A STUDY OF FRET POSITIONS OF TANBUR BASED ON AUTOMATIC ESTIMATES FROM AUDIO RECORDINGS Gedik, Ali C., Dept.

of Electrical and Electronics Eng., Izmir Institute of Technology, a.cenkgedik@musicstudies.org Bozkurt, Baris, Dept. of Electrical and Electronics Eng., Izmir Institute of Technology, barisbozkurt@iyte.edu.tr Cirak, Cem, State Conservatory of Turkish Music, Ege University, cemcirak@hotmail.com
ABSTRACT Background in music theory Tanbur (or tambur) is the common name of long-necked, fretted, plucked lutes of the Middle East and Central Asia [1]. The studies of the earliest theorists such as Al-Frb (d 950) and Saf al-dn (d 1294) based on the fretting of tanbur for the description of tuning systems show the importance of tanbur in traditional art musics of this wide geographical region. The first analytical study on tanbur kebir trki (great Turkish tanbur) which forms a basis for the theory and practice of traditional Turkish art music (TTAM) was presented by Dimitrie Cantemir (16731723) in his treatise, Edvar-i musiki (Textbook of music) [2]. Therefore there is a strong relationship between the fretting of tanbur and the tuning system of TTAM. However due to the divergence of theory and practice in TTAM [3] the fret positions of tanbur is still an open topic of research: there exists no agreed standard among theoreticians, musicians and instrument makers. Since early 20th Century, various tuning theories have been proposed in Turkey [4] with varying number of pitches used in an octave: 24, 29, 41, 53, 79, etc. Today, the number of frets still varies due to personal choice and instrumental tractates provide alternative fret locations according to this or that musician [5]. Background in computing The first computational study on tanbur is presented by [6]. However this study focuses on the acoustic features of tanbur rather than fretting. An automatic analysis method for the tuning analysis of traditional Turkish art music (TTAM) recordings is recently presented by [7]. The algorithms applied in tuning analysis are as follows: pitch frequency analysis pitch histogram computation, histogram alignment and histogram template construction on large audio databases containing recordings with non-standard tuning frequencies. These algorithms have been successfully used to study theory-practice mismatches in TTAM [4]. Aims Primarily we present an automatic analysis method that can estimate fret positions of a tanbur given a collection of recordings. We also study the problem of fret positions by comparing the estimated fret locations to those specified in the literature. Main Contribution We first discuss the importance of the instrument with all of its appearances in music: the theory, practice and production. We summarize the literature of tanbur fretting and show that there is an important lack of reliable information in the domain. We finally introduce a method for estimation of fret placements from audio recordings and show via tests that the method is reliable. Implications The presented algorithm is potentially applicable in tuning studies and research on tanbur education and production. and includes the description of various tanburs of their time. It is accepted that the oldest tanbur description that resembles most to the tanbur performed in traditional Turkish art music (TTAM) today is given by Dimitrie Cantemir (16731723) who was also respected as a master of the instrument of his time [2]. A comparison of tanbur drawn by Cantemir and tanbur performed today is presented in Figure 1. The first analytical study which

1.

INTRODUCTION

Tanbur (or tambur) is the common name of long-necked, fretted, plucked lutes of the Middle East and Central Asia [1]. Manuscripts of the theorists such as Al-Frb (d 950) and Saf al-dn (d 1294) are considered as common sources for the theory of traditional art musics of the Middle East

forms a basis for the theory and practice of traditional Turkish art music (TTAM) is also presented by Cantemir in his treatise, Edvar-i musiki (Textbook of music) [2].

a)

b)

Figure 1. Tanbur in TTAM: a) the drawing of Cantemir (16731723) [8] and b) today [6]. The interest of theorists on tanbur continues in the 20th century. The founder of the modern theory of TTAM, Rauf Yekta Bey whose article on TTAM was published in the Encyclopedie le da Musique et Dictionaire du Conservatoire in 1922 was also a tanbur player. In his article he discusses the tuning theory of TTAM based on tanbur and underlines the importance of the instrument: Tanbur is the most beloved instrument of Turks. While old Arabian and Iranian writers consider ud as the most perfect instrument, Turkish writers consider tanbur holding such an honor. For a comparison, it can be said that the tanbur plays the similar role which piano plays for Western composers. Similarly, most of the Turkish composers are tanbur players, also. [19] Rauf Yekta Bey together with the most prominent tanbur player Mesut Cemil represented Turkey in The Congress of Arab Music held in Cairo in 1932 [9]. Cemil also performed in the congress which was recorded by the recording committee of the congress. The congress was a historical turning point toward standardizing the theory and practice of traditional art musics of the Arabian geography. Yekta and Cemil, as members of the musical scale committee, rejected the proposal of some Egyptian members on the use of equal-tempered quarter-tone scales, for its inappropriateness in measuring the Near Eastern pitch [9]. As reported by Racy, tanbur was favored by some committee members for its rich and enchanting

sound, and because its fretted long neck was useful for devising a standard theoretical scale. [9]. Although ud and kanun started to dominate the genre by displacing tanbur to a degree in 1900s [10], tanbur preserves its central role in education and performance of the genre at least in certain circles such as few private educational institutions, state conservatories, vocal and instrumental ensembles of state TV and radios, and amateur choruses where the tradition survived. Therefore, while it is more likely to hear and see ud and kanun performances then tanbur performances in more popularized discourses of the genre in private TV, radios and concerts, tanbur is still indispensably performed in all forms of the genre, either vocal or instrumental in those rather restricted circles in Turkey, today. However, the most prominent and respected performers of TTAM are considered as tanbur players such as Tanburi Cemil Bey, Mesut Cemil, Necdet Yaar, Ercment Batanay, zzettin ke, Fahrettin imenli, etc in any circle of the genre. Furthermore tanbur still preserves its central role in theoretical writings also, today. Recently Yavuzoglu [11] proposed a new tuning system for the genre and applied the system on a tanbur which was also performed in a recent conference on TTAM after the presentation of the proposed theory. Consequently, the tanbur occupies a distinctive place in education, performance and theory of TTAM both historically and currently. However this is the point where the tanbur also stands at the center of various problems about the genre in the 20th century. These problems mainly result from the divergence of theory and practice in TTAM. Although the divergence of theory and practice is an historical fact of the genre, the appearance of this fact as a problem occurs due to the westernization and nationalization of TTAM in the 20th century [3]. Historically, practice and theory have a loose relation where the former is based on oral tradition and the letter is rather a combination of speculation and a musicological scientific method. While the theory was rather based on verbal descriptive information, theorists in the 20th century intended to formulate the theory based on more analytical approaches using a terminology similar to western music theory, such as scale degrees, tetrachords, pentachords, etc. Therefore a new discourse started to dominate the genre especially after the institutionalization of the TTAM by the state conservatories and ensembles: the theory should generate practice [12]. The theory of Arel-EzgiUzdilek became the official theory of TTAM by the foundation of state conservatory of TTAM in 1976. The outcome of the institutionalization of TTAM both in education and performance is the appreciation of theory more seriously by the performers than ever before. However this consequence did not solve the problem of divergence but generates new discourses among the

musicians, resulting the description of practice with respect to theory. On the one hand, performance of certain pitches which contradict with the theory are defined with respect to theory by using a terminology such as a little higher, a little lower or minus a comma [13]. As stated in [3]: although the performances diverge from the theory, the Arel theory is highly respected among performers, and they hesitate to contradict the theory when the pitch intervals of their performances are measured by musicologists1. On the other hand, the theorists observing the divergence of the Arel-Ezgi-Uzdilek theory and practice, proposed new theories which target converging the theory to practice. These theories which proposed various numbers of pitches for an octave such as 24, 29, 41, 53, 79 etc., were considered and computationally evaluated by [4]. As a result, reflection of the divergence problem on tanbur can be summarized as follows: While the theory dictates standardization in practice, there is almost no standardization in any appearance of tanbur in practice. This can be listed more specifically under three headings: i. Production: The dimensions of the sound box and the neck, and the number of frets and their placements can vary due to each instrument maker. ii. Performance: Once a player owns a tanbur from the instrument maker, the player reorganizes the number of frets and their placements according to personal choices. iii. Education: Although TTAM is historically based on an oral tradition called mek, as a result of modernization processes education demonstrates a hybrid structure, based on both mek and written sources, today. Textbooks consist of theoretical information and, exercises and compositions written on western staff notation adapted to TTAM2 according to the theory (see Figure 2) are used even in amateur choruses as well as in conservatories. Master musicians as lecturers continue to perform with students within the context of mek. However due to the divergence of theory and practice, the pitches at the notations do not match exactly with the performed pitches. Consequently our study, estimation of fret placements from audio recordings, proposes solutions to these three problems for tanbur. The calculation of fret placements of master tanbur players clearly supplies useful information for the production, performance and education of tanbur instrument. However it should be noted that our study does not intend to constitute any standardizations in any of these practices of tanbur.

Figure 2. A transcription of a composition written on the western staff notation adapted to TTAM. [14] Our study for the computational analysis of fret placements estimated from audio recordings mainly consist of two phases: analysis of pitch frequency information and mapping the pitch information to the fretboard of the tanbur as fret placements. The method for the first phase of the study is already developed by Bozkurt [7] and successfully applied in a number of recent studies [4, 15]. In this algorithm, pitch information is represented as pitch-frequency histograms and the tonic for each piece is estimated by an automatic tonic detection algorithm based on histogram template matching. Finally pitch interval values are obtained from the pitch-frequency histogram aligned to the tonic. However the second phase of the study, mapping pitch interval values to the fretboard of the tanbur, bears challenging problems. Firstly, there is no source about the calculation of fret placements in tanbur, except tables for fixed string lengths according to this or that musician [5]. The literature is mainly based on western fretted string instruments [16, 17, 18]. The few studies on non-western fretted string instruments such as sitar, vina [20], Iraqian long necked lute [21] etc. do not consider the fret placement problems such as fret compensation. Mostly the fret placement of tanbur is given as a table with measured distances between frets and nut or bridge in the literature on TTAM. However this information is unreliable since neither the height of the strings measured from the nut and bridge is given nor the problems of fret placement in fretted string instruments is taken into account. The most comprehensive study on tanbur lists fret locations directly computed by applying the theoretical intervals on the string length [22]. The only computational study on tanbur is also far from dealing with fret placement of tanbur but presents the acoustic analysis of the instrument [6]. Consequently we present here a method for the automatic estimation of fret placements of tanbur from the audio recordings for the first time. Firstly we measured the string length, the height of the strings measured from the nut and distances of each fret from the nut of a tanbur and recorded a number of performances of a tanbur player with the same tanbur. After applying the automatic pitchfrequency analysis to the recordings and comparing the automatic fret mapping results to the actual fret placements

1 Karl Signell and M. Kemal Karaosmanolu (quoted from Can Akko) shared their measurement experiences with foremost performers.(personal communication with Signell and Karaosmanolu, 68 March 2008, stanbul) 2 The main difference of the notation staff used in TTAM and western music is the eight additional accidentals in TTAM notation.

measured, we have calculated parameters of the fret placements including the fret compensations. 2. TANBUR IN THEORY AND PRACTICE

2.1. Morphology of tanbur There are various tanburs with different dimensions specified by the string length (distance between the bridge and the nut) such as 100, 102, 104, 106, 108, 110, and 112 cm tanburs [23]. All other dimensions of the tanbur are determined by certain ratios reference to the string length in production. For example the ratio of the string length to the length of the fretboard is 4/3. The most common tanbur dimension used today is the tanbur with 104 cm string length and 78 cm length fretboard [22]. The frets along the fretboard cover a range of two octaves. The frontal and lateral views of tanbur performed in TTAM today are presented in Figure 1.b and Figure 3 respectively. The quasi-hemispheric soundbox is made of thin slices with 3 mm thickness, 3-4 cm width and 55-60 cm length. The number of slices depends on the dimensions of the tanbur [22]. Slices are made of hard woods such as ebony, rosewood, pearwood, walnut and cherry. The soundboard usually consists of 2 thin pieces cut from spruce with 1.80-2 mm thickness. Rosewood or juniper is used for the bridge and originally tortoise shell is used for the plectrum which is replaced by synthetic materials today. Finally the fretboard is made of ebony or juniper and the frets are made of nylon. The frets wounded on the fretboard are 0.4 - 0.6 mm wide. Either 7 or 8 strings are used at the instrument as pairs and tuned as shown in Figure 4. However the tuning of the first pair can be changed when transposition is necessary. Traditionally only the string couple tuned as yegah is used to play the melody of a composition or improvisation and the other string pairs function as resonators or to supply the tonic during improvisations. If 7 strings are used at the tanbur, then the 4th pair is reduced to one string. Two kinds of strings are used at tanbur today: plain steel strings and yellowish strings made of brass, copper and bronze or mixture of them. The diameter and the kind of the strings in millimeter are as follows: plain steel 1st pair strings (D) - 0.3, yellowish 2nd pair strings - 0.4, plain steel 3rd pair strings 0.3, yellowish 4th pair strings 0.5. The tension of the strings applied on the tanbur is around 75-80 kg. but this tension is subject to changes due to the transposition applied. 2.2. Fretting problems of tanbur As stated in the Introduction, the main difficulty about tanbur fretting raises due to the divergence of theory and practice in TTAM. Therefore the problem is not specific to an instrument. However, tanbur stands at the center of the discussions about this divergence problem due to its central

role in both theory and practice. Firstly the number of pitches in TTAM is still subject to discussions which lead to the question about the number of frets to be used in a tanbur. Yekta presents 49 frets (for two octaves) for a tanbur which is explicitly related with the tuning system he proposed as 24 pitch intervals within an octave. Similarly, the tanbur with 98 frets presented by Yavuzolu [11] is the application of his tuning system with 48 pitch intervals within an octave.

Figure 3. Cross section of tanbur performed in Turkey, today [6].

Figure 4. The tuning of the tanbur as couple of strings [22]. However none of the theoretical proposals for tanbur are applied in practice. The theory of Arel-Ezgi-Uzdilek (AEU) which is considered as the official theory of TTAM is simply the predecessor of the theory presented by Yekta. However it is not possible to find a tanbur with 49 frets in Turkey, today. Furthermore, the number of frets used by one of the most notable tanbur players, Necdet Yaar is 65. Although the increase in the number of frets is explained by the need of transposition, there is no standard also to meet this need. Therefore, it is also possible to find tanburs with 56 frets. Besides the problem of the number of frets, fret placement is another crucial point where the practice diverges from theory. In production, the commonly applied methodology is to use templates drawn on sheets which mark the fret placements on tanbur (every producer has his template). These templates are either derived from theoretical information or specific to this or that instrument maker. However there is no reliable scientific source about the procedure of producing these templates. Table 1 presents the pitch interval values as frequency ratios and the fret placements of a tanbur with 1064 mm

string length given by Yekta [19]. Only the first 25 frets are given in the table to save space. The next 24 frets are simply the octave shift of the first 25 frets which can be simply found by applying the frequency ratios. Consequently the frequency ratios reflects the tuning theory proposed by Yekta which supplies the pitch interval values used as official theoretical information as a result of AEU theory.
Fret Distance to Bridge (mm) 1064 1009.97 996.38 958.69 945.78 897.75 885.67 852.17 840.70 808.89 798 757.48 747.29 719.02 709.34 673.32 664.26 639.13 630.52 606.67 598.5 568.11 560.47 539.26 532 Pitch Names Turkish YEGH Nim Pest Hisar Pest Hisar Dik Pest Hisar HSEYNARAN Acemairan Dik Acemairan ARAK Gevet Dik Gevet RAST Nim Zengle Zengle Dik Zengle DGH Krd Dik Krd SEGH Puselik Dik Puselik ARGH Nim Hicaz Hicaz Dik Hicaz NEVA D C A G E Western D

No Frequency Ratios 1 2 3 4 5 6 7 8 9 11 12 13 14 15 16 17 18 19 21 22 23 24 25 1/1 256/243 2187/2048 65536/59049 9/8 32/27 19683/16384 8192/6561 81/64 4/3 1024/729 729/512 262144/177147 3/2 128/81 6561/4096 32768/19683 27/16 16/9 4096/2187 243/128 1048576/531441 2/1

10 2097152/1594323

20 8388608/4782969

mismatch is understandable to a degree. However we observe similar problems in comparatively very recent documents like the unique book on tanbur making written by a tanbur maker, Cafer An [23] as presented in Table 2. Table 3 presents three fretting systems1 given for a tanbur with 104 cm string length (given by tanbur maker Cafer An (CA)2 in [23]): fret locations of the tanbur of Necdet Yaar (NY) (a master of the instrument), the theory of Arel-Ezgi-Uzdilek (AEU) and Cafer Ans fretting system. Therefore Table 3 reflects available information in the literature on tanbur frettings for a respected performance (NY), the official theory (AEU) and the production (CA) in charge. The same mismatch of Yektas values (the double octave fret being at exactly one quarter of the string length) also holds true for the two fretting systems, CA and NY. The two systems exactly fit to the theoretical fretting of Yekta, except the additional frets used for the need of transpositions as can be observed from the Table 2. Furthermore it is also clear from the same table that the tuning system of AEU is exactly the same with the system of Yekta. Consequently, it can be said that the fret measures presented in the rather limited literature, referring to the tanbur of a master player and presented by a master maker are simply the application of the AEU/Yekta tuning system to a tanbur with a given string length which is far from the actual practical fret measures. Therefore it is clear that the reliability of literature is questionable. Consequently a tanbur player has to move the frets to find the correct fret placements after owning a tanbur, almost without any guidance from the theory or instrument maker, today. There is a crucial need for reliable measurement tools to help this process. 3. A COMPUTATIONAL APPROACH TO STUDY THE FRET PLACEMENTS OF TANBUR

Table 1. The pitch interval values and the fret placements of a tanbur with 1064 mm string length given by Yekta [19]. The interesting point in the table is that fret placements are simply reflections of frequency ratios applied to a tanbur with 106.4 cm length. In fact it is not practically possible to have an octave relation (2/1) between the frets yegah and neva given as 1064 mm and 532 mm with reference to bridge, due to the change in the length and tension of the string when pressed by a finger. This mismatch is also valid for other fret measures. Since Yekta presented the fret measures in his article which presents the TTAM to the international community for the first time (and he was not a tanbur maker), this

In this section, we first compare the fretting systems specified in literature to actual measurements on a tanbur. Then we describe our method to detect fret locations from recordings. Further we present the fret location estimates obtained from recordings using the same tanbur.

1 2

Only the frets within the first octave are given to save space. The author is also a lecturer on tanbur making in the oldest school of TTAM, stanbul State Conservatuary of Turkish Music.

No
1 2 3

Frequency Ratios (Yekta)

1/1 256/243 2187/2048

Fret Distance to Bridge (mm)

NY 104 98.7 97.42 96.16 95.49 95.54 93.67 92.45 87.74 86.6 93.67 92.46 87.74 86.6 85.48 84.93 83.27 82.19 79.02 78 74.02 73.06 83.27 82.19 79.03 78 74.03 73.06 AEU 104 98.7 97.42 CA 104 98.7 97.42

Pitch Names
Turkish YEGH Nim Pest Hisar Pest Hisar Western D

4 5 6 7

65536/59049 9/8 32/27 19683/16384

93.67 92.46 87.74 86.6 85.48 84.93

Dik Pest Hisar HSEYNARAN Acemairan Dik Acemairan E

8 9 10 11 12 13

8192/6561 81/64 2097152/1594323 4/3 1024/729 729/512

83.27 82.19 79.02 78 74.03 73.06 72.12 71.65

ARAK Gevet Dik Gevet RAST Nim Zengle Zengle G

14 15 16 17

262144/177147 3/2 128/81 6561/4096

70.25 69.34 65.81 64.95 64.11 63.69

70.25 69.34 65.8 64.95

70.25 69.34 65.81 64.95

Dik Zengle DGH Krd Dik Krd A

18 19 20 21

32768/19683 27/16 8388608/4782969 16/9

62.45 61.64 59.27 58.5 57.74 56.99 56.62

62.45 61.64 59.27 58.5

62.45 61.64 59.27 58.5

SEGH Puselik Dik Puselik ARGH C

22 23 24 25

4096/2187 243/128 1048576/531441 2/1

55.52 54.8 52.69 52

55.52 54.79 52.68 52

55.52 54.8 52.68 52

Nim Hicaz Hicaz Dik Hicaz NEVA D

Table 2. Fret placements given for a tanbur with 104 cm string length (as given in [23]): the tanbur of Necdet Yaar (NY), fret locations derived from the theory of Arel-Ezgi-Uzdilek (AEU) and Cafer Ans fretting system. The number of frets for each fretting system is 65, 24 and 56 respectively. The frequency ratios presented by Yekta is also presented for the corresponding frets of the 3 fretting systems for comparison. Only the frets within the first octave are given to save space. Values are given with respect to the bridge. All grey shaded rows are the frets that do not exist in theory. Black shading: slight differences among the same frets. Dark grey: Additional frets of NY. Light grey: Additional frets of CA and NY. compared with the given fretting systems. The fret measures of tanbur used in the experiment can be found in 3.1. The present fretting/tunning system in comparison to measured fret locations on a tanbur Table 3 (indicated in the first column as M). Given the A well-known young tanbur player, the third author of the string length, it is also easy to find the fret measures of paper is involved to the study for experimental reasons. AEU system by simply applying the frequency ratios of The tanbur used in the experiments is measured in terms of Yekta. This fret measures are also given in the table as fret placements and string length. These measures are AEU.

No 1 2 3 4 5 6 7

8 9 10 11 79.3 74.9 12 74.4 13 73.4 72.6 72.3 14 15 70.6 67 16 66.3 17 65.5 64.7 63.9 18 62.9 19 20 59.7 21 58.4 56.9 56.4 22 55.6 23 55 54.4 24 25 53.2

M 104.7 99.8 98.6 96.5 94.6 93.8 88.9 87.8 86.4 85.4 84.6 83.6

AEU 104.7 99.4 98.1 94.4 93.1 88.3 87.2

error 0 0.4 0.5 0.2 0.7 0.6 0.6

83.9 82.7 79.6 78.5 74.5 73.5

0.3 0.3 0.8 0.1 0.1

70.8 69.8 66.3 65.4

0.2 0.8 0 0.1

62.9 62 59.7 58.9

0 0 0.5

No M 25 53.2 26 50.2 49.7 27 49.2 48.7 48 28 47.3 29 44.7 30 44.1 31 43.7 43.2 42.8 42.3 32 41.7 33 34 39.8 35 38.1 36 37.6 37 36.9 36.3 38 35.6 39 33.6 40 33.1 41 32.6 32.1 42 31.7 43 30.1 44 29 45 28.5 46 27.7 47 48 49 26.8

AEU 52.35 49.7 49

error 0.85 0.5 0.2

3.2. Estimation recordings

of

fret

placements from audio

47.2 46.5 44.2 43.6

0.1 1.8 0.1 0.1

41.9 41.4 39.8 39.3 37.3 36.8 35.4 34.9 33.1 32.7 31.5 31.0 29.9 29.5 28 27.6 26.5 26.2

0.2 0 1.2 0.3 0.1 0.1 1.3 0 0.1 0.2 0.9 0.9 1.0 0.3

0.6

55.9 55.2 53.1 52.4

0.3 0.2

0.8

Table 3. The measured (M) fret placements of the tanbur used in the experiment and the calculated (AEU) fret placements from the Yekta system for the tanbur used with 104.7 cm string length. Only the frets within the first octave are given to save space. Values are given with respect to the bridge. Grey shading: the frets either do not exist in the theory or in practice. The first observation is that although the tanbur used in the experiment has 56 frets as one of the tanbur type reported by An [23], most of these frets do not match with the theoretical system of AEU. Figure 5 presents the visual representation of the theory and practice in this sense. Although the figure gives the impression that the practice is almost similar to the theory, it can be seen from the Table 3 that the most important frets of a tanbur, neva, rast, dgah, argah, hseyni airan, hseyni, gerdaniye, muhayyer, tiz argah and tiz neva considerably diverge from theory with an average of 1cm which is not tolerable for the fret placements of tanbur.

Recently Gedik and Bozkurt [15] comprehensively discussed the challenges in computational studies on TTAM. These problems can be briefly listed as follows: Pitches demonstrate distributional characteristics instead of fixed frequency values. There is no reference frequency such as A4 = 440 Hz in western music. There is no reliable theory to consider as a reference as in western music The number of pitch intervals and their values are still hot topics of discussion. Bozkurt [7] presented a solution to the first problem by developing pitch-frequency histogram representation of TTAM. Related with the second problem, we have shown in our previous publications [7, 15] that automatic tonic detection can be very reliably achieved via template matching. In addition, by aligning pitch histograms with respect to tonics, automatic tuning analysis can be reliably performed for a given collection of recordings. Finally, instead of unreliable theoretical information, data-driven models were proposed and succesfully used for the final two problems [7, 15]. In this study, the algorithms are extended to achieve automatic detection of fret locations. The steps of the method applied to audio recordings for the estimation of fret placements are listed below: 1. Representation of a given recording as pitchfrequency histogram. 2. Automatic computation of tonic of the given recording by using the pitch-frequency histogram. 3. Mapping the pitch-frequency histogram to the fretboard of the tanbur using effective string lengths. a. Pitch-interval values are converted to frequency ratios. b. Frequency ratios are applied to a tanbur with given effective string lengths to find the fret locations. Since the first 2 steps of the method listed above are comprehensively considered in our previous papers, we focus only on the 3rd step of the method. The main contribution of this study from the computational point of view is to assign the string length in an adaptive manner (instead of considering it constant throughout the fretboard as applied in the theoretical approaches) for fret/stretch compensation purposes. Due to complexity of fret/stretch compensation problem, it is a common practice for instrument makers to detect the octave fret locations using harmonics and then compute and effective string length to be used in computations.

measured 0.14 yekta 0.13 0.12 0.11 0.1 0.09 0.08 25 30 35 40 45 50 fret distances to nut (cm) 55 60 65 70

Figure 5. Fret placements of tanbur measured, and derived from the Yekta-AEU system. The resulting fret estimations which lay within the dgah region (between 0 cm and 34.1 cm) with a margin of 5 cm We follow a similar approach and use some reference toward the bridge, are found as the fret placements for the fret measurements as effective string lengths. These first fret region. reference frets are the ones which can be easily tuned by ii. The effective string length, 103 cm, corresponding to the tanbur player manually using the harmonics: neva (1st the neva region and the equation (1) are used to estimate octave), dgah (1st perfect fifth), muhayyer (2nd perfect the fret placements. The resulting fret estimations which fifth) and tiz neva (2nd octave) with reference to yegah. lay within the neva region (between 34.1 cm+5 cm and Therefore the measured 4 reference frets of a given tanbur 51.5 cm) with a margin of 5 cm toward the, are found as with reference to nut are used as an input to the system. As the fret placements within the second fret region. a result the algorithm is applied as if there are four fret iii. The effective string length, 103.65 cm, regions on tanbur: The four fret regions are the dgah corresponding to the muhayyer region is used and the region (between the frets yegah and dgah), the neva equation (1) are used to estimate the fret placements. The region (between the frets dgah and neva), the muhayyer resulting fret estimations which lay within the muhayyer region (between the frets neva and muhayyer) and the tiz region (between 51.5 cm +5 cm and 69.1 cm) with a neva region (between the frets muhayyer and tiz neva). margin of 3 cm toward the bridge are found as the fret Given a recording and reference fret measures, the fret placements within the third fret region. estimation algorithm is first applied by using the first iv. The effective string length, 103.9 cm, corresponding reference fret region and the corresponding effective string to the tiz neva region and the equation (1) is used to length. Thus the frets which lay within the first reference estimate the fret placements. The resulting fret estimations fret region are found. Iteratively, the other reference fret which lay within the tiz neva region (between 69.1 cm regions and corresponding effective string lengths are used +5cm and 77.9 cm) with a margin of 1 cm toward the in the same way and estimated frets are found for each bridge are found as the fret placements within the third reference fret region. In summary, each estimated fret fret region. (distances to nut) is checked in order to find the reference Consequently, the estimated fret placements of a fret region of it. Therefore different effective string lengths recording performed by the tanbur are found by corresponding to each reference fret region are being used concatenating the estimated frets for each of the four fret to estimate frets. The details of the algorithm are presented regions. below: i. The effective string length, 102.3 cm, corresponding to the dgah region is used. The frequency ratios are 3.3. Tests converted to fret placements for this effective string length according to the equation below: 16 improvisational (taksim) performances of the player from 8 modes (makam) are recorded for the experiment. ESL As a result we tried to estimate the fret placements of the fret E = ESL , (1) tanbur from the 15 recordings and evaluate the success of Rf the system based on the measured values of the fret fretE : estimated fret placements, ESL : effective string length, Rf : placements. The distribution of 15 recordings are as frequency ratios follows: 2 improvisation/taksim recordings from each

mode/makam hicaz, rast, neva, nihavend, saba, uak, suzinak and one recording from the mode/makam hseyni. In order to evaluate the algorithm, the difference of the estimated fret distances and the measured fret distances are considered as error. The overall mean error of the proposed method for the 15 recordings is found as 0.2 cm where the mean error is also found to be 0.2 cm for each mode (makam). In order to demonstrate the success of the algorithm more clearly the error of our adaptive approach is compared with the error of the approach where constant string length is used, as suggested both in theory and production. In other words the frets are estimated again by our algorithm but by using the actual string length, 104.7 cm as constant. The overall mean error of the method for constant string length for the 15 recordings is found as 0.7 cm where the mean error ranges between 0.6 and 0.8 for the 8 modes (makam). Thus the adaptive approach is found more successful than the constant string length approach. Figure 6 enables us to compare visually the two approaches, constant and adaptive approaches, by presenting the fret placements estimated from 2 uak recordings in comparison to measured frets. The constant string length approach and the adaptive string length approach are presented in Figure 6.a and 6.b., respectively. The theoretical constant string length approach again fails,
0.2 dugah neva

since the reference frets explicitly do not match the estimated frets, as can be seen from Figure 6.a. While the adaptive approach gives very close fret estimation values, the constant approach does not give reliable fret estimation values even for the most important frets, dgah, neva and muhayyer. Consequently, the theoretical constant string length approach is unreliable also for the estimation of the fret placements from audio recordings. Finally Table 5 enables us to compare the two approaches numerically by presenting the frets estimated from the same two uak recordings, in comparison to the fret placements measured. Firstly it is possible to see the considerable deviation of the most important frets (light gray shaded) estimated by the theoretical constant string length approach. Secondly, success of the theoretical approach is worse for the most frequently used fret region, the main octave of the tanbur as shown by dark grey shading. While the mean error of the adaptive approach is 0.15 cm for this region, the mean error of the theoretical approach is 0.75 cm. Therefore our approach is considerably better than the constant string length approach for the most frequently used fret regions of the tanbur.

estimation of fret placements for constant string length

measured frets reference frets ussak1 muhayyer

0.15

ussak2

0.1

0.05 25

30

35

40

45 50 fret distances to nut (cm)

55

60

65

70

estimation of fret placements for effective string lengths 0.2 dugah neva

measured frets reference frets ussak1 muhayyer

0.15

ussak2

0.1

0.05 25

30

35

40

45 50 fret distances to nut (cm)

55

60

65

70

Figure 6. Mapping the fret placements estimated from 2 uak performances on the fretboard of tanbur a) theoretical constant string length approach, b) adaptive effective string lengths approach. Only the main octave performed in TTAM is presented.

Meas. frets
96.5 93.8 88.9 86.4 84.6 79.3 74.9 70.6 67 59.7 56.9 53.2 49.7 47.3 44.7 42.8 39.8 38.1 37.6 35.6 33.6

Recording : Uak#1 Ref. Constant SL Adaptive SL frets Estimated error Estimated error frets frets
96.2 93.3 88.7 86.1 84.3 79.1 74.9 70.6 66.8 59.9 56.7 53.4 49.5 47.3 44.8 42.9 39.9 37.9 37.3 35.8 34 0.3 0.5 0.2 0.3 0.25 0.2 0 0 0.2 0.2 0.2 0.2 0.2 0 0.1 0.1 0.1 0.2 0.3 0.2 0.4 0.2 E1 96 93.1 88.3 85.7 83.8 78.5 74.2 69.8 66 59.1 55.9 52.6 48.6 46.7 44.2 42.3 39.3 37.3 36.6 35.1 33.3 0.5 0.7 0.6 0.7 0.8 0.8 0.7 0.8 1 0.6 0.7 0.6 1.1 0.6 0.5 0.5 0.5 0.8 1 0.5 0.3 0.7

4.

DISCUSSION, CONCLUSION AND FUTURE WORK

G1 A1

D1 E2

G2 A2

Mean error Recor Meas. frets

93.8 88.9 83.6 79.3 70.6 67 65.5 56.9 53.2 47.3 44.7 39.8 38.1 37.6 35.6 33.6

Recording : Uak#2 Adaptive SL Ref. Constant SL frets Estimated error Estimated error frets frets
93.3 88.7 84 79.1 70.6 66.8 65.2 56.7 53.2 47.3 44.8 40.1 37.9 37.3 35.8 34 0.5 0.2 0.4 0.2 0 0.2 0.3 0.2 0 0 0.1 0.3 0.2 0.3 0.2 0.4 0.2 E1 93.1 88.3 83.5 78.5 69.8 66 64.5 55.9 52.3 46.7 44.2 42.3 39.4 37.3 36.6 35.1 0.7 0.6 0.1 0.8 0.8 1 1 1 0.9 0.6 0.5 0.5 0.4 0.8 1 0.5 0.7

In this study we discussed one of the most neglected and challenging issues in the literature from a music theory and computing perspective: the fret locations of tanbur. We first discussed the importance of the instrument with in various perspectives: the music theory, practice and tanbur production. Secondly, we presented the main difficulties within the context of the divergence of theory and practice. We summarized the literature of tanbur fretting and showed that there is a lack of reliable information in the domain. We introduced our method for estimation of fret placements from audio recordings and presented our test results which show that the method is indeed reliable. As a result a computational study for the study of tanbur frets is presented for the first time in the literature. Furthermore the current fretting systems proposed in theory and used in production were compared for the first time. However there are various lacks of the study which we hope to fill out in the future studies. First of all an ethnomusicological approach would supply invaluable contribution for the role of tanbur in performance. Secondly frets estimated from the recordings of the tanbur masters could be tested by tanbur players for a more complete evaluation. Nevertheless we hope that our study would pave the way for the colleagues from various disciplines interested in tanbur to contribute to the current state-of-art of the research on tanbur. 5. ACKNOWLEDGEMENT

G1 A1

This work is supported by Scientific and Technological Research Council of Turkey, TUBITAK (Project No: 107E024). 6. REFERENCES

D1 E2

G2 A2

[1] Hassan, S. Q. et al. ''Tanbr.'' Grove Music Online. Oxford Music Online. 22 Mar. 2009 http://www.oxfordmusiconline.com/subscriber/article /grove/music/52071 [2] Popescu-Judetz, E. Prens Dimitrie Cantemir, Trans. S. Alemdar, Pan Yay., stanbul, 2000. [3] Gedik, A. C. and Bozkurt, B. ''Evaluation of the Makam Scale Theory of Arel for Music Information Retrieval on Traditional Turkish Art Music''. Journal of New Music Research, (In press). [4] Bozkurt, B., Yarman, O., Karaosmanoglu, M.K. and Akkoc, C. ''Weighing Diverse Theoretical Models On Turkish Maqam Music Against Pitch Measurements: A Comparison Of Peaks Automatically Derived From

Mean error

Table 5. Measured and estimated placements of frets for 2 uak recordings, demostrating the success of adaptive and constant string length appproaches. Values are given as distance to bridge. SL: string length. Light grey shading: the most important frets of the tanbur. Dark grey shading: the most frequently used frets of the tanbur.

Frequency Histograms With Proposed Scale Tones''. Journal of New Music Research, (In press). [5] Akan, E. Tanbur metodu. alar Musiki Yaynlar, Istanbul, 2007. [6] Erkut, C. and Vlimki, V. ''Model-Based Sound Synthesis of Tanbur, a Turkish Long-Necked Lute''. Proc. Of the International Conference on Acoustics, Speech and Signal Processing, vol. 2, Istanbul, Turkey, pp. 769-772, June 5-9, 2000. [7] Bozkurt, B. ''An automatic pitch analysis method for Turkish maqam music''. Journal of New Music Research, 37(1), 1-13, 2008. [8] Yarman, O. Kantemir Fontu, http://www.ozanyarman.com/kantemirfontu.html, Access date: 15th Sept. 2009. [9] Racy, A. J. "Historical Worldviews of Early Ethnomusicologists: An East-West Encounter in Cairo, 1932," In Ethnomusicology and Modern Music History, eds. Stephen Blum, Philip V. Bohlman, and Daniel M. Neuman, p. 6891, Urbana: University of Illinois Press, 1991. [10] Feldman, W. Z., "Ottoman music". Grove Music Online. Oxford Music Online. http://www.oxfordmusiconline.com/subscriber/article /grove/music/52169, Access date: 10 Aug. 2009. [11] Yavuzolu, N. 21. Yzylda Trk Mzii Teorisi. Istanbul: Pan Yaynclk, 2008. [12] Thomas, A. E. "Intervention and reform of Arab music in 1932 and beyond". Conference on Music in the world of Islam, Assilah, 2007. [13] Marcus, S. "The interface between theory and practice: Intonation in Arab music". Asian Music, Vol.24, No.2. pp.39-56, 1993. [14] Mus2. URL: http://www.musiki.org/, Access date: 15th Sept. 2009. [15] Gedik, A. C. and Bozkurt, B. "Pitch frequency histogram based music information retrieval for Turkish music". Signal Processing, doi:10.1016/j.sigpro.2009.06.017 [16] Martin, D. "Innovation and the Development of the Modern Six-String Guitar". The Galpin Society Journal, Vol. 5, pp. 86-109, Jul., 1998. [17] Fletcher, N.H. and Rossing, T.D. The Physics of Musical Instruments. New York: Springer-Verlag, Second Edition 1998.

[18] Hopkin, B. Musical Instrument Design: Practical Information for Instrument Design. See Sharp Press, 1996. [19] Yekta, R. Trk Musikisi. (Translated from French by O. Nasuholu ), Pan Yay., stanbul, 1986. [20] Subramanian, K. S. "An Introduction To The Vina". Asian Music, Vol. 16, No. 2, pp. 7-82, Spring, 1985. [21] Hassan, S. Q. "The Long Necked Lute in Iraq". Asian Music, Vol. 13, No. 2 (1982), pp. 1-18. [22] ksz, M. A. Trk musikisinde tanbur saznn geliimi. Marmara niversitesi, PhD thesis, stanbul, 1998. [23] An, C. Tanbur Yapm Sanat ve Sanatlar. Bilgi Basmevi, stanbul, 2002.