Timbral and Melodic Characteristics of the Persian Singing Style of Avaz HAMA JINO BIGLARI

Transcription

1 Timbral and Melodic Characteristics of the Persian Singing Style of Avaz HAMA JINO BIGLARI Master of Science Thesis Stockholm, Sweden 2012

2 Timbral and Melodic Characteristics of the Persian Singing Style of Avaz HAMA JINO BIGLARI DT217X, Master s Thesis in Music Acoustics (30 ECTS credits) Single Subject Courses Royal Institute of Technology year 2012 Supervisor at CSC was Johan Sundberg Examiner was Sten Ternström TRITA-CSC-E 2012:026 ISRN-KTH/CSC/E--12/026--SE ISSN Royal Institute of Technology School of Computer Science and Communication KTH CSC SE Stockholm, Sweden URL:

3 Timbral and melodic characteristics of the Persian singing style of Avaz Abstract The floridly ornamented Persian singing style called avaz was studied, focusing on melody characteristics in melismatic pitch transitions, phonation types, and formant to harmonic relationships. Audio and EGG signals were simultaneously recorded from a professional male tenor range singer, who sang a Persian avaz song, scales, rapid tone reiterations, and alternations between two neighbouring tones. Voice source parameters and formant settings (F1 & F2) were measured from inverse filtering of the audio signal, using the custom made DeCap and S-naq (Svante Granqvist) and the commercial Soundswell softwares. Fundamental frequency F0 was measured from the EGG signal using the Soundswell CORR tool. In the melismatic embellishments, the pitch transition between melody tones of the modal register were sung via remarkably short falsetto episodes in which F0 quickly jumped up to a peak in order to immediately dive towards the next modal tone. Being produced in this way, tone repetitions and alternations were sung with continuous phonation and had identical voice source data in their modal melody tones. Moreover, for most vowels, the singer tuned F1 to H2 (F0 * 2) and sometimes also F2 to some higher harmonics, in his higher voice range, i.e. above about Bb4 (235 Hz, approximately). These findings are discussed in relation to western operatic formant strategies and some melodic ornaments of early Italian Baroque singing. Klangliga och melodiska egenskaper inom den persiska sångstilen avaz Sammanfattning Den rikligt ornamenterade persiska sångstilen avaz studerades med fokus på fonationstyper, melismatiska växlingar mellan meloditoner, samt förhållandet mellan formanter och övertoner. Audio- och EGGsignaler spelades in samtidigt, med en professionell manlig tenorsångare som sjöng ett stycke persisk avazsång, skalor, snabba tonupprepningar, samt alterneringar mellan två toner. Olika parametrar hos röstkällan samt formantpositionerna (F1 & F2) mättes genom inversfiltrering av audiosignalen med hjälp av de skräddarsydda mjukvaruprogrammen DeCap och S-naq (Svante Granqvist) samt det kommersiella programmet Soundswell. Med verktyget Soundswell CORR kunde grundtonsfrekvensen F0 fås ur EGGsignalen. I de melismatiska ornamenteringarna sjöngs växlingar mellan modalresgistrets meloditoner via anmärkningsvärt korta falsettepisoder där F0 snabbt hoppade uppåt till en höjdpunkt för att omedelbart dyka neråt mot nästa modalton. Sålunda framställda tonupprepningar och alterneringar sjöngs med kontinuerlig fonation och hade identiska röstkällor i modaltonerna. I övre delen av sitt omfång, dvs ovanför Bb4 (ca 235 Hz), placerade sångaren F1 på H2 (F0 * 2) för de flesta vokaler, och ibland placerades även F2 på högre övertoner. Dessa observationer diskuteras i förhållande till formantstrategier inom västerländsk operasång samt vissa melodiska ornament inom tidig italiensk barocksång.

4 1 Introduction Aim of the study Voice Science The Basics Inverse Filtering Voice Source Parameters Research on Iranian singing Repetitive Melodic Ornaments Method The subject Protocol Recording Analysis Definitions and scope Results The Fundamental Frequency (F0) Voice source Formants and Spectrum Harmonics The Voice Assessment Discussion Discussion on F Discussion on Voice Source Discussion on Formants Conclusions Acknowledgements Bibliography... 30

5 1 Introduction 1.1 Aim of the study This study aims to investigate some acoustic aspects of a type of traditional Iranian singing. More specifically, it aims to describe characteristic timbral and melodic features of the ornamental singing style known as Persian avaz, with focus on the following three areas: 1) Characteristic features of the melodic line, focusing on the fundamental frequency in melismatic embellishments 2) Voice source characteristics 3) Formant frequencies and their relationship with the harmonics 1.2 Voice Science The Basics When singing or speaking, sound is generated as the air pressure is repeatedly disturbed due to the oscillatory vibrations of the vocal folds. A vibration cycle in the glottis is often seen as divided into different phases. During glottal closure we have the closed phase which is followed by the opening phase, being defined as the tiny time slot that elapses while the glottis is moving towards being opened. Thereafter, the open phase is entered, during which there is transglottal airflow. The closing phase starts when the glottis moves towards being closed again. And finally, the glottis returns to the closed phase. The act of closing the glottis and also the act of maintaining glottal closure is known as adduction. (Sundberg 2001; Vennard 1968) During the open phase in each phonation cycle, an air pulse passes by the vocal folds due to the relatively high pressure in the lungs, i.e. the subglottal pressure. This transglottal airflow is decreased, minimized or completely interrupted, depending of how complete the glottal closure is during the closed phase, i.e. depending on whether the vocal folds have full contact or not. Thus, the transglottal airflow consists of an air pulse followed by a minimized or stopped air passage phase. The number of air pulses in a certain amount of time yields the fundamental frequency (F0) of the air pressure disturbances being radiated from the lips. The transglottal airflow is the voice source, i.e. the oscillatory component, in singing, and it provides an approximate picture of the vocal fold motion. (Sundberg 2001; Hall 2002) Due to the complex and abrupt motion of the vocal folds, the resulting vibrations in the air have not only the frequency of the vocal fold vibration, but also higher frequencies that are whole number multiples of F0, also known as partials, overtones and harmonics. That is, due to the opening and closing movements of the vocal folds and the collisions involved in the muscular-membranous motion, the air particles move with not only the F0 frequency but also with integer multiples of F0, i.e. 2xF0, 3xF0, 4xF0, etc. However, the energy level of the voice source partials is strongly decreasing, so that a frequency doubling in the partials means a decrease by 12 db. For example, when singing the tone A4 which has F0=440 Hz, the overtones in the harmonic spectrum of the voice source will be 880Hz, 1320Hz, 1760Hz, 2200Hz, etc. Then the air vibration, i.e. the air pressure disturbances, at 440 Hz will be 12 db stronger than at 880 Hz, which in turn will be 12 db stronger than the 1760 Hz harmonic, etc. (Sundberg 2001 & 1989; Hall 2002) Above the glottis we have sound; the disturbances of air pressure caused by the air pulses above the glottis are sound waves. On their way to the lips, the sound waves hit the inner walls of the space known as the vocal tract which encompasses the throat as well as the mouth and the nasal cavities. The sound waves are thus reflected or refracted, so that parts of the waves travel backwards towards the glottis in order to be reflected outwards again. When a returned wave belonging to a higher partial reaches back to the glottis in order to be reflected outwards again, it might accompany the next wave being produced by the transglottal air pulse, so that the energy of the reflected wave will be added to the energy of the newly produced wave. 1

6 Thus, the reflections in the vocal tract will result in that some partials are strengthened compared to their neighbouring partials. This phenomenon is known as resonance; it occurs whenever returning (reflecting) waves resonate synchronously along with new waves emerging from the oscillator. For a given voice source oscillating at frequency F0, the shape of the vocal tract at a certain point of time during phonation results in a specific set of partials being strengthened. In other words, the vocal tract functions as a filter that enhances some frequencies of the voice source partials so that the vibrations at those frequencies become stronger due to the addition of the returning waves to the new ones. Also the neighbouring partials are somewhat strengthened, while partials further away from the resonance frequencies remain unaffected. (Sundberg 2001 & 1989; Hall 2002) The specific frequencies that a given vocal tract having a certain shape is ready to enhance are known as formants 1. The formants are more or less evenly distributed over the harmonic spectrum; a straight tube, i.e. a cylinder produces resonances at 500 Hz, 1500 Hz, 2500 Hz etc, and the vocal tract more or less does the same. However, the positions of the formants within each of the 1 khz segments of the human voice spectrum vary, so that for example the two lowest formants can be at 300 and 2500 Hz. The two and sometimes three lowest formants determine which vowel is being pronounced. The first formant (F1), which roughly varies in the range Hz, is controlled by the amount of jaw lowering, so that a small opening produces a low F1 while a large opening gives a high F1. The second formant (F2) is controlled by the tongue position, so that having the tongue pushed forward produces high F2 while drawing the tongue back to the throat gives low F2. The third formant (F3) is usually related to the lip rounding and also to the cavity between the tongue and the lips. The rule of thumb is to count on 8-9 formants up to 8 khz in the harmonic spectrum. (Sundberg 2001) Thus, the harmonic spectrum of the resulting sound wave being radiated from the lips differs from the spectrum of the voice source in that the formants strengthen some partials. The timbre, i.e. the sound quality, is then determined by the combination of the voice source and the formant setting. One of the questions studied in this thesis is whether F1 and F2 systematically and intentionally are set equal or close to any harmonics for a given vowel, and if so, whether the formant sticks to that harmonic even when F0 and thereby also the higher partials increase or decrease. A formant is considered as tuned to a harmonic when it is on or close to the harmonic. When F0 is low, e.g. at about 100 Hz, the chances are much higher that at least some formant frequencies match some harmonics, whereas the same thing should not be as easy to find for high F0 values, e.g. in the upper tenor octave and in soprano ranges Inverse Filtering As mentioned above, the vocal tract ensures filtering of the voice source through the formants, so that the air pulses leaving the lips differ from the voice source in that they have been filtered. This means that if the filtering could be undone, the air stream at the lips would be the same as at the glottis, i.e. the voice source. But the voice source can be unfolded even from the filtered air steam being radiated. Instead of undoing the filtering, the filtered air stream is taken as input in order to be filtered with a filtering function that is the opposite of the one given by the vocal tract. This process is known as inverse filtering, and it requires that the singer sings into a mask, through which all air passes so that the air flow can be measured. (Rothenberg 1973) A similar and almost equivalent way of doing inverse filtering is by using the audio signal (recorded via microphone) instead of the air flow. The difference is that the air flow in the mask shows if there is any air leakage during phonation, i.e. if part of the transglottal air is unphonated which makes the singer sound breathy, whereas the recorded audio signal misses such leakage. (Sundberg 2001) The inverse filter function for a given formant frequency is nowadays calculated by software programs, such as DeCap which is made by Svante Granqvist. A point of time in the signal captured from either the air flow or the radiated sound wave (hereafter audio signal will be assumed) is selected and the search for the formant frequencies is started. F1 and F2 and sometimes also F3 can be located within their typical 1 Some scholars, for example in France, use different terminology, and call such frequencies resonances. 2

7 ranges for the vowel being sung, and the higher formants can be distributed about 1 khz apart over the spectrum range. When the user selects a formant frequency, the software program reverse filters the radiated audio signal so that the harmonic spectrum of the audio signal is modified accordingly. Then the position of each formant is iteratively changed back and forth by the user until the resulting harmonic spectrum is reasonably even and decreases with about 12 db per octave. Another criterion is that the resulting flow glottogram representing the voice source must have a reasonable shape, which often means that it must contain a pulse (of air) followed by a closed phase with less or no air flow, preferably with no ripples. Yet another criterion is to let the software program differentiate the EGG signal, so that the EGG derivative (degg) can be used to find a more reliable flow glottogram by fine tuning the selected formant frequencies. However, this requires that the beginning of the closing phase in the flow glottogram is recognizable in that the decrease rate of curve suddenly slows down. As a consequence, the final set of formant frequencies selected should be seen as approximate values. The flow glottogram sometimes may contain ripples in its closed phase. Moreover, the beginning of the closing phase may lack a sudden change and will therefore be impossible to locate, and a closing phase may be followed by an opening phase in which the curve increases in order to decrease again before rising towards the next pulse peak. Therefore, also the flow glottogram produced through inverse filtering can sometimes be hard to verify, which in turn can jeopardize the reliability of the obtained voice source Voice Source Parameters Data achieved from the air pulse amplitude along with the closed phase quotient reveal the amount of adduction (how hard the vocal folds are pressed together) and the subglottal air pressure level. These parameters in turn determine the phonation type, as there are different phonation types along a continuous scale extending from low adduction and low subglottal pressure to very high values for both parameters. Sundberg describes pressed phonation in terms of high adduction in the vocal folds and as being related to high position of the larynx and high subglottal pressure. He also states that in pressed phonation the maximum width of the glottis opening, i.e. its horizontal vibration amplitude, is at minimum. And meanwhile, the closed phase, i.e. the time when glottis is closed, is maximized, which of course means that the open phase has its shortest duration (Sundberg 2001:85). The difference in SPL between the first and second partials determines the air pulse amplitude and affects the radiated sound timbre. The first partial is indeed the same as F0, but in this context it is denoted as H1, and the second partial is H2. Therefore, the difference in SPL is denoted H1-H2 and is measured in db. QClosed (Closed Quotient) is another recurring voice source parameter, denoting the ratio of the closed phase to the cycle period. The closed phase denotes the duration of glottal closure, which can be either complete (the vocal folds have full contact so that the air flow through glottis is stopped) or partial (there is come air leakage as the vocal folds are not in full contact). The amplitude of air pulse amplitude in a flow glottogram shows the momentary air volume passing through the glottis. MFDR (Maximum Flow Declination Rate), given by the negative peak amplitude of the transglottal airflow derivative, shows the point of maximum closing speed of the vocal folds. MFDR turns out to be interesting to measure, as it determines the SPL being produced. It is also used in NAQ (Normalized Amplitude Quotient), which shows the amount of adduction and is defined as the pulse amplitude of the transglottal airflow curve divided by the period and MFDR. (Björkner et al 2006) Research on Iranian singing Some research has been conducted on different aspects of the various styles and traditions found in the music of Iran, both by Iranian and also by Western as well as East Asian scholars and musicians, but the subjects of those studies have mostly been other than technical or acoustic-physiological questions on singing. Simms (1996) transcribed and analyzed commercial recordings of the master of Persian avaz, Mohammad Reza Shajarian. He found the melodic repertoire to constitute a centonic style, in which some twenty basic blocks, each consisting of a few notes, recur through the quasi-improvised style. During 3

8 (1984) and Miller (1999), both lived in Iran where they studied different music styles, primarily the Persian radif, which are transcribed canonizations of the traditional melodic repertoire. However, their presentations of the singing styles do not include acoustic-physiological investigations. Nevertheless, there are two cases of exception regarding acoustic-physiological investigations of Iranian singing. Margaret Caton studied recordings of male Iranian singers of Persian avaz as well as those of folk music and she published an article at UCLA in 1974 (see Caton 1974). Caton presented time domain spectrograms of a commercial avaz recording, focusing on the takiyah, i.e. the ms short falsetto episodes, in which F0 quickly jumped upwards to a peak in order to descend as quickly towards the next melody tone. She discussed accentuated and non-accentuated takiyah and made statements on the differences between them, such as the usage of aspiration (phonation initiated with the phoneme [h]) and their strong and weak partials. She also discussed the provenience of takiyah types in Kurdistan and Azerbaijan, and she attempted to explain the glottal motion patterns for takiyah as well as the modal melody tones. (Caton 1974) Michèle Castellengo and her colleagues, among them Jean During, have investigated acoustic aspects of the falsetto episodes in Iranian tahrir. They studied the following parameters during the takiyah (without actually mentioning the term takiyah): the relationship between vowels and the sound intensity; the F0 jump interval as depending on dynamics, i.e. the relationship between the dynamics and the F0 interval between the takiyah peak and the modal tone; the F0 jump interval as depending on the starting frequency; open quotient comparisons between the takiyah and the modal tones; duration of the takiyah; and reproducibility of the takiyah peak frequency. (Castellengo et al 2009) Repetitive Melodic Ornaments Tone repetitions and alternation are considered as the two basic melodic ornaments of the late 16th century and the early Italian Baroque solo singing at the beginning of the 17th century. James Stark discusses the physiological implications of the 400 years old Italian descriptions on those ornaments in terms of modern voice science, asking which vocal folds motion patterns are physiologically (im-) possible for each type of tone sequence (Stark 1999). Those repetitive ornaments are the tremolo of Zacconi, which according to Stark described tone repetitions, as did the trillo of Caccini. Also Greenlee regarded tremolo and trillo as denoting tone repetitions, albeit some scholars would disagree, instead suggesting that tremolo and even trillo described vibrato (see Brown 1976). The ornament for alternation is the gruppo of Caccini, but also that has been open to debate and some scholars preferred to interpret even the gruppo as vibrato. (Greenlee 1985; Stark 1999; MacClintock 1976) Stark combines modern voice science with Manuel Garcia s theoretical model for the singing voice in his search for the physiological implications of Italian sources describing tone repetitions and alternations as being the two basic ornaments of the early Baroque. Stark considers rapid tone repetitions and alternations as impossible to sing with the same phonation type, i.e. with the same glottal motion pattern. In his view, rapid tone repetitions require what he labels as loose phonation, in which the arytenoid cartilages stay apart from each other, thereby creating an open triangle in the posterior part of the glottis during phonation (Stark 1999:24). Thus, there is unphonated transglottal airflow during Stark s loose phonation, which indeed is the phonation type which Sundberg labels breathy (Sundberg 2001). In other words, Stark argues that rapid tone repetitions require breathy phonation, since each repeated tone must be preceded by full glottal closure, during which the arytenoid cartilages are pushed together so that the posterior 3/8 part of the glottis will too be closed. (Stark 1999). Rapid alternations, on the other hand, are according to Stark not suitable for loose phonation, since the arytenoids cartilages would not be able to close and open quickly enough in combination with alternating pitch. Stark states that the suitable phonation mode for glottal alternation is what he calls anterior phonation, in which the arytenoid cartilages are brought together by the adductory muscles and remain fixed during phonation. In such phonation, only the anterior 3/5 (or in some cases 5/8) of the glottis length vibrates during phonation while the posterior non-membranous part remains closed (Stark 1999). Stark ascribes this kind of glottal motion to what Sundberg (2001) labels as pressed phonation, adding 4

9 that Sundberg s definition includes raised larynx position (Stark 1999). However, Stark s anterior phonation encompasses not only pressed phonation but also normal and flow phonation as defined by Sundberg (see Sundberg 2001). In Biglari (2009), the repetitive ornaments of the early Italian Baroque were discussed, and some of the descriptions in the early sources as well as their modern musicological interpretations were compared to the repetitive melodic ornaments found in Iranian singing. Since the data on Iranian singing were solely F0 and histograms for old commercial recordings, some questions remained unanswered. Margaret Caton s statements on the glottal motion during modal and takiyah phonation were the only publication that directly illustrated the vocal folds in a study on the avaz vocal style. That should suffice as a reason for the author to study also the voice source in this thesis, but the author had also touched upon Stark s discussion of the above mentioned physiological possibilities regarding early Baroque singing, which was the reason to study the voice source in tone repetitions and alternations in particular. In the 2009 study, it was shown that the phonation in the commercial Iranian recordings was continuous, so that tone repetitions did not contain any pauses. The melody tones were instead preceded by takiyah episodes, and the same pattern was observed in alternations. Although, it was not possible to determine whether the same phonation type was in use in both ornaments, it was shown that tone repetitions and alternations could be produced also in other ways than was being suggested by Stark. In other words, the BSc thesis showed that tone repetitions and alternations were sung rapidly with continuous phonation. However, while the existence of interleaving takiyah episodes overruled Stark s criteria on full glottal closure, it was still unclear if the Iranian singers sang the repeating modal melody tones with the same glottal motion pattern as the alternating modal melody tones. 1.3 Method The subject This study is based on studio recordings at KTH. The subject is a male professional singer and teacher of Persian avaz, who has learned the style the traditional way, i.e. by taking lessons during many years from some of the acclaimed masters of the art in Iran Protocol The main idea was to record an excerpt of a traditional Avaz song with free meter, containing a variety of pitches as well as of dynamics, sustained tones and typical ornamentations. In addition, scales sung on various vowels were added to the protocol so as to allow study of formant strategies, i.e. the relationships between the lower formants and spectrum partials. Prompted by the author s previous studies of melodic ornaments in commercial recordings of Persian avaz, the protocol was extended to include also isolated execution of two ornaments, namely tone repetitions, i.e. reiteration of the same melody tone, and alternations between two adjacent notes. The subject first sang the Avaz piece that he selected, which was in the mode of dastgah-e mahoor and therefore contained the steps of the major scale, or rather Ionian mode. Thereafter, when asked to sing a shortened version of the song, he repeated the introductory part. Then, the subject sang on the vowels /ɑ, ӕ, i/ an ascending scale starting from F3 (the tone F at about 175 Hz, not the third formant) and after a short pause followed by a descending scale starting on F4, i.e. at about 350 Hz. The subject also was asked to sing tone repetitions. The subject then sang two series of tone repetitions, each ending with an improvised phrase. First series: 5

10 - A sustained F4 tone on the vowel /ɑ/, about 1 second long, and followed by short repetitions of the same tone. - A sequence of short repetitions on vowel /e/ on the pitch of E4 - An improvised phrase ending containing faster repetitions on the vowel /ɑ/ on the pitches G4, F4, E3, arranged in groups of three. Second series: - Repetitions of the tone F4 on vowel /ɑ/, basically similar to phrase 1. - Sequence of short repetitions on vowel /ɑ/ and /e/ on the pitches of E4 and D4 - An improvised phrase, ascending from C4 to G4 and then descending again, where the tone on each pitch was sung twice Upon being asked to sing alternations between two adjacent tones, the subject sang three series of repeated alternations which were approximately on the following adjacent tone pairs C 4-D 4, D 4-E4, E4-F4. All the alternations were sung on the vowel /ɑ/ Recording The recordings were done in one session in a sound treated studio at KTH. The audio signal was recorded using a head mounted omni-directional microphone (TCM 110, AV-JEFE) at 10 cm distance. To calibrate the microphone, a dynamically steady tone was sung and recorded while the SPL was being measured by a level meter held next to the microphone; the difference between the recorded and the externally measured SPL could afterwards be used to adjust any recorded SPL value. Two electrode contacts belonging to a 2-channel electroglottograph (EGG) machine (Glottal enterprises, EG2) were attached to the throat of the subject. This was in order to measure the fundamental frequency in a reliable way. The EGG machine creates electric voltage at 2 MHz speed between the electrodes, so that electric current will flow through the vocal folds whenever they are in contact, thereby providing data that shows when the folds are in contact, over time. This shows the frequency of the glottal motion, which is taken to be same as the fundamental frequency. The audio and EGG signals were recorded using two channels in the commercial software program Swell, and the recordings were stored in.smp files Analysis Assessment of the Recording In order to assess how representative our recordings were for Persian avaz (in other words: how typical the Persian avaz excerpt sung by the subject sounded), a panel of experts on Persian music were asked to do an audio-visual listening test. 17 excerpts from our recordings and 21 other recordings of 11 singers (the author s studio recordings of 5 singers, and commercial recordings of 6 singers) were rated on a nongraded horizontal line denoting clearly typical at one end and clearly untypical at the other end. 10 excerpts were duplicated so that they occurred at least twice in the test, thus making 51 sound excerpts in total. The duplications covered both typical and non-typical (as perceived and purposefully selected by the author) examples as well as our recordings of the subject. The 17 excerpts of the subject covered phrases on various pitches from the avaz song, including both sustained tones and series of fast melismatic ornaments, ascending and descending scales, and also the isolated melodic ornaments for tone repetition and alternation, as shown in Table I. 6

11 Number of unique Number of Type of melody Category recordings duplications Alternation Alternation 1 0 Scale Scale 4 0 Tone repetitions Tone repetition 1 0 Persian avaz Song 11 3 Table I The types of excerpts sung by the subject included in the audio-visual voice assessment. Among the other 21 recordings listed in Table II were the author s studio recordings covering isolated tone repetitions, alternations and scales along with Persian avaz as well as some non-persian excerpts, namely typical Kurdish and Azerbaijani songs. The commercial recordings were excerpts of Persian folkloric traditions as well as Persian avaz, including episodes with tone repetitions and repeated alternations. The test participants were asked to neglect linguistic and dialectal characteristics and instead focus on timbre, the singing technique and the ornamentation style. Type of melody Category # Recordings (dupl. not incl.) # Duplications Type of recording Persian avaz song, by Persian singer Song 3 2 Commercial recordings Persian avaz song, by non-persian singer Song 1 1 Author s KTH recording Kurdish song (on free meter, like avaz) Song 2 0 Author s KTH recordings Azerbaijani song Song 2 0 Author s KTH recordings Non-avaz genres of Persian singing Song 3 1 Commercial recordings Scales Scale 6 2 Author s KTH recordings Different alternation by Azerbaijani singer Alternation 1 2 Commercial recording Alternation by Persian singer Alternation 1 1 Commercial recording Tone repetitions Tone repetition commercial recording, 1 author s KTH recording Table II The excerpts sung by other singers than the subject, either from commercial recordings or from studio recordings at KTH Acoustical Analysis The commercial software Swell was used to convert the recorded EGG signal to an F0 curve consisting of EGG frequencies versus time. In order to get a reliable F0 curve, the autocorrelation Swell tool CORR was used to derive from the EGG signal an F0 curve free of errors. The resulting data was also stored in an SMP file. F0 histograms were generated in Swell for selected sections of the F0 curves. Since both EGG and audio had been recorded by Swell and thereby were synchronized, it was possible to relate any point in the obtained F0 curve to its corresponding point in the audio signal. The recorded audio signal was opened in the freeware Wavesurfer software in order to obtain its SPL curve in db versus time. Wavesurfer was also used for the frequency domain representation of the harmonic spectra of the audio, i.e. the levels of the spectrum partials (in db) for any given point in time in the audio signal. DeCap, the custom made software program by Svante Granqvist, was used for inverse filtering. The SMP files containing both audio and EGG signals were opened in DeCap while the same file and its corresponding F0 curve were opened in Swell. The flow glottogram along with the EGG derivative as well as the voice source partials were viewed in DeCap. The DeCap and the Swell windows were synchronized through the link function in both programs, so that a selected spot in the Swell audio channel automatically redrew the flow glottogram and the other curves for the same time point in DeCap. Due to the distance of ca 27 cm between the glottis and the microphone (ca 17 cm from glottis to the lips, plus another 10 cm to the microphone), the EGG derivative in DeCap was delayed by 0.8 ms ( t = 27 cm / (35000 cm/s) = s). 7

12 The points in time to be inverse filtered were selected as result of listening experience; the author listened through every vowel sung and selected one or several points distributed over the duration of the vowel, thereby covering various colors and phases of long shifting vowels. Upon selecting a point of time in the audio channel in Swell, the formants were set one by one in DeCap and they were adjusted until the following criteria were fulfilled, basically in the following order: - The shape of the flow glottogram became satisfactory, which in most cases meant that the closed phase was as ripple-free as possible. - The beginning of the closed phase in the flow glottogram and the negative peak of the EGG derivative coincided. - The partial peaks in the harmonic spectrum descended more or less evenly - The two lowest formants were in accordance with the vowel sung. For each inverse filtered point of time, eight or nine formants, i.e. F1-F8/F9, were set in the partial frequency range up to 8 khz, and the resulting inverse filtered flow glottogram as well as the formant frequencies were stored. Thereafter, the two lowest formants were plotted versus F0 so as to reveal their relationship with the spectrum harmonics appearing at integer multiples of F0. For the voice source characteristics to emerge, voice source data derived from inverse filtering needed to be analyzed. Although some voice source parameters such as H1-H2 and QClosed are visually measurable in DeCap, it was more convenient to use another software program of Svante Granqvist for that task, namely S-naq. Each SMP file, containing the inverse filtered flow glottogram for a specific time coordinate, was opened in Swell and linked to S-naq, so that the EGG derivative and the flow glottogram could be made visible in the S-naq window. Finally, the period as well as the beginning and the end of the closed phase were marked. Then the voice source parameters NAQ, H1-H2, and QClosed were stored. In this way the vowels /ɑ, æ, i/ were analyzed for all modal pitches in the scale. The same parameters were compared also for the vowels /ɑ, æ, e, i/ in the avaz song recordings. In addition, QOpen was compared between modal and falsetto registers within series of tone repetitions. 1.4 Definitions and scope Some of the F0 axes in the diagrams presented in this thesis show the absolute frequency in Hz. However, in most diagrams the frequency is represented relatively, as an interval above certain frequency, which usually is 220 Hz, i.e. the tone A2. In such cases, the unit is written as [ST above A2], or [ST rel Hz], where the abbreviation ST denotes semitones. Subglottal pressure (P sub) was not measured during the recording, although the aim initially had been to measure it. While the calibration of the SPL was done as to compensate for the absence of P sub, the author has decided to include P sub measurement in future studies. All the db values for SPL are relative values, with 20 Pa being the reference level of pressure. Tone repetitions and alternations were included in the protocol partly in reference to the author s previous discussions on those repetitive melodic ornaments. In a BSc thesis in musicology which was done in 2009 at Uppsala University, the author analyzed F0 curves and histograms for excerpts of commercial recordings of Persian avaz as well as similar styles of singing from the Kurdish and Azerbaijani traditions. Nevertheless, some questions remained unanswered in the BSc thesis, and they will be addressed in this thesis. Therefore, some of the issues discussed in the BSc thesis will be briefly presented in this thesis so that the central (unanswered) questions can be discussed in light of the new data, including the idea that Iranian and some other Eastern singing techniques appear to be similar to Italian Baroque ornaments, as described in the historic sources. 8

13 Some musicological literature on early Italian singing is referred to in this thesis, mostly regarding issues discussed in the author s BSc thesis where the early Barocque vocal ornaments for tone repetition and alternation were discussed. Both the historic sources (Vicentino, Confurot, Bovicelli, Zacconi, Caccini) and the musicological literature discussing them (Greenlee, MacClintock, Galliver, Brown, Stark) can be neglected by a reader who would be mostly interested in the main questions of this thesis, namely some melodic and timbral characteristics of Persian avaz in terms of F0, voice source and formant settings. Caton s (1974) mentioning of genre-related, ethnical and geographical characteristics for various types of takiyah has not been considered in this thesis, which is about the Persian singing style of avaz, specifically. Neither does this thesis deal with Caton s dividing of takiyahs into accentuated and non-accentuated based on differences in aspiration (phonation initialized with the phoneme [h]) and partial strength in the spectrogram. The author would be interested in examining different kinds of takiyah in future studies. 9

14 [a:] [o:] [ae] [ i: ] 2 Results 2.1 The Fundamental Frequency (F0) Modal and falsetto registers were used both in the avaz song and in the scales and the repetitive ornaments, i.e. alternation and repetition. All melody tones were sung in modal register, while melismatic transitions from one melody tone to the next were sung in falsetto, whereby F0 quickly jumped up to a frequency peak well above the next melody tone and then quickly dived towards it. These short falsetto episodes are known as takiyah, which literally means leaning/support, i.e. appoggio (Caton 1974; in Simms 1996 spelled tekye in accordance with modern Farsi pronunciation). Some examples of takiyah are shown in the F0 curve in Fig Caton studied the takiyah as the basic stone in ornamentation, and she presented different kinds of takiyah which she ascribed to different genres. Also this thesis follows Caton s idea, focusing on takiyah rather than tahrir 2. F0 [ST rel A2] F0 for melismatically ornamented melody in Persian avaz Takiyah (lit:appoggio) Tahrir = melismatic ornamented melody (here with 17 takiyahs) Time [ms] Freq Pitch transition from one modal tone to the next (mostly) goes via takiyah Continuous phonation Register breaks Takiyah, ornament, falsetto Melody tone, modal register Peak frequencies vary Not a tone; No distinct pitch Interval not significant; can be off scale Fig (Left): Examples of takiyah and tahrir marked o the F0 curve for melismatically ornamented melody in avaz song. (Right): Schematic view of pitch melismatic pitch transition in avaz. The modal melody tones in the avaz song covered the pitch range G3-A b 4. The lowest takiyah peak was 150 cents above G3 and occurred between the lowest modal tones F3 and G3 in the scales. The highest takiyah peak was at A4 in the avaz song. Occasionally, takiyahs preceding the same modal melody pitch had varying peak frequencies, especially in the scale recordings, for example the peaks between F3 and G3 varied from 150 cents above G3 to C4. Such differences in peak were seen between ascending and descending scales on the same vowel as well as between the different vowels /ɑ, æ, i/, and even between two recordings of descending scale on the same vowel, as shown in Fig Time F0 [Hz] Ascending scales on /?,?, i/ / ɑ, ӕ, i / /ɑ/ /ӕ/ /i/ Time [ms] Descending scales on /i/ Fig F0 in scales. (Left): The takiyah peaks differ between ascending scales on the vowels /ɑ, æ, i/. (Right): The takiyah peaks differ also between two recordings of a descending scale on /i/. F0 [ ST above A2 ] Time [ ms ] 2 In other studies not dealing with voice science, the discussion is rather tahrir-oriented, probably due to the fact that the takiyah is sometimes not discussed at all by singers and teachers of avaz. Even when takiyah occurrences are marked between the melody notes in avaz transcriptions, the discussion often remains tahrir-oriented, e.g. in Tatsumura (1980) and Simms (1996). 10

15 ( The takiyah peak frequencies in the avaz song were more even and thus more in control when the two recordings of the same episode were compared; the fundamental frequency in several phrases containing series of modal-takiyah-modal transitions were found to be more or less identical between the two avaz recordings, as shown in Fig Two recordings of Avaz song (ca 700 ms silence added in REC 2) REC 1 REC 2 F0 [ST above A2] Time [ ms ] Fig The two recordings of the avaz song are rather similar. The difference in timing is minor and non-significant, since the song is free of meter. The differences in takiyah peaks in the tahrir in the initial 4 seconds are not musically significant. However, takiyah peaks preceding the same modal pitch varied much more when frequencies in the same recording were compared, both within the same exhalation and between two different phrases. Sometimes also two takiyahs that were very close in time, e.g. preceding two consecutive modal tones of the same pitch, had different peaks. Variations in peak could be seen even between two takiyahs (T1 and T2) preceding not only the same modal pitch M2 but also departing from the same modal pitch M1. Note that the two simplest cases of such scenarios are tone repetition (M1-T1-M1-T1-M1- ) and alternation (M1- T2-M2-T3-M1-T2-M2- ), which are shown in Fig Tone Repetition #3 F0 Histogram Tone repetition #1 Tone Repetition #3 14 0, , ,12 0, ,06 0, a Time [ms] a F0 [ST above 201 Hz] b Time [ms] b F0 [ST rel A2] Occurrence F0 [ST rel 201 Hz] Occurrence F0 Histogram Tone Repetition # F0 [ST above 201 Hz] F0 [ST rel A2] Alternation in Avaz song 21 Occurrence F0 Histogram for alternation in Avaz song 0,09 Alternation #2 F0 Histogram Alternation # ,06 0,06 9 0, ,03 8 0, c Time [ms] c F0 [ST rel A2] d Time [ms] d F0 [ST above 201,5 Hz] Fig Takiyah peak variations in tahrirs made of tone repetitions and alternations. a-b) The takiyah peaks in tone repetitions vary about 100 cent, but the modal tone frequency is the only significant frequency according to the histogram. c-d) The takiyah peaks preceding the same pitch vary also in alternations, but again the histograms show that only the modal tones are dominating. F0 [ST rel 201,5 Hz] Occurrence 0,08 In the excerpts presented in Fig , the takiyah peaks preceding modal tones at the same pitch in the repetitive sequences mostly vary within about 100 cent. But the histograms in the figure also reveal that tone repetitions and alternations are being sung; the histogram clearly shows that the dominating frequency in tone repetitions belongs to the repeated modal tone. Similarly, the histograms for the alternations show that the two melody tones were sung with equal durations, although the upper tone produces a less stable histogram curve with more tails due to the fact that both takiyah peaks occur above the upper tone. Thus, the histograms clearly show that tone repetitions and alternations were produced when the F0 curves show that those were produced with interleaving takiyah episodes. 11

16 Only a few takiyah peaks were more than 500 cents above the next melody tone in the avaz song. In the scales, the peaks were higher, sometimes as high as ca 700 cents above the next melody tone, whereas in tone repetitions and alternations, they were well below 500 cents. The difference between the lowest and the highest takiyah peak frequencies within each sequence of tone repetitions was approximately cents, with the lower peaks usually in the beginning and the highest peak usually somewhere after the initial 6-8 repetitions. The last few tone repetitions had decreasing takiyah peaks and were almost twice as slow as the initial ones. Thus, the envelope of the F0 curve reminded of a parable as it ascended up to the maximum peak midway through the duration of the repetition sequence in order to descend somewhat thereafter. The SPL dropped during the takiyah episodes, regardless of which vowels were being sung, as is shown in Fig Whenever the F0 curve jumped up towards a takiyah peak, the SPL curve dropped to a negative peak and rose again when F0 was diving from the falsetto peak towards the next modal tone. Considering the modal melodic pattern only, however, the SPL moved in the same direction as the modal tones; the SPL increased and decreased in parallel with ascending and descending modal tone patterns Time [centiseconds] Time [centiseconds] SPL [db] 111 Scale on /i/ F0 SPL Avaz song Tahrir on vowel /i/ F0 [ST rel A2] 290 F0 [ Hz ] SPL [ db ] F0 [ Hz ] Scale on /ɑ/ 340 F0 SPL Scale on / i/, SPL & F0 SPL [ db ] F0 SPL Scale on /?/, SPL & F Time [centiseconds] Fig F0 and SPL in scales on the open vowel /ɑ/ and the closed vowel /i/, and also on /i/in a tahrir in the avaz song. SPL dropped at all takiyah episodes regardless of vowel. Our data did not indicate any clear correlation between SPL and takiyah peak intervals, neither during the modal tones nor the takiyah episodes, which is shown in Fig In the alternation, the SPL alternated along with F0 for the modal tones and it dropped during the takiyah peaks. Still, the takiyah peaks mostly increased although the SPL of each modal tone pitch remained unchanged. It also seemed that the SPL mostly dropped less when the falling interval between the takiyah peak and its following modal tone increased. On the other hand, the SPL increased during the initial short tones of each tone repetition sequence and reached its maximum near the repeated modal tone that had the highest takiyah peak. That is, it seemed that increasing dynamics to some extent lead to higher takiyah peaks when the modal tones had the same pitch. However, while the SPL level stayed high throughout the few longer modal tones at the end, the takiyah peaks in those last repetitions decreased. It seemed that the SPL of the modal tones as well as of the takiyah episodes in some cases increased without affecting the jump intervals between the takiayh peaks and the modal tones Time [centiseconds] 25 F Time [centiseconds] Fig SPL and F0 during tone repetitions and alternation Time [centiseconds] F0 [ST rel A2] SPL 112 SPL [db] 103 Tone repetitions 26F0 F0 [ST rel A2] SPL [db] F0 [ST rel A2] SPL [db] 109 SPL Tone repetitions SPL F0 Alternation: F0 & SPL

17 2.2 Voice source The voice source data (NAQ, H1-H2, QClosed) for the scales sung in modal register on the vowels /ɑ, æ, i/ are shown versus F0 in Fig , for the pitch range F3-F4, approximately. In a nearsighted reading we can see that the NAQ data was scattered for /æ/ while it remained low for /ɑ/. For /i/ it tended to increase linearly with F0, meaning that the value was nearly doubled when a tone was raised by one octave. The values of H1-H2 mostly remained in the range of 4-6 db for /ɑ/ and /i/. QClosed mostly tended to increase with F0 for /æ/, to decease for /i/, and to mostly remain within the range for /ɑ/. 0,2 Scales 16 Scales 0,7 Scales / ɑ / NAQ 0,15 0,1 H1-H2 [db] QClosed 0,5 0,3 / ӕ / 0,05 0 0, F0 [Hz] F0 [Hz] F0 [Hz] Fig Voice source data (NAQ, H1-H2, QClosed) in the ascending scales sung on the vowels /ɑ, æ, i/. / i / The same voice source parameters (NAQ, H1-H2, QClosed) for the vowels /ɑ, æ, i, e/ sung in the avaz song are presented versus F0 for the pitch range Hz in Fig For /ɑ/, NAQ remained more or less constant around 0.1 over the entire F0 range, while for /i/ it tended to increase with F0, and for /æ, e/ it was more scattered. Thus, the NAQ values for /ɑ, æ, i/ in the avaz song showed the same tendencies as in the scales. The H1-H2 values in the avaz song tended to be higher than the scale values; some values for /æ/ in the lower part of the F0 range were above 12 db. Also most other H1-H2 values in the avaz song were a couple of db higher than in the scales. The QClosed values in the avaz song were mostly in the same range as in the scales, i.e. ca , with the difference that fewer values in the avaz song were in the lower part of the range, and also that /ɑ, i/ were more scattered and did not show any tendencies to increase with F0. 0,2 Vowels in avaz song 16 Vowels in avaz song 0,7 Vowels in avaz song / ɑ / NAQ 0,15 0,1 0, F0 [Hz] H1-H2 [db] F0 [Hz] Fig Voice source data (NAQ, H1-H2, QClosed) in the avaz song on the vowels /ɑ, æ, i, e/. QClosed 0,5 0,3 0, F0 [Hz] / ӕ / / i / / e / There may be several reasons why some voice source data from the scales and the song differ. The narrow and fragmented coverage of F0 for the vowels /i, e/, make it more difficult to discern a systematic variation with F0. It might also be relevant that the scale tones were sung in one phrase and each tone was sung once, producing basically one data point per pitch for each vowel. In the avaz song, on the other hand, tones at the same pitch occurred in several phrases, thus producing more than one value for a given F0 on a given vowel. The NAQ values varied in the range , which indicates great variation in the level of adduction. The vowel /ɑ/, which covered the widest F0 range, A3-G4, approximately, had the least variation in NAQ, which mostly stayed around 0.1, thereby indicating a consistently high level of adduction. But it needs to be added that on the approximate pitch F, the NAQ range was , which indicates a 13