Analysis
of Acoustical Features of Biphonic Singing Voices
Male
and Female Xöömij and Male Steppe Kargiraa
TAKEDA,
Shoichi and MURAOKA, Teruo
–
1
E-mail: takeda@thu.ac.jp E-mail: muraoka@ee.ec.musashi-tech.ac.jp
ABSTRACT
This paper clarifies
spectral features of Mongolian or Tuva’s biphonic singings such as Xöömij,
Steppe Kargiraa, etc. Spectra of five types of Xöömij sounds sung by male
singers showed that a resonance with a high Q value is necessary if a
listener is to perceive two pitches, and the spectra of all the sounds were
found to have second-formant peaks corresponding to the higher-pitch voice.
Similar second-formant peaks were observed in Xöömij sounds sung by a female
singer. In Steppe Kargiraa /a/ sounds sung by a male singer, we found that
first formants have acute peaks instead.
INTRODUCTION
Traditional Asian biphonic singings, among which the Mongolian "Xöömij
[1]" may be best known, are produced by a single singer articulating two
voices simultaneously: a "drone," which is bass voice of almost
constant low-pitch, and a "melody tone" of high-pitch. Xöömij is most
popular in West Mongolia [2], and its singing technique is thought to have
spread to European countries and been used in epical chants such as "two
voices from a single mouth" in
The origin of Xöömij
is still uncertain. It was once thought to have been a kind of conjuration, but
today is most widely believed to have sprung from a vocal imitation of
murmuring streams or the echoes in the Altai mountain-chain [3, 4]. It has also
been suggested that Xöömij is an imitation of the sounds of the Morin Khuur [3]
and was used to pacify female animals separated from their young; a way in
which it is still used [5].
This paper pursues the
process of Xöömij generation by using the results of spectral analysis. Taking into
account the results of previous acoustical analyses [6, 7],
we formulated the following three hypotheses:
1. There actually is, in
addition to a glottal source, an independent sound source (such as a whistle). (Hypothesis
of Independent Sound Sources) [8]
2. Some portion of the
vocal tract vibrates at a high frequency, and the product of the modulation of that
high frequency vibration with a glottal source is perceived as the melody tone.
(Hypothesis of Modulation)
3. A sharp resonance formed
by a peculiar vocal tract shape selectively enhances some harmonics of the
glottal source, and this resonance is perceived as the melody tone. (Hypothesis
of Resonance)
Past soundspectrographic
analyses [1], [9] did not prove any of the hypotheses because the amount of
data analyzed was insufficient and the measurements were not accurate enough.
We [6, 7], [17] first tested whether the “Hypothesis of Resonance” would be
supported by the results of a detailed spectral analysis of a typical example
of Xöömij singing and then repeated the analysis [18] using a Xöömij recording
obtained under better conditions and using a state-of-the-art computer system.
We then examined whether or not our results would hold for other types of
Xöömij singing [11-13]. We first investigate the mechanism of Xöömij generation
by using numerical speech signal analyses such as short-time FFT analysis, LPC
analysis, and cepstrum analysis. Observing the harmonic structures of Xöömij
sound waveforms and tracing the transitions of formant frequencies and the accompanying
Q (quality of resonance) values, we obtained results consistent with the
“Hypothesis of Resonance.”
Adachi & Yamada
recently also used FFT and LPC as part of their research on vocal tract shapes during
Xöömij singing [10], [16]. They used four Xöömij samples sung by one singer
(the type of Xöömij is unknown), and their results also support the Hypothesis
of Resonance.
NUMERICAL SIGNAL
ANALYSES [18]
We investigated the three
hypotheses using Xöömij material. After careful auditory examinations, we selected
a recording of unaccompanied single Xöömij singing entitled "Gooj
Nanaa" (the singer is unknown) recorded on the LP "Folk Songs [Asian
version]" (JVC SKX25017 25018, Japan). The signal was digitized (16-bit
samples) at a sampling rate of 22.05 kHz for calculation of formant frequencies,
bandwidth, and Q values. For spectrum display the sampling rate was only
11.025 kHz. Short-time FFT was again applied to 1024 data samples and LPC
analyses were carried out with a 30-msec Hamming window weighting. The order of
LPC analysis for a sectional spectrum display was 10 and that for a 3D
time-varying spectrum pattern display was 12. The orders were determined empirically
by observing each spectrum.
Figure 1 is an expanded
view of the middle part of a Xöömij waveform, where the waveform is considered
almost stationary. The melody-pitch heights that were obtained by music score transcription
approximately coincided with the second formant frequency F2. This
suggests that the movements of F2 are perceived as melody in Xöömij
singing. To trace the variation of F2, we calculated the successive
spectrum envelopes shown in
Fig. 2.
A distinctive feature of our analysis that a formant
that forms the melody tone is revealed by the use of the LPC method. As shown in Fig. 2, this formant is extracted clearly
and quantitatively. Notable findings are that the intensities of the second formants
of Xöömij sound waveforms are quite different from
those of normal speech and that the Q values of F2 range from 6
to 98 and have an average value of 32.
According to the data in
the literature [14, 15], the estimated Q of formants in normal speech is
at most 30. The spectra of a Xöömij sound signal have a harmonic structure
consistent with the Hypothesis of Resonance.
SPECTRAL FEATURES OF
VARIOUS XÖÖMIJ ARTICULATIONS [11-13], [17, 18]
The detailed spectral
investigation described in the previous section supports the Hypothesis of Resonance
but was based on the analysis of only a single Chest Xöömij sample. A stronger conclusion
could be drawn from the analysis of many samples of Xöömij with different articulations.
We further investigated
samples of five types of Xöömij singing in order to find out whether there are
spectral differences between the different types. The samples we analyzed were
(1) Nasal Xöömij, (2) Oral-Nasal Xöömij, (3) Glottal Xöömij, (4) Chest Xöömij,
and (5) Throat Xöömij.
This classification is
based on where the singer believes the resonance point to be, and there is no
proof that the resonance is actually at that place. These Xöömij samples were
sung by male Mongolian singer Ganbold and were recorded on a CD entitled
"Mongolian Songs" (KING RECORD, KICC-5133, Japan (1988)).
For sound pieces in which
each of the present authors perceived two tones, sharp peaks could be observed
in their spectra. These peaks correspond to the second formant frequencies F2,
which thus are strikingly enhanced and are heard as the melody tone. This was
commonly found for each type of Xöömij investigated in the present study, thus
supporting the Hypothesis of Resonance.
FORMANT TRANSITIONS FROM
We also tried to
clarify the spectral features of the transition from the sounds of normal 
vowels to Xöömij sounds. It is widely recognized
that the phonetic impressions of Xöömij sounds somehow resemble [i], [e], or
[u] sounds and that Xöömij initially sounds similar to an [u] when the melody
tone is not heard clearly. We asked a Japanese Xöömij singer to articulate [(1)
As shown in the F1-F2
diagram in Fig 3, shifts of the F1-F2 combinations toward the
region of [i] were always observed. This suggests that the location of the stricture
during Xöömij singing is almost the same as its location during the
articulation of the vowel [i]. In the transitions from vowels to Xöömij, F1
shifted to about 250 Hz, while F2 shifted into the range of 1.8 kHz 2.3 kHz
and its remarkable Q-increases were also observed. The frequency range
of F2 is almost the same as that of the melody tone.
ACOUSTICAL FEATURES OF
FEMALE XÖÖMIJ
VOICES
This section describes
acoustical features of female Xöömij voices. It is known to be difficult for females
to sing Xöömij songs.
Analysis was conducted using voices of Mongolian female singer Sainkho Namtchylak recorded on a CD
entitled "Lost Rivers" (FMP CD 42, Germany (1992)).
The signal was digitized
(16-bit samples) at a sampling rate of 16 kHz for spectrum display. Short-time
FFT and LPC analyses were carried out with a 30-msec Hamming window weighting.
Figures 4 (a) shows a short-time spectrum of monophonic part of a
female Xöömij sound waveform, and (b) shows that of biphonic
part. A sharp peak can be observed in the spectrum in Fig. 4 (b), whose sound
is perceived as two pitches. This peak corresponds to the second formant frequency
F2, which is strikingly enhanced and is heard as the higher pitch. This
was commonly found for each sample of female Xöömij voices investigated in the
present study, thus supporting again the Hypothesis of Resonance.
A conspicuous difference
from male Xöömij voices is in that the harmonic structure of the spectrum of a
female Xöömij sound waveform is coarse compare to that of a male one.This
coarse harmonic structure may be the reason why it is difficult for female
singers to control melody tones.
ACOUSTICAL FEATURES OF
MALE STEPPE KARGIRAA VOICES
Another interesting biphonic singing is a
Tuva’s singing method called “Steppe Kargiraa,” which is characterized by an
extremely low fundamental pitch. Recently the voice-production process has been
explained by Imagawa, Sakakibara, Konishi, and Niimi using a glottal source
model based on a “false vocal fold [19].” In this section we describe the results
of spectral analysis of Steppe Kargiraa sound waveforms that have an auditory
impression near a vowel /a/.
Analysis was carried out
using voices of two male singers, Fedor Tau and Gundenbiliin Yavgaan. Tau’s
voices were recorded on a CD entitled "TUVA Voices from the Center of Asia"
(Smithsonian Folkway CD SF 40017, USA (1990)), and Yavgaan’s
voices on a CD entitled “Mongolian Xöömij” (King KICW 1004, Japan (1999)). The
signal was digitized (16-bit samples) at a sampling rate of 16 kHz for spectrum
display. Short-time FFT and LPC analyses were carried out with a 30-msec
Hamming window weighting.
Like Xöömij sound
waveforms, the spectrum of a Steppe Kargiraa waveform in Fig. 5 (b) shows a
prominent formant peak; while that of a normal vowel /a/ in Fig. 5 (a) does
not. An interesting finding here is that the peaks yielding melody tones are
not the second formant frequencies F2 but the first formant frequencies F1
CONCLUSIONS
We have analyzed spectral
features of two types of biphonic singing: Xöömij in
To further test this
hypothesis, we evaluated samples of four types of Xöömij singing classified according
to where the singer believes the resonance point to be. Sharp peaks were found
in the spectra of all types of Xöömij. These results support the Hypothesis of
Resonance, in which glottal waves and the sharp resonance of their higher
harmonics are perceived as biphonic tones.
Another important finding
in this work is that the first formant frequencies of Xöömij sound waveforms are
constant. Investigating the transitions of formant frequencies from normal
vowels to Xöömij sounds, we found that the F1-F2 combination
always shifts toward the [i] region, with the first formant frequencies
shifting to about 250 Hz.
The results of analyses of
spectral features of female Xöömij and male Steppe Kargiraa singings also showed
sharp formant peaks in the spectra that yield perception of melody tones. A
conspicuous feature of spectra of female Xöömij sound waveforms is that the
harmonic structure is coarse compared to those of male Xöömij sound waveforms,
which may make female singers control melody tones difficult.
ACKNOWLEDGMENTS
The authors express their
sincere appreciations to Professor Kiyoko Motegi at
Ochiai at
Plankton Co. for their offering valuable information on Xöömij. Finally, the authors would like to appreciate Messrs.
Masashi Itoga, Katsuhisa Tadokoro, and Masashi Miyashita, former students at
the Te ikyo University of Technology (presently
This research was partly
supported by Grant -in-Aid from
Grant-in-Aid for Scientific
Research on Priority Areas (2) "Diversity of Prosody and its Quantitative
Description"
from the Ministry of Education, Culture, Sports, Science and
REFERENCES
[1] Trân
Q. H. and D. Guillou, "Original research and acoustical analysis in
connection with the Xöömij style of biphonic
singing," Musical Voices of Asia, Individual research reports |
[2] M.
Yamada, "Mongolian biphonic singing Xöömij," Journal of the
Acoustical Society of Japan Vol. 54-9, pp.680-685 (1998).
[3]
" A general survey of Mongolian music," Asian traditional
performing arts 1978," The Japan Foundation, pp.5-9 (1978.11).
[4] Batzengel, "Urtin
duu, Xöömij, and Morin xuur," Musical Voices of Asia, Seminar information
and documentation |
[5] H.
Hasumi, "Understanding Mongolian music," Musical Voices of Asia,
Seminar information and documentation |
[6] T.
Muraoka, K. Wagatsuma, and M. Horiuchi, "Acoustic Analysis of the
Mongolian singing Xöömij," Preprint of the
Acoustical Society of Japan 2-3-9, pp.385-386 (1983.10).
[7] T.
Muraoka, K. Wagatsuma, Y. Tsuchikane, and M. Horiuchi, "On a Consideration
of Mongolian Singing Xöömij and its
Specialities," Preprint of the seminar on Musical acoustics, The
Acoustical Society of Japan MA84-1, pp.1-6 (1984).
[8] B.
Chernov, and V. Maslov, "Larynx -double sound generator," Proc. 11th
Int'1. Cong. Phonetic Sci., pp.40-43 (Tallin, Estonia, 1987).
[9] S.
Gunji, "An acoustical consideration of Xöömij," Musical Voices of
Asia, Individual research reports |
[10] S.
Adachi, and M. Yamada, "An Acoustical Study of Sound Production in
Biphonic Singing, Xöömij," Proceedings of 1997
[11] S. Takeda, M. Itoga,
Y, Sato and Y, Ueda, “Analysis of Acoustical Features of Mongolian Singing “Khöömij”," Proc. Acoust. Soc. Jap. 2-7-15, pp605-606 (Oct, 1992).
[12] S. Takeda, M. Itoga,
“On the differences in Spectra in Accordance with the Phonemic and Tone-height
Differences in Mongolian Singing “Khöömij”," Proc. Acoust. Soc.
Jap. 2-3-3, pp.499-500 (March, 1993).
[13] S.
Takeda, M. Itoga, “Analysis of Acoustic Features of Mongolian Singing “Khöömij”,"
Technical Report on Musical Information Sci.1-4, pp.1-4 (April, 1993).
[14] J.
Ohizumi, and Y. Fujimura, Onsei kagaku (Science of Human Voices), Tokyo
University Publishing (1972).
[15] K.
Nakata, Onsei (Human voices), Acoustic Engineering Series by the
Acoustical Society of Japan (Corona Publishing Co., Ltd., Tokyo, 1977).
[16] S.
Adachi, and M. Yamada, "An Acoustical Study of Sound Production in
Biphonic Singing, Xöömij," Journal of the
Acoustical Society of America, 105, pp.2920-2932 (May, 1999).
[17] T.
Muraoka,
[18] T. Muraoka,
[19] H. Imagawa, K.
Sakakibara, T. Konishi, and S. Niimi, “Glottal Source Model for Throat Singing Based
on Vocal Fold and False Vocal Fold Vibrations,” Proc. Acoust. Soc. Jap. 1-6-14, pp.255-256 (March 2001).