, Spain

2Applied Mathematics Department, School of Computer Science,

Polytechnic University of Madrid, Spain

3Music Technology Group, Universitat Pompeu Fabra, Barcelona, Spain

4Department of Sonology, Escola Superior de Música de Catalunya,

Barcelona, Spain

5Department of Audiovisual Communication, Publicity and Literature,

University of Seville, Spain

6 Department of Applied Mathematics, School of Engineering,

University of Seville, Spain

Address correspondence to:

Francisco Gómez Martín

Departamento de Matemática Aplicada

Escuela Universitaria de Informática

Ctra. de Valencia, Km. 7. 28031 - MADRID

FAX: +34913367877


2/28

Abstract

The aim of this work is to compare different approaches to melodic and tonal analysis of flamenco a cappella singing styles. The ultimate goal would be to perform a multidisciplinary study to compare different versions of the same style, and, as a consequence, to clarify the roots of styles and their evolution. As a first step, we have gathered a representative corpus of music, which in fact is a significantly difficult task due to the variety of media and quality of existing flamenco recordings, and also due to how scant and scattered those recordings are. We focus on specific singing styles, namely, deblas, martinetes and tonás. Once the entire corpus is selected, a manual melodic and tonal analysis is carried out by flamenco experts. From their analysis we obtain musical features with which we design a system to perform automatic melodic description. Such a system output a symbolic representation of singing melodies. We then use two distance measures, the correlation coefficient between pitch histograms and the edit distance, in order to measure the similarity between those representations. Those distances produced similarity distance matrices. With them we have used the tool SplitTrees to generate phylogenetic trees. Those trees were able to classify the different styles and revealed interesting clusters and relationships within the studied singing performances.


3/28

Introduction

Motivation

Research on flamenco music is a unique case. Flamenco music is both recognized and appreciated as a fascinating and vital musical tradition. Because of its richness and impact, flamenco music should have been carefully studied by musicologists and historians; however, they have long shown an unaccountable lack of interest in this music. Those who first studied flamenco music were enthusiastic connoisseurs, frequently without any academic qualification. Their sources and methods are not rigorous by current research standards and, since their analyses relied on limited sources, their work was often biased (Cruces, 2002). This situation slowly changed in the 1980s, when the first courses on flamenco music were taught at conservatoires. Much later, the basic problems in flamenco music were incorporated into the mainstream of research. Although much research on flamenco music has since been generated, the prevailing tendency is to study flamenco music through musicological methods such as source studies, style criticism, iconography or artists’ biographies. Although little attention has been paid to pure musical analysis, this paper is a token of such a change of approach.

In flamenco, singing styles are called cantes. This paper is concerned with a cappella singing styles. A cappella cantes are among the primary styles within the flamenco repertoire. Until very recently, flamenco singers have learned them only through oral transmission. In oral transmission, melody is a primary musical facet to be listened to, remembered, elaborated and spread. Moreover, melody has helped flamenco enthusiasts and musicians to remember and identify variants of a particular style or genre. Questions frequently discussed among flamenco scholars include how to quantify the similarity between two melodies, how to discriminate between different styles and variants among performers, and how to study the roots and the evolution of flamenco styles.

Goals and Structure of this Paper

The main goal of this paper is to classify certain a cappella singing styles known generically as tonás. The classification was first carried out manually by flamenco experts, who extracted the musical features characterizing the cantes. These features were then incorporated into an automatic system (a computer-based one) that performed a second classification. This automatic system was built to study different versions of a cante, several cantes belonging to the same style, and the evolution of cantes, (since we sometimes have historical and temporal information). In our judgment, human-based and computer-based methods are not opposed; rather the latter are a tool at the former’s service.

This paper is organized as follows. The next two sections describe the main characteristics of a cappella cantes and flamenco singing. Next, an account of the corpus used for the classification is presented. After that, manual description and annotation of flamenco cantes discussed and the classification carried out by the flamenco experts is presented. Next, the tools used in the automatic classification are described. This section includes a discussion of melodic transcription and measures of melodic similarity as well


4/28

as the use of phylogenetic trees. In the final section we analyze our results and draw conclusions.

A Cappella Flamenco Cantes

The Style of Tonás

We have restricted our analysis to songs without instrumentation, or in some cases with some percussion, known in flamenco music as cantes a palo seco. This corpus is composed of tonás, songs without musical accompaniment that generically include martinetes, deblas, saetas, tonás and carceleras. Our study was divided into two parts. In the first part, we concentrated on two main styles, namely, deblas and martinetes, of which we considered a small sample of 24 cantes. In the second part, we enlarged the sample to 101 cantes and incorporated the tonás. Since, in flamenco music, the word toná refers to both the style and one of its substyles, we will refer to “toná group” as the whole style and toná as the substyle.

Let’s now examine the musical features of the cantes in the tonás group. These cantes are played in free rhythm (sometimes the pattern of seguiriya is used as rhythmic accompaniment). Each singer chooses his or her own tonality. Modality, depending on the particular style, may be either the major mode or the Andalusian mode (Phrygian mode with a tonic major chord), though frequently both modes alternate within the same song (Fernández, 2004). Lyrics of songs range widely. A classification of the tonás group mainly based on lyrics was carried out by Lefranc (2000). Blas Vega (1967) studied the tonás group from a historical standpoint.

A toná typically possesses verses formed by words either of three or four syllables, or eight syllables, where the second and the forth verses may have assonant rhythm; often the final verse finishes with an imperfect tercet (off-rhyme tercet). Although of uncertain origin, many people in the flamenco world believe that tonás are the mother of all flamenco styles (cantes madres), from which all other styles are derived. Early researchers who tried to support this theory without a rigorous methodology dated its origins to the 18th century in Jerez and Triana (Molina and Mairena, 1963); however, more recent studies have proposed a later appearance (Jaramillo, 2002, among others).

The debla is a song based on the toná. Great melismatic ornamentation, more abrupt than the other songs of the tonás group, characterizes its melody.

The martinete is also considered a variant of the toná. It differs from deblas in the theme of its lyrics and in its melodic model, which always finishes in the major mode. The debla’s style is usually sad, played without guitar accompaniment, like the other tonás. Martinetes however are usually accompanied by the sound of a hammer struck against an anvil.

There are three clear variants of martinetes. The first one, to be called martinete 1, has no introduction, whereas the second one, to be called martinete 2, starts with a couple of verses from a toná. The third one, to be called martinete 3, is a concatenation of a toná


5/28

and some of the previous variants of martinetes; the toná of martinete 2 and 3 is called toná incipit. A trait of martinete 3 is the singing of several poems, all having the same theme as in medieval romances.

Characteristics of Flamenco Singing

Flamenco singing is different from other musical styles. Some of its features relate to the melodic structure (which is peculiar to the flamenco style or palo); other features relate to the expressive resources of the singer, and still others relate to the characteristics of the singer’s timbre. The absence of scores and the oral nature of flamenco make discrimination among diverse expressive assets difficult, whether one is looking at a particular style or at an individual singer. This difficulty reflects the long-standing discussion within the flamenco community regarding the definition of styles and variants.

Melody can have different representation levels. Energy and fundamental frequency curves are the primary low-level melodic features. Energy is associated with loudness and fundamental frequency is related to its perceptual correlate, pitch. Henceforward, the term pitch will refer to fundamental frequency.

The following are among the most important features of sung flamenco melodies (Fernández, 2004):

• Instability of pitch and energy envelopes.

• Microtonality: use of intervals smaller than the equal-tempered semitones of Western classical music.

• Pitch glides: portamenti or smooth transitions between notes. Notes are not very clearly attacked.

• Short pitch range or tessitura: normally limited to a sixth interval and characterized by the insistence on a note and those contiguous. Flamenco female singers are usually contralto (for greater intelligibility), while male singers are tenors or baritones.

• Enharmonic scale: microtonal intervallic differences between enharmonic notes.

• Melismas: high degree of ornamentation.

• Different timbres: trends in timbre characteristics of flamenco singers depend on the period. In general, there is a distinction between Gypsy and non-Gypsy voices. The Gypsies are fine interpreters of flamenco music and have created trends in flamenco singing.

• Breathiness in the voice.

• Formants: absence of high frequency (singer) formant, which is characteristic of classical singing styles.


6/28

These characteristics contrast with classical singing styles, in which precise tuning and timing is important, and in which timbre is characterized by stability, absence of breathiness, and high-frequency formants (i.e., the singer formant) (Sundberg, 1987). Some attempts have already been made to manually and graphically represent the expressive features of flamenco singing (Donnier, 1997) using the Gregorian chant neumes.

Methodology

The Corpus of Music

We chose the tonás group as the corpus of music for several reasons. Initially, we had at our disposal 365 pieces of the toná group. These were not festive cantes, but had a rather ritual mood. They could not be considered as commercial cantes since they had not been frequently recorded, and as a result had evolved more slowly than other flamenco styles.

Tonás have been removed from the original corpus. There are more than 45 types of tonás and for many there are insufficient recordings to permit a serious study. Also, we removed recordings of poor quality. Finally, we reduced our corpus to martinete 1, martinete 2, the toná incipit of martinete 3 and deblas. In total, we conducted our study on 101 pieces.

We gathered a representative music collection for this study. The corpus was found in diverse media (mostly vinyl and magnetic tapes) and comprises recordings from the 40’s through today. Criteria for selecting the pieces were musicological, geographical and historical. For example, Tomás Pabón is known as the singer who established the debla style, while Antonio Mairena and Naranjito de Triana are known as important contributors from the 80’s who influenced all subsequent singers.

One goal for this study was for the computer to recognize a style in the same way as would a flamenco aficionado. With that in mind, we decided to use only the first phrase of each piece, since that phrase presents the main melodic theme which is adequate for a flamenco aficionado to recognize the style.

For the first part of our study we selected 12 singers and the most representative deblas and martinetes (a total of 24 pieces). Table 1 shows the details of the singers, dates, locations and schools.

INSERT TABLE 1 HERE.

For the second part of the study, we expanded the corpus to 101 pieces. The above 24 martinetes of type 1 and deblas were included plus 27 martinetes of type 1, 20 martinetes of type 2, 11 deblas and 19 tonás incipit.

Manual Description and Annotation

A flamenco expert who also had formal musical training performed a manual melodic analysis of our initial corpus. For this analysis we first normalized the melodic contour


7/28

with respect to the tonic, and approximated notes to the closest in the equal-tempered scale. At this point, let us introduce the notion of perceived notes. Not all notes of a musical piece have the same importance. Lundin (1967, pp. 77-79) identified three particular melodic attributes, namely propinquity, repetition and finality, which allow the listener to recover the melodic contour and recognize the melody. Although not a formal definition, for the purpose of this study we define perceived notes as notes that are characteristic of the melody. The flamenco experts extracted the perceived notes and formed melodic contours of the pieces. From their contours we studied the musical features of the tonás group and integrated that knowledge into our computer program. For example, the flamenco experts removed certain ornamentation, or melisma, according to their previous knowledge of the particular style under analysis. Rhythmic representation is not straight, as manual annotation follows perception of rhythmic pulses, and not note durations. Figures 1-a and 1-b show graphical representations of the main melodic contour of a debla performed by Antonio Mairena and Tomás Pabón respectively. The reader may observe short-time differences, as well as a similar overall melodic contour. Figures 2-a and 2-b show graphical representations of the main melodic contour of a martinete sung by Antonio Mairena and Tomás Pabón, respectively. Here we also found some short-time differences and a similar overall melodic contour.


8/28

INSERT FIGURES 1 AND 2 HERE

With these manual patterns we conducted a similarity study based on multi-dimensional scaling, by using the Euclidean distance between the vector representations of the sequential values of melodies. A two-dimensional projection of the d-dimensional data is presented in Figure 3. We observe that martinetes (M) and deblas (D) are located in different regions of the 2-D plot. Note that martinetes from Mairena and Pabón (Tomás) are close to each other, whereas their deblas are very far away from each other.

INSERT FIGURE 3 HERE

We also performed a manual analysis of the reference pitch of each piece, which we identified by means of key labeling. The analyzed deblas provide a modal context, according to the following tonal structure: Tonic (I) – Subdominant (IV) – Tonic (I). As shown in Table 2 we observe a wide range of keys in the pieces analyzed.

INSERT TABLE2 HERE

We identified pieces in six different keys: G Major (Chocolate), A Major (Vargas), Bb Major (Pabón, Lobato and Heredia), B Major (Almadén and Turronero), C Major (Mairena, Simón and Naranjito) and C# Major (Pepe de Lucía and Talegón). We also observed microtonality in the modulations, mostly due to tuning problems. In fact, the flamenco singer is not trained to perform accurate modulations in traditional flamenco singing. The reference pitch mostly relates to the particular tessitura of the singer, but we might wonder if the chosen reference might also relate to a similarity among singing styles.

Computational Analysis

Because flamenco has traditionally been transmitted orally, the number of transcriptions is scant and some are incomplete. Transcriptions that were manually provided by experts in the field display characteristics of flamenco music that were implicitly considered in the transcription process. Although manual annotation is time consuming and might be subjective or prone to errors, manual analyses provide accurate and expert information. These problems were pointed out by Toiviainen & Eerola (2006) and Lesaffre et al. (2004) in different musical contexts. For flamenco music an additional problem relates to transcription since traditional Western notation may not be able to represent all expressive resources used in flamenco. This is particularly significant when considering singing voice compared to guitar playing.

In order to automate the melodic transcription and use alternative similarity distances as proposed in the literature a computational analysis was made. The latest techniques in automatic melodic analysis of audio allow us to obtain different representation levels of a music recording (Gómez et al., 2003). There are various representation levels for melody. Energy (intensity) and fundamental frequency (pitch) curves are the main low-level melodic features. At a higher structural level, note duration and pitch provide symbolic


9/28

representation, which may input to higher-level analyses. Finally, deviations of the analyzed recording with respect to the obtained score relate to expressivity. Some attempts have been made to apply these techniques to analysis of flamenco music (Donnier, 1997, Gómez and Bonada, 2008).

Melodic Transcription

Melodic transcription allows us to compute the fundamental frequency envelope and a MIDI-like representation of singing melodies (onsets, offsets and frequency of each pitch event). It is important to note that because each singer has his/her own reference tone and sings each note relative to the scale constructed on that tone (Haus and Pollastri, 2001), we must estimate this tuning frequency. The steps for melodic description can be summarized as follows:

1. Estimate low-level features. We compute a set of low-level features from the audio signal: energy in different frequency bands and pitch. The pitch estimation algorithm is based on amplitude correlation in the frequency domain.

2. Estimate tuning frequency. When analyzing singing voice performance, we do not know the reference frequency (with respect to 440 Hz) used by the singer. To locate the main pitches, we perform an initial estimation of the tuning frequency (i.e. reference frequency used by the singer to tune the piece), assuming an equal-tempered scale system. This tuning frequency is computed by minimizing the estimated instantaneous pitch error weighted average. The weights are computed by combining energy and first and second pitch derivatives.

3. Transcribe short notes. The audio signal is segmented into short notes using a dynamic programming algorithm that finds the segmentation which maximizes a set of probability functions that consider pitch error, energy variations and note durations.

4. Iteratively consolidate notes and refine tuning frequency. The estimated tuning frequency is refined according to the obtained notes by minimizing the note pitch error weighted average, where weights depend on note durations. Then, existing consecutive notes with the same pitch and a soft transition are consolidated until there is no further consolidation.

5. Estimate vibrato. Vibrato depth and rate descriptors are computed from the zero-crossings obtained by subtracting the transcribed note nominal pitch to the pitch function.

Figures 4 shows an example of the obtained representation for a flamenco audio excerpt sung by a female singer. The top view represents the waveform. The middle view shows the obtained note transcription, including quantized notes (up to a semitone) and the instantaneous pitch envelope, where the presence of vibratos, pitch glides and melismas can be observed. The bottom figure represents the energy en, which relates to the sound intensity.


10/28


Figures 5-a and 5-b provide graphical representations of the extracted melodic contour of a debla performed by Antonio Mairena and Tomás Pabón respectively. Figures 6-a and 6-b provide graphical representations of the extracted melodic contour of a martinete also performed by Antonio Mairena and Tomás Pabón respectively. We observed some differences with respect to manual annotations.

INSERT FIGURES 5 AND 6

Some errors in automatic transcriptions are due to the lack of tuning, but most are due to simplifications and assumptions made when manually labeling the melodies (i.e. melisma are not considered, note durations are neglected and only the main melodic anchor points are notated).

Once all of the pieces are converted into MIDI notes, we normalized them with respect to key, so that the melodic representation is invariant to transposition. To obtain this we computed the intervals between consecutive pitches, instead of working with the absolute pitch values. We also wanted the representation of the pieces not to be affected by tempo. We normalized note duration with respect to the duration of the previous note.

Measures for Similarity Computation

There is a vast body of research on how to measure similarity between melodies using computational models, usually inspired by methods for string comparison (Crawford et alt., 1998, Mongeau and Sankoff, 1990). With respect to rhythm, and specifically in flamenco, Díaz-Bañez et al. (2004) studied rhythmic similarity of flamenco patterns. Our preliminary task was to implement several melodic similarity measures to determine the most appropriate one for our purpose. We selected the following two measures that gave satisfactory results:

• Correlation coefficient between note histograms. We measured how interval distribution correlates without regard to its order. Correlation indicates the strength and direction of a linear relationship between two random variables. We obtained this coefficient for each pair of pieces and produced a similarity matrix.

• Edit distance (Mongeau and Sankoff 1990). This metric counts the difference between two sequences under certain operations. The edit distance between two strings is determined by the minimum number of operations needed to transform one string into the other, where an operation is an insertion, deletion, or substitution of a single note. Both pitch and note duration information was considered in the algorithm, as they increase the discrimination (López and Rocamora, 2005).

Phylogenetic Trees

In addition to similarity analysis, a phylogenetic analysis was made to study relationships among the transcribed performances. Several techniques exist for generating


11/28

phylogenetic trees from distance matrices. The choice of technique depends on the specific application. To study martinetes and deblas we utilized BioNJ algorithm (Gascuel, 1997) since the set of cantes was small and a tree structure could reasonably be expected. The BioNJ algorithm imposes the structure of a tree on the similarity matrix, though when changes are due to many causes this may not be advisable and where the structure is a general graph, networks should probably be considered. For the actual computing, we used the tool SplitsTree (Huson 1998). SplitsTree computes a tree (or more generally a graph) with the property that the distance in the drawing between any two nodes reflects as closely as possible the true distance between the corresponding two pieces in the distance matrix.

Case Study 1: Martinetes and Tonás

As previously mentioned, we selected the most representative deblas and martinetes (total 24, 12 in each style). Flamenco experts (see Figure 1) extracted manual annotations and then our automatic system was fed with the audio files and generated corresponding transcriptions (see Figure 5). In general, we consider that the automatic system extracted features similar to those extracted by the flamenco experts. The next step was to compute the correlation coefficient between interval histograms, which resulted in a similarity matrix, shown in Figure 7.


The code color used to display the matrix shows us that not all the deblas and martinetes were well discriminated by the correlation scheme. This is mainly due to the fact that this measure does not consider note sequences, but only consider interval distributions. This distance measure has been used to establish a baseline for latter analyses.

In Figure 8 we can see a tree generated by the BioNJ algorithm with the similarity matrix given by the edit distance. As expected, due to the different melodic contour of each style, two clusters appear: the first includes the deblas, and the second includes the martinetes.


Case Study 2: the Entire Corpus

The entire corpus was formed by adding 27 martinetes 1, 20 martinetes 2, 11 deblas and 19 tonás incipit. Manual transcription for the entire transcription was not performed, since it would have been a highly time-consuming task.

The first experiment we conducted was to compute the phylogenetic tree for the set of martinetes of type 1 and 2. Our intention was to test out our system on the discrimination of those styles. Figure 9 shows a BioNJ tree generated by the SplitTree software, where melodic similarity is based on edit distance. The legend of the figure is the following: M1-x, for x=36,…, 62, are the martinetes of type 1; M2-x, for x=63,…, 82, are the


12/28

martinetes of type 2. Notice the high Lsfit obtained, a 99.87%, which means that the graph correctly reflects the distances of the matrix.


On the tree three main clusters, which are highlighted in the figure, can be observed. There are two clusters comprising most of the martinetes of type 1. On the cluster located on the left side there is only a mismatch, the martinete M2-66. The other cluster of martinetes of type 1 located on the right side and is smaller than the previous cluster. Finally, the cluster corresponding to martinetes of type 2 has only a mismatch, the martinete M1-62. Outside those clusters there is a mixture of both types of martinetes.

Our next phylogenetic graph was computed on the entire corpus, also using edit distance. It is shown in Figure 10 below. The LSfit percentage for this tree is 99.84%.


The legend of the figure is the following: D-x, for x=1,…,12, are the deblas included in the first study; D2-x, for x=13,…,23, are the deblas incorporated in the second study, M-x, for x=24,…35, are the martinetes of the first study; M1-x, for x=36,…, 62, are the martinetes of type 1; M2-x, for x=63,…, 82, are the martinetes of type 2; T-x, for x=83,…, 101, are the tonás incipit.

In general the accurateness of the classification is smaller when considered the entire corpus. Several clusters are, however, noticeable. There is a big cluster containing the most of deblas (located on the rightmost part of Figure 10). Moreover, there are two big clusters that encompass most of martinetes in general, although both types of martinetes are mixed within the clusters. Again, outside of the main clusters there is a high degree of mixture among the styles.

Tonás are spread over the tree, except for the debla cluster, and are quite mixed with the martinetes. It is not as surprising as it may first seem since there is a great degree of improvisation in the toná incipit. This fact makes the tonás not to stay in a fixed cluster.

Conclusions

Conclusions

This paper deals with the automatic classification of flamenco a cappella singing styles. We have seen that melody is one of the most important aspects to be considered in the analysis of these styles, and we have proved that standard automatic melodic description and similarity techniques can be effectively used to identify the most relevant a cappella singing styles in flamenco music.

This first study combines manual and automatic analysis. Current results show that the computational approach provides a good discrimination between deblas and martinetes, although some confusion is found among the different martinete variants. This coincides to some extent with the discrimination ability of a flamenco enthusiast: distinguishing martinete variants is more difficult than identifying a debla, because the debla style has a


13/28

very characteristic melody followed by all singers, who preserve the original version by Pabón.

We believe that the proposed automatic tools open a path to obtain quantitative and qualitative measures that can complement historical and musical information (and other types) on the roots and evolution of orally transmitted flamenco styles.

Future work

Although current results are encouraging, there are still several areas to improve. Regarding automatic melodic description, vibratos on top of glissandos are not well detected, and rapid short notes are sometimes considered to be part of longer notes. Besides, in a cappella singing, it is not that rare that the reference frequency varies along the song, but the current implementation of the transcription algorithm does not allow that frequency variation. Regarding melodic similarity measures, an interesting area to explore is to compare the use of distances between pitch envelopes, which would include both melodic information and expressive nuances. Another aspect to research is the quantitative evaluation of the distance measures inside each of the styles. One idea is to compare mathematical measures with human ratings and analyze the influence of musical and flamenco expertise, as proposed by Guastavino et al. (2008) in the context of flamenco rhythmic patterns. The clusters obtained from the phylogenetic trees suggest several venues for further research. Those clusters pointed out melodic similarities that would be worth understanding.

Acknowledgments

This research project is supported by the Andalusian Agency for the Development of Flamenco Music under the COFLA Project.


14/28

References

Blas Vega, J. (1967). Historia de las tonás, ed. Guadalhorce, Málaga.

Crawford,T., Iliopoulos, C.S., Raman, R. (1998), String matching techniques for musical

similarity and melodic recognition, Computing in Musicology.

Cruces, C. (editor) (2002). Historia del Flamenco. Ed. Tartessos, Sevilla.

De Mulder, T., Martens, J. P. Lesaffre, M., Leman, M., De Baets, B., De Meyer, H.

(2003), An Auditory Model Based Transcriber of Vocal Queries, Proceedings of the

International Conference on Music Information Retrieval.

Díaz-Báñez J. M., G. Farigu, F. Gómez, D. Rappaport, G. T. Toussaint (2004). El

Compás Flamenco: A Phylogenetic Analysis. Proceedings of BRIDGES: Mathematical

Connections in Art, Music, and Science, Winfield, Kansas, pages 61-70.

Donnier, P (1997). Flamenco: elementos para la transcripción del cante y la guitarra,

Proceedings of the IIIrd Congress of the Spanish Ethnomusicology Society.

Dress, A., Huson, D. and Moulton, V. (1996). Analysing and visualizing sequence and

distance data using SPLITSTREE. Discrete Applied Mathematics, 71:95-109.

Fernández, L (2004). Flamenco music theory. Acordes concert.

Gascuel, O. (1997). “BIONJ: an improved version of the NJ algorithm based on a simple

model of sequence data,” Molecular Biology and Evolution, 14, pp. 685-695.

Gómez, E., Bonada, J. Automatic transcription of flamenco singing, Conference on

Interdisciplinary Musicology, Thessaloniki, Greece, 2008.


15/28

Gómez, E. Klapuri, A. Meudic, B. (2003). Melody Description and Extraction in the

Context of Music Content Processing, Journal of New Music Research Vol.32 .1

Guastavino, C., Gómez, F., Toussaint, G., Marandola, F., Gómez, E. (2009). Perception

of rhythmic similarity in Flamenco music, in preparation.

Hernández Jaramillo, J. M. (2002). La Música Preflamenca, Sevilla, Consejería de

Relaciones Institucionales - Junta de Andalucía.

Haus, G. and Pollastri, E. (2001). An Audio Front End for Query-by-Humming Systems.

In Proc. of ISMIR 2001, Bloomington, IN, Usa, Oct.

Huson, D. H. (1998). SplitsTree: Analyzing and visualizing evolutionary data,

Bioinformatics, 14:68-73.

Israel J. Katz. Flamenco. Grove Music Online ed. L. Macy (Accessed 16 May, 2006),

www.grovemusic.com

Pierre Lefranc. (2000). El Cante Jondo. Del territorio a los repertorios: tonás, seguiriyas,

soleares. Publicaciones de la Universidad de Sevilla, colección Cultura Viva, n° 16.

Lesaffre, M., Leman, M., De Baets, B. and Martens, J.-P. (2004). Methodological

considerations concerning manual annotation of musical audio in function of algorithm

development, Proceedings of the International Conference on Music Information

Retrieval.

López, E., Rocamora, M. (2005) Tararira: Sistema de búsqueda de música por melodía

cantada, 10th Brazilian Symposium on Computer Music,

http://iie.fing.edu.uy/investigacion/grupos/gmm/proyectos/tararira/index


16/28

Lundin, R. W. (1967). An objective psychology of music (2nd ed.). New York: Ronald

Press.

Molina, R., Mairena, A. (1963). Mundo y formas del cante flamenco, Ed. Revista de Occidente, Madrid, 1963.

Mongeau, M., Sankoff, D. (1990), Comparison of musical sequences, Computers and the

Humanities, 24:161–175.

Ryynänen, M. P. (2006). Singing transcription, in Signal processing methods for music

transcription (A. Klapuri and M. Davy, eds.), Springer.

Sundberg, J. (1987). The Science of the Singing Voice. DeKalb, IL: Northern Illinois

Univ. Press.

Toiviainen, P. & Eerola, T. (2006). Visualization in comparative music research. In A.

Rizzi & M Vichi (Eds.), COMPSTAT 2006 - Computational Statistics. Heidelberg:

Physica-Verlag, 209-221.


17/28

Table Captions

Table 1. Information about the corpus of martinetes and deblas.

Table 2. Keys of the analyzed deblas.


18/28

Figure Captions

Figure 1-a: Manual transcription of a debla performed by Antonio Mairena.

Figure 2-a: Manual transcription of a martinete performed by Antonio Mairena.

Figure 1-b: Manual transcription of a debla performed by Tomás Pabón.

Figure 2-b: Manual transcription of a martinete performed by Tomás Pabón.

Figure 3: Multi-dimensional scaling based on manual transcription.

Figure 4: Example of an obtained melodic transcription.

Figure 5-a: Automatic transcription of a debla performed by Antonio Mairena.

Figure 5-b: Automatic transcription of a debla performed by Tomás Pabón.

Figure 6-a: Automatic transcription of a martinete performed by Antonio Mairena.

Figure 6-b: Automatic transcription of a martinete performed by Tomás Pabón.

Figure 7: Similarity matrix based on pitch histogram correlation.

Figure 8: A tree generated by BioNJ based on edit distance for the 12 deblas and 12 martinetes.

Figure 9: The BioNJ tree for martinetes 1 and martinetes 2. Melodic similarity is based on the edit distance measure.

Figure 10: The BioNJ tree for the entire corpus. Melodic similarity is based on the edit distance measure.


19/28

Singer Year Location School

Antonio Mairena 1960 Mairena del Alcor Pabón

Chano Lobato 2002 Cádiz Pabón Mairena

Chocolate 1999 Jerez de la Frontera Pabón

Mairena

J. Almadén 1985 Ciudad Real Pabón

Mairena

Jesús Heredia 2002 Écija Pabón

Mairena

M. Simón 1985 Jerez de la Frontera Pabón

M. Vargas 1972 Cazalla de la Sierra Mairena

Naranjito 2002 Triana (Sevilla) Pabón

Mairena

Pepe de Lucía 1963 Algeciras Pabón

Talegón 2002 Córdoba Pabón

Mairena

Tomás Pabón 1950 Triana (Sevilla) El Baboso

Turronero 1989 Utrera Pabón

Mairena

Table 1: Information about the corpus of martinetes and deblas.


20/28

Singer Key

Antonio Mairena C Major Chano Lobato Bb Major Chocolate G Major J. Almadén B Major J. Heredia Bb Major M. Simón C Major M. Vargas A Major Naranjito C Major Pepe de Lucía C# Major Talegón C# Major Tomás Pabón Bb Major Turronero B Major Table 2. Keys of the analyzed deblas.


21/28

0

2

4

6

8

10

12

-6

-4

-2

0

2

4

6

8

10

Fig. 1-b. Manual transcription of a debla

performed by Tomás Pabón. Fig. 2-b. Manual transcription of a martinete

performed by Tomás Pabón.

Fig. 1-a. Manual transcription of a debla

performed by Antonio Mairena.

0

2

4

6

8

10

12

Fig. 2-a. Manual transcription of a martinete


-6

-4

-2

0

2

4

6

8

10


22/28

Figure 3: Multi-dimensional scaling based on manual transcription.


23/28

Figure 4: Example of computed melodic transcription.


24/28

Figure 5-b. Automatic transcription of a

debla performed by Tomás Pabón.

Figure 6-b. Automatic transcription of a

martinete performed by Tomás Pabón.

Figure 5-a. Automatic transcription of a debla


Figure 6-a. Automatic transcription of a

martinete performed by Antonio Mairena.

0 5 10 15 20 25 30 35 40 45 5054

56

58

60

62

64

66

68

70

note index

0 5 10 15 20 25 30 35 40 4554

56

58

60

62

64

66

68

70

note index

MIDI pitch

0 5 10 15 20 25 3050

52

54

56

58

60

62

64

66

note index

MIDI pitch

0 5 10 15 20 2550

52

54

56

58

60

62

64

66

note index

MIDI pitch


25/28

Figure 7: Similarity matrix based on pitch histogram correlation.


26/28

D-TURRONERO

M-TURRONERO

M-PABON

D-PABON

D-LUCIA

M-LUCIA

M-LOBATO

D-LOBATO

D-MAIRENA

D-HEREDIA

D-CHOCOLATE

D-ALMADEN

M-TALEGON

D-NARANJITO

M-VARGAS

D-SIMON

M-SIMON

D-VARGAS

M-HEREDIA

M-MAIRENA

M-CHOCOLATE

M-ALMADEN

D-TALEGON

M-NARANJITO

PHILOGENETIC TREE

Figure 8. A tree generated by BioNJ, based on edit distance, for the 12 deblas and 12 martinetes.


27/28

Figure 9. The BioNJ tree for martinets 1 and martinets 2,

Melodic similarity is based on the edit distance measure.


28/28

Figure 10. The BioNJ tree for the entire corpus, Melodic similarity is based on the edit distance measure.

Comparative Analysis of A Cappella Flamenco Cantespersonal.us.es/dbanez/papers/cofla-journal.pdf ·...

Documents

Transcript of Comparative Analysis of A Cappella Flamenco Cantespersonal.us.es/dbanez/papers/cofla-journal.pdf ·...