Voice onset times and burst frequencies of four velar stop
consonants in Gujarati
Manish K. Rami, Joseph Kalinowski, Andrew Stuart, and Michael P. Rastattera)
Department of Communication Sciences and Disorders, East Carolina University, Greenville,
North Carolina 27858-4353
共Received 12 July 1999; revised 29 July 1999; accepted 20 August 1999兲
Voice onset times 共VOT兲 and burst frequencies of two aspirated 共i.e., /kh /, /gh/) and two unaspirated
共i.e., /k/, /g/兲 Gujarati velar stop consonants were investigated in an effort to provide characteristic
acoustic information. Stop consonants in a monosyllabic vowel–consonant–vowel 共VCV兲
production were obtained from eight native speakers of Gujarati. Differences in mean VOT and
burst frequency as a function of voicing and aspiration were examined. A significant voicing by
aspiration effect was found for VOT (p⫽0.026). The two voiced stops, while not significantly
different from each other (p⫽0.278), had significantly shorter VOTs than voiceless stops. The
aspirated /kh/ had a significantly longer VOT than the unaspirated /k/ (p⫽0.0013). With respect to
burst frequency, voiced stops had significantly higher burst frequencies than voiceless stops (p
⫽0.002). There was no significant difference between mean burst frequencies of aspirated and
unaspirated stops (p⫽0.058). © 1999 Acoustical Society of America. 关S0001-4966共99兲00312-4兴
PACS numbers: 43.70.Fq, 43.70.Kv 关AL兴
INTRODUCTION
It has been shown that phoneme categories for stop consonants can be differentiated by differences in the temporal
aspects of the onset of the periodic laryngeal vibration or
glottal pulsing and the articulatory events associated with the
release of the consonant burst 共Fant, 1969; Ladefoged, 1996;
Lieberman, 1977兲. This period, between the release burst of
the stop sound and the onset of vibration of the vocal folds,
is generally referred to as voice onset time 共VOT兲. While
VOT is quite effective in separating phonemic categories in
most two- and three-category languages, it is inadequate in
demarcating stop-consonant contrasts in four-category languages such as Hindi and Marathi 共Ladefoged and Maddieson, 1996; Lisker and Abramson, 1964兲. A four-category
language is one in which there are four stops at one place of
articulation 共e.g., four velars, four dentals, and so on兲. In
these languages, aspiration is used in conjunction with the
voicing element to produce four categories of stop consonants.
In a seminal cross-language study of initial stops in 11
languages, Lisker and Abramson 共1964兲 reported VOT values for the voiced unaspirated and voiced aspirated stops in
four-category Hindi and Marathi languages. Although there
were orderly differences in mean values of VOT, there was
an overlap in the ranges of variation in distributions of VOTs
of the two voiced stops, and the two proved to be nearly
similar. Lisker and Abramson indicated that the voiced aspirates differed from the other voiced category by the presence
of a low-amplitude ‘‘buzz’’ assimilated with noise in the
interval following stop release. It has also been suggested
that aside from formant transitions, the burst frequency could
also provide phonetically salient information 共Halle, Hughes,
a兲
Author to whom correspondence should be addressed. Electronic mail:
rastatterm@mail.ecu.edu
3736
J. Acoust. Soc. Am. 106 (6), December 1999
and Radley, 1957; Stevens, 1972, 1980兲 in these cases. The
frequency of the release burst is determined by turbulent airflow at the release of the stop.
The Gujarati language, similar to Hindi and Marathi, has
four velar stops. Specifically, Gujarati contains two aspirated
共i.e., /kh/, /gh/) and two unaspirated 共i.e., /k/, /g/兲 velar stop
consonants. Gujarati is spoken by 44 million people worldwide and is ranked twenty-third in the top 100 languages by
population 共Grimes, 1996兲. While the linguistic percept of
these sounds to the native Gujarati speaker is apparent, information pertaining to the acoustic properties is unknown.
While perceptual discrimination of the four velar Gujarati
stops may be based on VOT distinctions, it remains possible
that the information provided by the burst spectrum of each
consonant might provide salient perceptual information as
well. Toward that end, VOTs and the consonant-stop burst
frequencies of the four velar Gujarati stop consonants: /k/,
/g/, /kh/, and /gh/ were examined.
I. METHOD
A. Participants
Eight normal native speakers of Gujarati (M ⫽33.6
years, s.d.⫽18.5; six males and two females兲 served as participants. While participants had learned Gujarati as their
first language, all were bilinguals with English as their second language.
B. Apparatus
Speech samples were recorded in a quiet room with a
microphone 共Apple model Plaintalk兲 placed approximately
40 cm from lips with an orientation of 0° azimuth and 20°
altitude. The microphone output was line-fed into a personal
computer 共Apple model Power Macintosh 7100/80兲 interfaced with an analog-to-digital input/output board 共Digidesign model Audiomedia NuBus兲. Amplitude waveforms
0001-4966/99/106(6)/3736/3/$15.00
© 1999 Acoustical Society of America
Downloaded 08 Nov 2011 to 134.129.171.185. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
3736
were generated by using a commercially available speech
and sound signal analysis program 共InfoSignal, Inc. model
Signalyze 3.12兲. A fast Fourier transfer spectral analysis of
the speech samples was also undertaken to obtain spectrograms and spectra that were used for acoustic analysis. The
analog speech signals were digitally sampled at 44.1 kHz
and quantized at 16 bit.
TABLE I. Grand means and standard deviations of voice onset time 共ms兲
and burst frequency 共Hz兲 for the four velar Gujarati stop consonants (n
⫽8). Standard deviations are presented in parentheses.
Velar stop consonant
Voiced
Voice onset time
C. Procedure
Each participant produced 25 utterances of each of the
four velar stop consonants of Gujarati 共i.e., /k/, /g/, /kh/, and
/gh/ in serial order兲 in a single test session. These consonants
were recorded in a carrier phrase using a neutral schwa before and after the target consonant 共i.e., h#ve a VCV tʃhe).
Each utterance was produced after a normal breath with a
pause before the target sound to decrease the effects of varying lung volume on VOT 共Hoit, Soloman, and Hixon, 1993兲.
Participants successively produced the phrases for each stop.
VOTs were measured from the amplitude waveforms of
each stop. VOT was defined as the duration between the
consonantal burst and the first glottal cycle 共Lisker and
Abramson, 1964兲. Burst frequency was measured from the
spectra of each consonant. Spectra were obtained by placing
a 10-ms window centered at the point of highest amplitude in
the burst. The burst frequency was defined as the peak with
the highest amplitude 共Stevens, 1980兲.
Ten percent of the data were reanalyzed in an effort to
provide an index of intrajudge and interjudge agreement. Intrajudge agreement for VOT and burst frequency, as measured by Pearson Product correlation, was 0.98 (p⬍0.001)
and 0.99 (p⬍0.001), respectively. The mean standard error
was also used to index intrajudge reliability 共Ferguson,
1976兲. The intrajudge mean standard errors for VOT and
burst frequency measurements were 0.2 ms and 2.19 Hz,
respectively. Interjudge agreement was evaluated with a second independent judge unaware of the purpose of the study.
Pearson Product correlations indexing interjudge agreement
were 0.97 (p⬍0.001) and 0.98 (p⬍0.001) for VOT and
burst frequency, respectively. Interjudge mean standard errors of measurement proved to be 0.39 ms for VOT and 2.97
Hz for the burst frequency, respectively.
Burst frequency
Voiceless
h
/g/
/g /
⫺37.3
共9.8兲
⫺29.2
共7.3兲
1555
共198兲
1531
共176兲
/k/
/kh/
40.6
共21.1兲
75.0
共33.2兲
1451
共146兲
1427
共140兲
less stop, /kh/, was significantly longer than the unaspirated
voiceless stop /k/ 关 F(1,7)⫽26.8, p⫽0.0013]. VOTs for
voiced stops /gh/ and /g/, although significantly shorter than
the voiceless stops, were not statistically different from one
another 关 F(1,7)⫽1.379, p⫽0.278].
Differences in mean burst frequency as a function of
voicing and aspiration were also examined with a repeated
two-factor ANOVA. Results showed a significant main effect of voicing 关 F(1,7)⫽25.2, p⫽0.002, ␻ 2 ⫽0.729]. The
main effect of aspiration 关 F(1,7)⫽5.14, p⫽0.058, ␻ 2
⫽0.315] and the interaction between voicing and aspiration
关 F(1,7)⫽0.005, p⫽0.947, ␻ 2 ⫽0] were not significant. In
other words, voiced stops had significantly higher burst frequencies than voiceless stops. There was no significant difference between mean burst frequencies of the aspirated and
unaspirated stops.
III. DISCUSSION
The findings of this study suggest that the voicing feature of velar stop production in Gujarati parallels that of the
two four-category languages of Hindi and Marathi 共Lisker
and Abramson, 1964兲. That is, the two voiced aspirated /gh/
and unaspirated /g/ Gujarati stops reside on the negative side
of the VOT continuum while the voiceless aspirated /kh/ and
unaspirated /k/ are located in the positive half of the VOT
II. RESULTS
Mean individual values of VOT and burst frequency
were obtained from each participants’ 25 of the four velar
stop-consonant tokens. Grand means and standard deviations
of VOT and burst frequency for the four velar stop consonants of all the participants are presented in Table I.
A repeated two-factor analysis of variance 共ANOVA兲
was used to assess differences in mean VOT as a function of
voicing and aspiration. Significant main effects were found
for voicing 关 F(1,7)⫽153.48, p⬍0.0001, ␻ 2 ⫽0.94] and aspiration 关 F(1,7)⫽13.86, p⬍0.007, ␻ 2 ⫽0.58]. As well, a
significant voicing by aspiration interaction was observed
关 F(1,7)⫽7.99, p⫽0.026, ␻ 2 ⫽0.43]. This interaction is depicted in Fig. 1. Two orthogonal signal-df comparisons were
undertaken to investigate the voicing by aspiration interaction. It was found that the mean VOT for the aspirated voice3737
J. Acoust. Soc. Am., Vol. 106, No. 6, December 1999
FIG. 1. Mean voice onset time as a function of voicing and aspiration for
the four velar stop consonants in Gujarati. Error bars represent plus/minus
one standard deviation of the mean.
Rami et al.: Letters to the Editor
Downloaded 08 Nov 2011 to 134.129.171.185. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
3737
continuum 共see Table I and Fig. 1兲. Burst frequency seems to
be an additional conspicuous acoustic property delineating
the voiced and voiceless Gujarati stops. Statistically significant higher burst frequencies were associated with the voiced
stops 共see Table I兲. Burst frequencies above 1500 Hz were
associated with voiced velar stop productions, while those
below defined the voiceless velar stops. Given that distinctions exist in VOT and burst frequency between the voiced
and voiceless Gujarati stop consonants, it is conjectured that
information provided by these acoustic properties could contribute to the perceptual distinctiveness of voicing.
While VOT and burst frequency appear to be a conspicuous acoustic property delineating voicing, differences
between the aspirated and unaspirated stops within each
voicing category remain less clear. Although significant differences existed in VOT between the voiceless stops, such
was not the case with the voiced stops. Compare mean differences in VOTs of approximately 35 and 8 ms for the
voiceless and voiced stops, respectively. It may be the case
that burst frequency provides an additional cue for differentiating the voiced aspirated /gh/ and unaspirated /g/ Gujarati
stops. Although differences in burst frequency were not significant, the effect size was large 共Cohen, 1988兲. Higher
burst frequencies were evident with the unaspirated voiced
stops. Such a finding suggests that there may be a slight shift
in point of constriction between the aspirated and unaspirated velar stops. Acoustical information contained in the
burst frequency may provide additional salience for the discrimination of the aspirated versus unaspirated voiced velar
stops in the Gujarati. It may also be the case that yet an
undetermined acoustic property not explored in this study
may be the conspicuous property that listeners cue into for
perceptual distinctiveness. This speculation can only be explored in future acoustical and perceptual experiments.
In the case of Hindi, an acoustic and fiberscopic study
by Benguerel and Bhatia 共1980兲 has shown that there are
differences in VOT for the voiceless aspirated and unaspirated and the voiced unaspirated stops. However, for the
voiced aspirated the authors suggested that a third type of
voicing known as murmur existed 共i.e., the simultaneous
presence of voicing and turbulence兲. Though similar data for
Gujarati are currently unavailable, it is conjectured that a
murmur may be found in the Gujarati voiced aspirated stop
and could be an additional salient acoustic property that aids
listeners in distinguishing the unaspirated from the aspirated
stops.
In conclusion, the current findings show systematic
acoustic differences among VOTs and burst frequencies of
the four Gujarati velar stop consonants. It appears that these
differences may be the conspicuous acoustic properties that
contribute to the perceptual discrimination of voicing. In ad-
3738
J. Acoust. Soc. Am., Vol. 106, No. 6, December 1999
dition, VOT may distinguish aspiration for the voiceless velar stops /kh/ and /k/. Additional information provided by
burst frequency may be used to make aspiration distinctions
between the voiced stops aspirated /gh/ and unaspirated /g/
as the VOT continuum overlaps for these consonant pairs.
Unlike the general phonetic types of two- and three-category
languages, the four-category languages seem to be employing attributes of alternate acoustic events for determining
phonetic boundaries, such as information provided by the
burst frequency of the consonant as found in the current
study. It should be pointed out that it is difficult to comment
about the exact voice source characteristics and the laryngeal
control as has been done in the case of Hindi aspirated
voiced stops 共Benguerel and Bhatia, 1980兲 until similar data
from other places of articulation 共e.g., palatal, alveolar, dental, etc.兲 in Gujarati are available. Research employing techniques designed to observe perceptual and physiological differences among the velar Gujarati aspirated and unaspirated
consonant pairs will also help answer these questions more
completely.
ACKNOWLEDGMENTS
The authors thank and acknowledge the thoughtful comments of Dr. Arthur S. Abramson, Dr. Peter Ladefoged, and
Dr. Anders Löfqvist on an earlier version of this paper. Presented in part at the North Carolina Speech, Language, and
Hearing Association Annual Convention, Charlotte, NC, 24
April 1998.
Benguerel, A.-P., and Bhatia, T. K. 共1980兲. ‘‘Hindi stop consonants: an
acoustic and fiberscopic study,’’ Phonetica 37, 134–148.
Cohen, J. 共1988兲. Statistical Power Analysis for the Behavioral Sciences,
2nd ed. 共Erlbaum, Hillsdale, NJ, 1988兲.
Fant, G. 共1969兲. ‘‘Stops in CV syllables,’’ in Speech Transmission Labs
Quarterly Progress Reports, 41 共Royal Institute of Technology, Stockholm兲.
Ferguson, G. A. 共1976兲. Statistical Analysis in Psychology and Education,
4th ed. 共McGraw–Hill, New York兲.
Grimes, B. F. 共Ed.兲. 共1996兲. Ethnologue: Languages of the World, 13th ed.
共Summer Institute of Linguistics, Dallas兲.
Halle, M., Hughes, G. W., and Radley, J. P. A. 共1957兲. ‘‘Acoustic properties
of stop consonants,’’ J. Acoust. Soc. Am. 29, 107–116.
Hoit, J. D., Soloman, N. P., and Hixon, T. J. 共1993兲. ‘‘Effect of lung volume
in voice onset time 共VOT兲,’’ J. Speech Hear. Res. 36共3兲, 516–520.
Ladefoged, P. 共1996兲. Elements of Acoustic Phonetics, 2nd ed. 共University
of Chicago Press, Chicago兲.
Ladefoged, P., and Maddieson, I. 共1996兲. The Sounds of the World’s Languages 共Blackwell, Cambridge, MA兲.
Lieberman, P. 共1977兲. Speech Physiology and Acoustic Phonetics: An Introduction 共Macmillan, New York兲.
Lisker, L., and Abramson, A. S. 共1964兲. ‘‘A cross language study of voicing
in initial stops: Acoustical measurements,’’ Word 20, 384–422.
Stevens, K. N. 共1972兲. ‘‘Quantal nature of speech,’’ in Human Communication: A Unified View, edited by E. E. David, Jr. and P. B. Denes
共McGraw–Hill, New York兲.
Stevens, K. N. 共1980兲. ‘‘Acoustic correlates of some phonetic categories,’’
J. Acoust. Soc. Am. 68, 836–842.
Rami et al.: Letters to the Editor
Downloaded 08 Nov 2011 to 134.129.171.185. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp
3738