- Email: [email protected]
Laryngeal adjustments in Japanese voiceless sound production
Journal of Phonetics (1982) 10, 1-10
Laryngeal adjustments in Japanese voiceless sound production Hirohide Yoshioka Haskins Laboratories, 270 Crown Street, New Haven, Connecticut 06510, US.A. and Research Institute of Logopedics and Phoniatrics, Faculty of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113, Japan
Anders Lofqvist Haskins Laboratories, 2 70 Crown Street, New Haven Connecticut 06510, US. A. and Department of Phonetics, Lund University, Helgonabacken 12, S-223 62 Lund, Sweden
Hajime Hirose Research Institute of Logopedics and Phoniatrics, Faculty of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113, Japan Received 11th October 1980
Abstract:
As part of a series of investigations on the production of sequence of unvoiced sounds in different languages, the current experiment was conducted using the combined techniques of photo-electric glottography, fiberoptic filming and laryngeal electromyography. Particular attention was paid to devoiced vowel production in various voiceless consonantal environments including geminates. The data show that the glottal opening gesture during a voiceless sequence containing a devoiced vowel is characterized by a monomodal pattern, unless the vowel occurs between a voiceless fricative and a geminated one, such as /siQs/, where a bimodal pattern may occur. The movement results also suggest that the velocity and size of the glottal opening gesture vary according to the nature of the adjacent voiceless obstruents. The speed of the opening phase is slow when a stop precedes the vowel, and fast when a fricative precedes it. The peak glottal opening attained during the devoiced vowel is larger when a fricative either precedes or follows than when the vowel is surrounded on both sides by single or geminated stops. Furthermore, it is revealed that the. peak velocity of the initial opening gesture occurs at almost the same time in relation to the voicing offset of the preceding vowel, regardless of the properties of the surrounding voiceless obstruents and, thus, irrespective of variations in the magnitude of velocity and opening size.
Introduction At the 97th meeting of the Acoustical Society of America, we reported how voiceless sound sequences, such as voiceless obstruent clusters, are organized in terms of their glottal opening and closing gestures, using native speakers of American English (Yoshioka et al., 1979) and Swedish (Lbfqvist & Yoshioka, 1980). The conclusion of those studies was that, in the production of sequential unvoiced sounds, the glottal opening gesture is characterized by a one, two or more than two peaked pattern in a regular fashion according to the nature of 0095-44 70/82/010001 + 10 $02.00/0
© 1982 Academic Press Inc. (London) Ltd.
2
H. Yoshioka, A. Lofqvist and H. Hirose
the voiceless segments. A voiceless obstruent characterized by aspiration or frication noise tends to require a separate opening gesture, while an unaspirated stop in a voiceless environment can be produced within the opening gesture attributed to an adjacent aspirated stop or fricative. For example, an /sk#sk/sequence in English was produced in most cases with two separate opening gestures. In contrast, an /sks#k/ string was in general accompanied by three opening gestures (Yoshioka et al., 1979). Furthermore, the velocity of the initial opening movement was shown to vary depending on the properties of the initial voiceless segment. When the first unvoiced segment in the cluster was a fricative, the speed of the abduction was significantly faster than when the initial voiceless sound was an aspirated or unaspirated stop, regardless of the nature of the following voiceless segments. This also meant that the difference in velocity during the initial opening phase heid true despite the fact that, for most cases beginning with a voiceless unaspirated stop, peak glottal opening occured during the following fricative segment. In order to examine the validity of these notions across different languages, the current experiment was carried out using the same combined techniques of photo-electric glottography, fiberoptic filming and laryngeal electromyography, in co-operation with a native speaker of Japanese. The phonology of the Japanese does not allow voiceless "pure" obstruent clusters other than geminates. Syllable-final obstruents also rarely occur in this language . On the other hand, in conversational speech of the Tokyo dialect, there is a well-known phenomenon of vowel devoicing in that a high vowel, such as /i/ and /u/, surrounded by voiceless obstruents on both sides is often produced without any vocal fold vibrations during the vowel segment (e.g. Hattori, 1951; Han, 1962; Fujimura, 1971; Sawashima, 1973). Therefore, we paid particular attention to devoiced vowel production in various voiceless consonantal environments including geminates.
Method and procedure The techniques used in the present experiment were simultaneous recordings of photoelectric glottography, fiberoptic filming and laryngeal electromyography (EMG), in parallel with the audio signal. The EMG data were obtained using bipolar hooked-wire electrodes (Basmajian & Stecko, 1962; Hirano & Ohala, 1969). The electrodes, consisting of a pair of platinum- tungsten alloy wires (50 pm in diameter with isonel coating), were inserted perorally into the posterior cricoarytenoid muscle (PCA) under indirect laryngoscopy with the aid of a specially designed curved probe (Hirose et al., 1971 ) . Before insertion, topical anaesthetic was applied to the mucous membrane of the hypopharynx using a small amount of 4% Lidocaine spray (Xylocaine ). For verification of electrode position, the subject was instructed to perform several non-speech and speech maneuvers that are well understood in terms of the manner of the PCA involvement, such as inspiration and expiration during breathing, voluntary swallowing, pitch changes including register shifts, glottal attack, and voiced-voiceless sound constrasts. The EMG signal was monitored on an oscilloscope not only during the verification gestures but also during the entire recording session. The interference voltages of the EMG signals, after high-pass filtering at 80Hz, were recorded on a multi-channel FM recorder together with the audio signal. After full-wave rectification and integration over a 5 ms time window , the action potentials were fed into a computer at a sampling rate of 200Hz for further processing to obtain the muscle activity patterns for ensemble-averaged tokens with a 35 ms time constant (Kewley-Port, 1977). The figures to be presented in this paper represent activity patterns aligned with reference to the voicing offset of the vowel preceding the voiceless sequence.
Laryngeal adjustments in Japanese
3
For the movement data, the glottal view through a flexible laryngeal fiberscope (Olympus VF-0 type, 4.5 mm in outer diameter) was photographed with a cine camera at a rate of 60 frames/s . A synchronization signal was registered on the FM recorder to identify each frame. Then, frame by frame analyses were made with the aid of mini-computer to calculate the distance between the vocal processes; this distance is considered one of the indicators of glottal width (Sawashima & Hirose, 1968; Sawashima, 1976). A cold DC light source (Olympus CLS), providing illumination of the upper glottal area, also served as the light source for the photo-electric glottography. The amount of light passing through the glottis was sensed by a photo-transistor (Philips BPX 81) placed on the neck just below the lower edge of the cricoid cartilage. The electrical output was recorded on another channel of the FM tape. These signals were sampled at 200Hz and processed with the digital system. A native male speaker of the Tokyo dialect, one of the authors, served as the subject. Among the various voiceless environments surrounding a devoiceable vowel /i/, the combination of /s/ and /k/ is optimum in forming the greatest possible number of meaningful words in Japanese. Therefore , as is shown in Table I. We chose the test words that contain a devoiceable vowel /i/ in the middle of voiceless obstruents composed of the phonemes /s/ and /k/. For example , the production of the first word in this list, /kikee/, which means "anomaly" in Japanese , may be transcribed as having an unvoiced string [kbk] --a [k] plus a devoiced [i] plus a slightly aspirated [k]. Table I
Test words and the carrier sentence (Q =geminate phoneme)
Japanese
r-rtt,;J:. _ ~
_ :r-9J
"sorewa
_desu" ( unvoiced string)
ff< :!
/kik ee /
5
/ k iQk ee /
[kj kk )
~J!
lfiil
/ k isee /
[kjs )
5
1!t If<
/ kiQsee /
[ kj ss )
~~ ~
!ill
/ siQkee /
/ sikee /
[kj k )
c IJk l c1j kk )
~
~
/ sisee /
[ \ js)
~
iEi1
I s iQsee /
[I j ss)
For the first two to three repetitions of each test word embedded in the frame sentence, "sorewa ------ desu." , "we call it ------ -." , simultaneous recordings of photo-electric output and fiberoptic filming were made together with the audio signals, followed by 14- 28 additional recordings of only EMG and photo-electric signals. During the latter part of the session, the glottal image was constantly monitored through the fiberoptic view finder. Such careful monitoring is mandatory to obtain reliable interpretations of large amounts of photo-electric recordings, as we have discussed elsewhere (Yoshioka eta!., 1979; LOfqvist & Yoshioka, 1980). Results Figure 1 illustrates the results for the test word /siQsee/. Since the glottal opening patterns obtained by photo-electric glottography have been shown to be practically identical to those
4
H. Yoshioka, A. Lofqvist and H. Hirose /siQsee/ [Jjsse: ) x 28--------------- [/isse :) x 2
G:JJ'\
L
h_ L
I
Pc:oo~~ AE
t.4f~ Voicing offset of preceding vowel
Figure 1
Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentence containing the test word /siQsee/. Twenty-eight devoiced tokens (left) and two voiced tokens (right), respectively.
obtained by plotting the distance between the vocal processes from the fiberoptic cine-films, we will focus on the photo-electric glottograms as an index of glottal width change during the pertinent voiceless sequence production. Here, the top row (GW) represents the averaged glottograms for two allophonic groups. One is devoiced, and the other is voiced. Among the 30 repetitions, 28 tokens were produced with devoiced vowels, while only two of them had fully voiced vowels. These variations were easily detectable in audio waveforms, in sound spectrograms and by listening the recorded tape. As for laryngeal EMG, the figure contains the corresponding averaged activity patterns of the abductor muscle -posterior cricoarytenoid muscle (PCA)- that has been demonstrated to substantially control the glottal aperture (Hirose, 1976; Yoshioka, 1979). These signals were aligned with respect to the voicing offset of the preceding vowel. It is obvious that, when it is uttered with a fully voiced vowel, two clear separate glottal opening gestures are found for the /siQs/ production at both the movement and the electromyographic levels. In contrast, the averaged curves for the devoiced group is a little unclear. The abductor muscle (PCA) activity curve in the middle may be characterized by two opening gestures : The first is associated with a high and steep peak, followed by a second that is broad but of moderate activity level. The glottographic pattern for this devoiced group is more complicated, in that one might describe it as having two peaks or, alternatively, a sort of plateau. Since all the other test words, except the one containing /siQs/ mentioned above, were always produced with a devoiced vowel, all the averaged curves henceforth from Fig. 2 are those for completely devoiced groups. Figure 2 shows the averaged glottographic pattern and the corresponding averaged abductor muscle activity pattern for the devoiced /sis/ sequence, in comparison with those for the devoiced /siQs/ shown in Fig. 1 and repeated in Fig. 2. Here, several points are worth mentioning. First, the averaged glottogram for the non-geminated /sis/ is clearly distinguished by a monomodal curve, while that for the geminated /siQs/ is characterized by a broad or bimodal pattern as mentioned above. This finding seems to be reflected in the EMG. The averaged PCA activity curve for the non-geminated /sis/ has a single peak around the line-up point, while that for the geminated /siQs/ is, as mentioned before, characterized by two separate activity patterns. In addition, despite the
Laryngeal adjustments in Japanese /sisee/
5
/siQs ee/
G:vbJ~ [\~
PC~l~J~~ AEf ~ [,,~ A
Voicing offset of preceding vowel
Figure 2
A
~
~
Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /sisee/ and /siQsee/, respectively.
differences in the overall modality between these two utterance types at both movement and EMG levels, the initial opening phases are quite similar. The peak glottal openings are approximately of the same size and reached almost at the same time. As for the PCA activity, both the curves have their peaks at a same time, i.e. the line-up point. Figure 3 . contains the activity patterns for a devoiced vowel /i/ surrounded on both sides by a pair of single or geminated stops. In comparison with those for the devoiced vowel /i/ occurring between voiceless fricatives shown in Fig. 2, the glottographic curves for these cases have a single and considerably smaller peak. The slopes of the glottographic curves for this stop group are also more gradual than for those surrounded by voiceless fricatives in Fig. 2 . It implies that the glottal opening gesture during the voiceless sequence containing a devoiced vowel may vary according to the nature of the surrounding consonants; slow for the voiceless stop and fast for the voiceless fricative.
/ki keel
G:1:~ u
/kiQkee/
hL
P~~o~~ AEf
JA& ....
Voicing offset of preceding vowel
Figure 3
f..~ .... Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /kikee/ and /kiQkee/, respectively.
6
H. Yoshioka, A. Lofqvist and H. Hirose /sikee/
G:J~\1
/siQkee/
(\ ~
PC3:0~~ AE
t
~
f
Voicing offset of preceding vowel
Figure 4
rh•.>D}.. Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /sikee/ and /siQkee/, respectively.
Figures 4 and 5 show the patterns for the utterance types that contain a devoiceable vowel /i/ in a voiceless sequence composed of two different obstruents, such as /sik/, /siQk/, /kis/ and /kiQs/. It is evident that all the glottographic curves are characterized by a monomodal pattern. In addition, and more interestingly, the difference in the slopes during the initial opening phase depends on the phonetic properties of the initial segments. When a voiceless fricative precedes the devoiced vowel, the opening movement is faster than when a voiceless stop precedes the vowel. Furthermore, peak glottal opening during these devoiced sequences coincides approximately with the peak amplitude of the audio envelope during the devoiced vowel segments. As for the EMG signals, the noise level is too high for a detailed discussion. Nevertheless, it may be mentioned that the peak PCA activity for these utterance types, as well as the others mentioned above, occurs around the line-up, i.e. at the voicing offset of the preceding vowel, regardless of utterance type.
/kiQsee/
/kisee/
G:v~~
P::o~~ AEt
~ Voicing offset of preceding vowel
Figure 5
t
~ Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelop curves for the sentences containing the devoiced test words /kisee/ and /kiQsee/, respectively.
Laryngeal adjustments in Japanese
Voicing offset of pr eceding vowel
Figure 6
7
( ms)
Superimposed curves of the averaged glottograms for all the test voiceless sequences.
For a detailed comparison of the characteristics of the glottal opening gesture for all the utterance types containing various combinations of voiceless sounds, Fig. 6 presents all the glottal movement data superimposed during the pertinent voiceless portions. These averaged curves are again aligned with respect to the voicing offset of the preceding vowel in the frame sentence. The solid lines represent the voiceless sequences beginning with a fricative, while the group of the dotted lines corresponds to those beginning with a voiceless stop. The two bottom graphs show separately these two groups, i.e. the sequences beginning with /s/ and /k/, respectively. First of all, with respect to the peak value of the opening gesture, the maximum opening is smaller when the devoiceable vowel is surrounded on both sides by single or geminated stops. In addition, what might be more interesting is that the timing of the peak opening is early and relatively fixed for sequences beginning with fracative /s/, whereas the timing for words beginning with /k/ is comparatively late and more variable than for the /s/ group. Incidentally, it is evident that, except for the word containing /siQs/, all the utterance types are equally characterized by a single peaked monomodal pattern. In other words, only the type /siQs/ is unique, in that the curve has a plateau-type or two peaks, as stated before. As for the speed of the glottal movement, it seems faster for the solid lines, i.e. those for the /s/ group, than for the dotted lines of the /k/ group, in general. In order to reveal the details of the characteristics of the velocity patterns, Fig. 7 shows the velocity curves for all the utterance types. These plots were made by successive substractions at 5 ms increments of the glottal width change, using the displacement data in Fig. 6. Positive numbers indicate abduction and negative numbers mean adduction. The bottom two graphs are again grouped according to the nature of the initial segments. It is clear that the velocity during the opening phase is faster for sequences beginning with a voiceless fricative than for those beginning with a voiceless stop. Moreover, another interesting finding is the timing of the peak abduction velocity. The location of peak abduction velocity is almost fixed across both groups, irrespective of the difference in peak amplitudes. Taken together, we may conclude that, although the peak velocity as well as the peak displacement and its timing are all clearly different between the /s/ and /k/ groups, the
8
H. Yoshioka, A. Lofqvist and H. Hirose
"'
.<::: c
Q)
"-
0
"'
c
·u; 0
u "0 Q)
·100
"'
"-
0
(/)
Voicing
Figure 7
10 0
200
30 0
400
offset of preceding vowel ( ms )
Superimposed curves of the first derivative of the averaged glottograms for all the test voiceless sequences.
timing of peak abduction velocity is more or less constant in relation to the line-up point, i.e. the voicing offset of the vowel preceding the voiceless sequence.
Discussion There are several experiments directed towards understanding the laryngeal adjustments during Japanese voiceless sequence production at both movement and electromyographic levels. For example, Sawashima (1969) showed photo-electric glottograms for single tokens using two native speakers of the Tokyo dialect. According to the data, when the devoiced vowel occurred between two voiceless fricatives , the glottographic patterns tended to be characterized by a slight depression in the middle of the curve for one subject, while the other subject showed a single peaked pattern even in fricative environments. In his later study using fiberoptic filming (Sawashima, 1971), a more comprehensive examination was made , including combinations such as /kis/ and /kik/ which were also used in the present study. He concluded that the fiberoptic data for these voiceless sequences were all characterized by a single peaked curve, although the utterance list did not contain the one having a devoiced vowel surrounded on both sides by voiceless single and/or geminated fricatives. Recently, Sawashima and his colleagues have reported on simultaneous fiberoptic and electromyographic recordings for single tokens , showing a two peaked patterns during /siQs/ sequence production for two subjects (Sawashima et al., 1978). The present results, although limited to a single object and presented as ensemble-averages, appear to be generally in good agreement with the previous works. When the devoiced vowel occurs between a voiceless fricative and a geminated one, such as /siQs/, the glottal opening gesture may be characterized by a bimodal or, at least, a plateau-type pattern in terms of a temporal closing movement or a temporal cessation of the continuous opening movement
Laryngeal adjustments in Japanese
9
during the voiceless sequence. In contrast, all the other glottal opening patterns during voiceless sequence production are characterized by a rather simple, single peaked pattern. Of course, it should be taken into consideration that these findings might reflect speakerspecific and/or token-specific aspects (e.g. Sawashima, 1969). Nevertheless, it is always found, including in the other studies of Japanese, that a voiceless fricative environment, and typically the one containing a geminate, seems to require two separate opening gestures. In addition, the current data also reveal the detailed characteristics of the averaged photoelectric grottograms, demonstrating the dependence of the abduction gesture on the phonetic nature of the segments. When the voiceless sequence contains a voiceless fricative /s/, the peak value of the glottal opening is larger than that for the one without a fricative. Moreover, the timing of first peak opening varies according to the property of the initial segments; early and relatively fixed for the fricative initial group , and late and more variable for the stop initial group. This finding is also consistent with our recent studies using American English (Yoshioka et al., 1979), Icelandic (Lofqvist & Yoshioka, in press) and Swedish (LOfqvist & Yoshioka, 1980), although the phonologies Of these languages differ among other things in the significance of stop aspiration. Therefore, we are inclined to conclude that at least the difference in the peak value between a voiceless fricative and a voiceless stop is universal. Furthermore, the plots of the velocity curves add another new dimension. Despite the clear difference of the peak value of the velocity between stop and fricative initial groups, the peak timing of the abduction velocity is almost fixed across the two groups. It should be mentioned that the line-up point was determined as the voicing offset of the preceding vowel regardless of the nature of the initial voiceless segment. In considering the fact that the glottis is usually slightly open at this moment, in particular when the initial segment is a voiceless fricative, peak velocity for the fricative initial group might occur a little later than for the stop group, if the beginning of the opening movement, defined as the inflection point in the movement curve, was chosen as the lineup point. These results in conjunction with other studies of ours using different languages may be interpreted in several ways. From a phonetic viewpoint, the faster and larger opening for a voiceless fricative may be related to the necessary supply of air during the voiceless fricative segment to produce adequate turbulent noise by a quick reduction of laryngeal resistance. On the other hand, in order to stop glottal vibrations at the implosion of a voiceless stop and to assist in the buildup of oral pressure, a slight opening gesture may be sufficient in combination with the closing gesture of the supraglottal articulators. As for the fixed timing of peak abduction velocity, regardless of the phonetic nature of the sequence, the interpretation seems open. From a physiological aspect, it is however possible that it reflects a basic nature of the voluntary movement control of the glottis particularly in relation to oral gestures. It could be the case that the velocity timing control is physiologically constrained in the temporal domain, while the adequate degree of velocity and displacement are attained within such a temporal framework tel meet different phonetic requirements. This work was supported by NINCDS Grants NS-13617 and NS-13870, and BRS Grant RR-05596 to Haskins Laboratories. We are grateful to David Zeichner and Richard Sharkany for technical assistance, to Agnes McKeon for preparing the illustrations, and to Ruiko Inoue for typing the manuscript. References Basmajian, J. & Stecko, G. (1962). A new bipolar indwelling electrode for electromyography. Journal of Applied Physiology, 17, 849.
10
H. Yoshioka, A. Lofqvist and H. Hirose
Fujimura, 0. (1971). Hatsuon no katei. Japan Journal of Logopedics and Phoniatrics, 12, 11- 21. Han, M.S. (1962). Unvoicing of vowels in Japanese. Study of Sound, 10,81- 100. Hattori, S. (1951). Onseigaku. Tokyo: Iwanamishoten. Hirano, M. & Ohala, J. (1969). Use of hooked-wire electrodes for electromyography of the intrinsic laryngeal muscles. Journal of Speech and Hearing Research, 12, 362- 373. Hirose, H. (1976). Posterior cricoarytenoid as a speech muscle. Annals of Otology, Rhinology and Laryngology, 88, 334-343. Hirose, H., Gay, T., Strome, M. & Sawashima, M. (1971). Electrode insertion technique for laryngeal electromyography. Journal of the Acoustical Society of America, 50, 1449-1450. Kewley·Port, D. (1977). EMG signal processing for speech research. Haskins Laboratories Status Report on Speech Research, SR-50, 123- 146. Lofqvist, A. & Yoshioka, H. (1980). Laryngeal activity in Swedish obstruent clusters. Journal of the Acoustical Society of America, 68 , 792-801. Lofqvist, A. & Yoshioka, H. (in press). Laryngeal activity in Icelandic obstruent production. The Nordic Languages and Modern Linguistics (Hovdhaugen, E. ed.) Vol. 4. Oslo : Universitetsforlaget. Sawashima, M. (1969). Devoiced syllables in Japanese : a preliminary study by photo-electric glottography. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 3, 35--41. Sawashima, M. (1971). Devoicing of vowels. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 5, 7- 13. Sawashima, M. (1973). Hatsuonji no koto chosetsu. Onseikagaku (Hiki, S. ed.). Tokyo: Tokyo University Press. Sawashima, M. (1976). Current instrumentation and techniques for observing Speech organs. Technocrat, 9--4, 19-26. Sawashima, M. & Hirose, H. (1968). A new laryngoscopic technique by use of fiberoptics. Journal of the Acoustical Society of America, 43, 168-169. Sawashima, M., Hirose, H. & Yoshioka, H. (1978). Abductor (PCA) and adductor (INT) muscles of the larynx in voiceless sound production. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 12, 5 3-60. Yoshioka, H. (1979). Laryngeal adjustments during Japanese fricative and devoiced vowel production. Haskins Laboratories Status Report on Speech Research, SR-58, 14 7-160. Yoshioka, H., Lofqvist, A. & Hirose, H. (1979). Laryngeal adjustments in the production of consonant clusters and geminates in American English. Haskins Laboratories Status Report on Speech Research, SR-59/60, 127-151.
Laryngeal adjustments in Japanese voiceless sound production Hirohide Yoshioka Haskins Laboratories, 270 Crown Street, New Haven, Connecticut 06510, US.A. and Research Institute of Logopedics and Phoniatrics, Faculty of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113, Japan
Anders Lofqvist Haskins Laboratories, 2 70 Crown Street, New Haven Connecticut 06510, US. A. and Department of Phonetics, Lund University, Helgonabacken 12, S-223 62 Lund, Sweden
Hajime Hirose Research Institute of Logopedics and Phoniatrics, Faculty of Medicine, University of Tokyo, 7-3-1 Hongo, Bunkyo, Tokyo 113, Japan Received 11th October 1980
Abstract:
As part of a series of investigations on the production of sequence of unvoiced sounds in different languages, the current experiment was conducted using the combined techniques of photo-electric glottography, fiberoptic filming and laryngeal electromyography. Particular attention was paid to devoiced vowel production in various voiceless consonantal environments including geminates. The data show that the glottal opening gesture during a voiceless sequence containing a devoiced vowel is characterized by a monomodal pattern, unless the vowel occurs between a voiceless fricative and a geminated one, such as /siQs/, where a bimodal pattern may occur. The movement results also suggest that the velocity and size of the glottal opening gesture vary according to the nature of the adjacent voiceless obstruents. The speed of the opening phase is slow when a stop precedes the vowel, and fast when a fricative precedes it. The peak glottal opening attained during the devoiced vowel is larger when a fricative either precedes or follows than when the vowel is surrounded on both sides by single or geminated stops. Furthermore, it is revealed that the. peak velocity of the initial opening gesture occurs at almost the same time in relation to the voicing offset of the preceding vowel, regardless of the properties of the surrounding voiceless obstruents and, thus, irrespective of variations in the magnitude of velocity and opening size.
Introduction At the 97th meeting of the Acoustical Society of America, we reported how voiceless sound sequences, such as voiceless obstruent clusters, are organized in terms of their glottal opening and closing gestures, using native speakers of American English (Yoshioka et al., 1979) and Swedish (Lbfqvist & Yoshioka, 1980). The conclusion of those studies was that, in the production of sequential unvoiced sounds, the glottal opening gesture is characterized by a one, two or more than two peaked pattern in a regular fashion according to the nature of 0095-44 70/82/010001 + 10 $02.00/0
© 1982 Academic Press Inc. (London) Ltd.
2
H. Yoshioka, A. Lofqvist and H. Hirose
the voiceless segments. A voiceless obstruent characterized by aspiration or frication noise tends to require a separate opening gesture, while an unaspirated stop in a voiceless environment can be produced within the opening gesture attributed to an adjacent aspirated stop or fricative. For example, an /sk#sk/sequence in English was produced in most cases with two separate opening gestures. In contrast, an /sks#k/ string was in general accompanied by three opening gestures (Yoshioka et al., 1979). Furthermore, the velocity of the initial opening movement was shown to vary depending on the properties of the initial voiceless segment. When the first unvoiced segment in the cluster was a fricative, the speed of the abduction was significantly faster than when the initial voiceless sound was an aspirated or unaspirated stop, regardless of the nature of the following voiceless segments. This also meant that the difference in velocity during the initial opening phase heid true despite the fact that, for most cases beginning with a voiceless unaspirated stop, peak glottal opening occured during the following fricative segment. In order to examine the validity of these notions across different languages, the current experiment was carried out using the same combined techniques of photo-electric glottography, fiberoptic filming and laryngeal electromyography, in co-operation with a native speaker of Japanese. The phonology of the Japanese does not allow voiceless "pure" obstruent clusters other than geminates. Syllable-final obstruents also rarely occur in this language . On the other hand, in conversational speech of the Tokyo dialect, there is a well-known phenomenon of vowel devoicing in that a high vowel, such as /i/ and /u/, surrounded by voiceless obstruents on both sides is often produced without any vocal fold vibrations during the vowel segment (e.g. Hattori, 1951; Han, 1962; Fujimura, 1971; Sawashima, 1973). Therefore, we paid particular attention to devoiced vowel production in various voiceless consonantal environments including geminates.
Method and procedure The techniques used in the present experiment were simultaneous recordings of photoelectric glottography, fiberoptic filming and laryngeal electromyography (EMG), in parallel with the audio signal. The EMG data were obtained using bipolar hooked-wire electrodes (Basmajian & Stecko, 1962; Hirano & Ohala, 1969). The electrodes, consisting of a pair of platinum- tungsten alloy wires (50 pm in diameter with isonel coating), were inserted perorally into the posterior cricoarytenoid muscle (PCA) under indirect laryngoscopy with the aid of a specially designed curved probe (Hirose et al., 1971 ) . Before insertion, topical anaesthetic was applied to the mucous membrane of the hypopharynx using a small amount of 4% Lidocaine spray (Xylocaine ). For verification of electrode position, the subject was instructed to perform several non-speech and speech maneuvers that are well understood in terms of the manner of the PCA involvement, such as inspiration and expiration during breathing, voluntary swallowing, pitch changes including register shifts, glottal attack, and voiced-voiceless sound constrasts. The EMG signal was monitored on an oscilloscope not only during the verification gestures but also during the entire recording session. The interference voltages of the EMG signals, after high-pass filtering at 80Hz, were recorded on a multi-channel FM recorder together with the audio signal. After full-wave rectification and integration over a 5 ms time window , the action potentials were fed into a computer at a sampling rate of 200Hz for further processing to obtain the muscle activity patterns for ensemble-averaged tokens with a 35 ms time constant (Kewley-Port, 1977). The figures to be presented in this paper represent activity patterns aligned with reference to the voicing offset of the vowel preceding the voiceless sequence.
Laryngeal adjustments in Japanese
3
For the movement data, the glottal view through a flexible laryngeal fiberscope (Olympus VF-0 type, 4.5 mm in outer diameter) was photographed with a cine camera at a rate of 60 frames/s . A synchronization signal was registered on the FM recorder to identify each frame. Then, frame by frame analyses were made with the aid of mini-computer to calculate the distance between the vocal processes; this distance is considered one of the indicators of glottal width (Sawashima & Hirose, 1968; Sawashima, 1976). A cold DC light source (Olympus CLS), providing illumination of the upper glottal area, also served as the light source for the photo-electric glottography. The amount of light passing through the glottis was sensed by a photo-transistor (Philips BPX 81) placed on the neck just below the lower edge of the cricoid cartilage. The electrical output was recorded on another channel of the FM tape. These signals were sampled at 200Hz and processed with the digital system. A native male speaker of the Tokyo dialect, one of the authors, served as the subject. Among the various voiceless environments surrounding a devoiceable vowel /i/, the combination of /s/ and /k/ is optimum in forming the greatest possible number of meaningful words in Japanese. Therefore , as is shown in Table I. We chose the test words that contain a devoiceable vowel /i/ in the middle of voiceless obstruents composed of the phonemes /s/ and /k/. For example , the production of the first word in this list, /kikee/, which means "anomaly" in Japanese , may be transcribed as having an unvoiced string [kbk] --a [k] plus a devoiced [i] plus a slightly aspirated [k]. Table I
Test words and the carrier sentence (Q =geminate phoneme)
Japanese
r-rtt,;J:. _ ~
_ :r-9J
"sorewa
_desu" ( unvoiced string)
ff< :!
/kik ee /
5
/ k iQk ee /
[kj kk )
~J!
lfiil
/ k isee /
[kjs )
5
1!t If<
/ kiQsee /
[ kj ss )
~~ ~
!ill
/ siQkee /
/ sikee /
[kj k )
c IJk l c1j kk )
~
~
/ sisee /
[ \ js)
~
iEi1
I s iQsee /
[I j ss)
For the first two to three repetitions of each test word embedded in the frame sentence, "sorewa ------ desu." , "we call it ------ -." , simultaneous recordings of photo-electric output and fiberoptic filming were made together with the audio signals, followed by 14- 28 additional recordings of only EMG and photo-electric signals. During the latter part of the session, the glottal image was constantly monitored through the fiberoptic view finder. Such careful monitoring is mandatory to obtain reliable interpretations of large amounts of photo-electric recordings, as we have discussed elsewhere (Yoshioka eta!., 1979; LOfqvist & Yoshioka, 1980). Results Figure 1 illustrates the results for the test word /siQsee/. Since the glottal opening patterns obtained by photo-electric glottography have been shown to be practically identical to those
4
H. Yoshioka, A. Lofqvist and H. Hirose /siQsee/ [Jjsse: ) x 28--------------- [/isse :) x 2
G:JJ'\
L
h_ L
I
Pc:oo~~ AE
t.4f~ Voicing offset of preceding vowel
Figure 1
Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentence containing the test word /siQsee/. Twenty-eight devoiced tokens (left) and two voiced tokens (right), respectively.
obtained by plotting the distance between the vocal processes from the fiberoptic cine-films, we will focus on the photo-electric glottograms as an index of glottal width change during the pertinent voiceless sequence production. Here, the top row (GW) represents the averaged glottograms for two allophonic groups. One is devoiced, and the other is voiced. Among the 30 repetitions, 28 tokens were produced with devoiced vowels, while only two of them had fully voiced vowels. These variations were easily detectable in audio waveforms, in sound spectrograms and by listening the recorded tape. As for laryngeal EMG, the figure contains the corresponding averaged activity patterns of the abductor muscle -posterior cricoarytenoid muscle (PCA)- that has been demonstrated to substantially control the glottal aperture (Hirose, 1976; Yoshioka, 1979). These signals were aligned with respect to the voicing offset of the preceding vowel. It is obvious that, when it is uttered with a fully voiced vowel, two clear separate glottal opening gestures are found for the /siQs/ production at both the movement and the electromyographic levels. In contrast, the averaged curves for the devoiced group is a little unclear. The abductor muscle (PCA) activity curve in the middle may be characterized by two opening gestures : The first is associated with a high and steep peak, followed by a second that is broad but of moderate activity level. The glottographic pattern for this devoiced group is more complicated, in that one might describe it as having two peaks or, alternatively, a sort of plateau. Since all the other test words, except the one containing /siQs/ mentioned above, were always produced with a devoiced vowel, all the averaged curves henceforth from Fig. 2 are those for completely devoiced groups. Figure 2 shows the averaged glottographic pattern and the corresponding averaged abductor muscle activity pattern for the devoiced /sis/ sequence, in comparison with those for the devoiced /siQs/ shown in Fig. 1 and repeated in Fig. 2. Here, several points are worth mentioning. First, the averaged glottogram for the non-geminated /sis/ is clearly distinguished by a monomodal curve, while that for the geminated /siQs/ is characterized by a broad or bimodal pattern as mentioned above. This finding seems to be reflected in the EMG. The averaged PCA activity curve for the non-geminated /sis/ has a single peak around the line-up point, while that for the geminated /siQs/ is, as mentioned before, characterized by two separate activity patterns. In addition, despite the
Laryngeal adjustments in Japanese /sisee/
5
/siQs ee/
G:vbJ~ [\~
PC~l~J~~ AEf ~ [,,~ A
Voicing offset of preceding vowel
Figure 2
A
~
~
Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /sisee/ and /siQsee/, respectively.
differences in the overall modality between these two utterance types at both movement and EMG levels, the initial opening phases are quite similar. The peak glottal openings are approximately of the same size and reached almost at the same time. As for the PCA activity, both the curves have their peaks at a same time, i.e. the line-up point. Figure 3 . contains the activity patterns for a devoiced vowel /i/ surrounded on both sides by a pair of single or geminated stops. In comparison with those for the devoiced vowel /i/ occurring between voiceless fricatives shown in Fig. 2, the glottographic curves for these cases have a single and considerably smaller peak. The slopes of the glottographic curves for this stop group are also more gradual than for those surrounded by voiceless fricatives in Fig. 2 . It implies that the glottal opening gesture during the voiceless sequence containing a devoiced vowel may vary according to the nature of the surrounding consonants; slow for the voiceless stop and fast for the voiceless fricative.
/ki keel
G:1:~ u
/kiQkee/
hL
P~~o~~ AEf
JA& ....
Voicing offset of preceding vowel
Figure 3
f..~ .... Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /kikee/ and /kiQkee/, respectively.
6
H. Yoshioka, A. Lofqvist and H. Hirose /sikee/
G:J~\1
/siQkee/
(\ ~
PC3:0~~ AE
t
~
f
Voicing offset of preceding vowel
Figure 4
rh•.>D}.. Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelope curves for the sentences containing the devoiced test words /sikee/ and /siQkee/, respectively.
Figures 4 and 5 show the patterns for the utterance types that contain a devoiceable vowel /i/ in a voiceless sequence composed of two different obstruents, such as /sik/, /siQk/, /kis/ and /kiQs/. It is evident that all the glottographic curves are characterized by a monomodal pattern. In addition, and more interestingly, the difference in the slopes during the initial opening phase depends on the phonetic properties of the initial segments. When a voiceless fricative precedes the devoiced vowel, the opening movement is faster than when a voiceless stop precedes the vowel. Furthermore, peak glottal opening during these devoiced sequences coincides approximately with the peak amplitude of the audio envelope during the devoiced vowel segments. As for the EMG signals, the noise level is too high for a detailed discussion. Nevertheless, it may be mentioned that the peak PCA activity for these utterance types, as well as the others mentioned above, occurs around the line-up, i.e. at the voicing offset of the preceding vowel, regardless of utterance type.
/kiQsee/
/kisee/
G:v~~
P::o~~ AEt
~ Voicing offset of preceding vowel
Figure 5
t
~ Voicing offset 100 ms of preceding vowel
Averaged glottograms, PCA activity patterns and audio envelop curves for the sentences containing the devoiced test words /kisee/ and /kiQsee/, respectively.
Laryngeal adjustments in Japanese
Voicing offset of pr eceding vowel
Figure 6
7
( ms)
Superimposed curves of the averaged glottograms for all the test voiceless sequences.
For a detailed comparison of the characteristics of the glottal opening gesture for all the utterance types containing various combinations of voiceless sounds, Fig. 6 presents all the glottal movement data superimposed during the pertinent voiceless portions. These averaged curves are again aligned with respect to the voicing offset of the preceding vowel in the frame sentence. The solid lines represent the voiceless sequences beginning with a fricative, while the group of the dotted lines corresponds to those beginning with a voiceless stop. The two bottom graphs show separately these two groups, i.e. the sequences beginning with /s/ and /k/, respectively. First of all, with respect to the peak value of the opening gesture, the maximum opening is smaller when the devoiceable vowel is surrounded on both sides by single or geminated stops. In addition, what might be more interesting is that the timing of the peak opening is early and relatively fixed for sequences beginning with fracative /s/, whereas the timing for words beginning with /k/ is comparatively late and more variable than for the /s/ group. Incidentally, it is evident that, except for the word containing /siQs/, all the utterance types are equally characterized by a single peaked monomodal pattern. In other words, only the type /siQs/ is unique, in that the curve has a plateau-type or two peaks, as stated before. As for the speed of the glottal movement, it seems faster for the solid lines, i.e. those for the /s/ group, than for the dotted lines of the /k/ group, in general. In order to reveal the details of the characteristics of the velocity patterns, Fig. 7 shows the velocity curves for all the utterance types. These plots were made by successive substractions at 5 ms increments of the glottal width change, using the displacement data in Fig. 6. Positive numbers indicate abduction and negative numbers mean adduction. The bottom two graphs are again grouped according to the nature of the initial segments. It is clear that the velocity during the opening phase is faster for sequences beginning with a voiceless fricative than for those beginning with a voiceless stop. Moreover, another interesting finding is the timing of the peak abduction velocity. The location of peak abduction velocity is almost fixed across both groups, irrespective of the difference in peak amplitudes. Taken together, we may conclude that, although the peak velocity as well as the peak displacement and its timing are all clearly different between the /s/ and /k/ groups, the
8
H. Yoshioka, A. Lofqvist and H. Hirose
"'
.<::: c
Q)
"-
0
"'
c
·u; 0
u "0 Q)
·100
"'
"-
0
(/)
Voicing
Figure 7
10 0
200
30 0
400
offset of preceding vowel ( ms )
Superimposed curves of the first derivative of the averaged glottograms for all the test voiceless sequences.
timing of peak abduction velocity is more or less constant in relation to the line-up point, i.e. the voicing offset of the vowel preceding the voiceless sequence.
Discussion There are several experiments directed towards understanding the laryngeal adjustments during Japanese voiceless sequence production at both movement and electromyographic levels. For example, Sawashima (1969) showed photo-electric glottograms for single tokens using two native speakers of the Tokyo dialect. According to the data, when the devoiced vowel occurred between two voiceless fricatives , the glottographic patterns tended to be characterized by a slight depression in the middle of the curve for one subject, while the other subject showed a single peaked pattern even in fricative environments. In his later study using fiberoptic filming (Sawashima, 1971), a more comprehensive examination was made , including combinations such as /kis/ and /kik/ which were also used in the present study. He concluded that the fiberoptic data for these voiceless sequences were all characterized by a single peaked curve, although the utterance list did not contain the one having a devoiced vowel surrounded on both sides by voiceless single and/or geminated fricatives. Recently, Sawashima and his colleagues have reported on simultaneous fiberoptic and electromyographic recordings for single tokens , showing a two peaked patterns during /siQs/ sequence production for two subjects (Sawashima et al., 1978). The present results, although limited to a single object and presented as ensemble-averages, appear to be generally in good agreement with the previous works. When the devoiced vowel occurs between a voiceless fricative and a geminated one, such as /siQs/, the glottal opening gesture may be characterized by a bimodal or, at least, a plateau-type pattern in terms of a temporal closing movement or a temporal cessation of the continuous opening movement
Laryngeal adjustments in Japanese
9
during the voiceless sequence. In contrast, all the other glottal opening patterns during voiceless sequence production are characterized by a rather simple, single peaked pattern. Of course, it should be taken into consideration that these findings might reflect speakerspecific and/or token-specific aspects (e.g. Sawashima, 1969). Nevertheless, it is always found, including in the other studies of Japanese, that a voiceless fricative environment, and typically the one containing a geminate, seems to require two separate opening gestures. In addition, the current data also reveal the detailed characteristics of the averaged photoelectric grottograms, demonstrating the dependence of the abduction gesture on the phonetic nature of the segments. When the voiceless sequence contains a voiceless fricative /s/, the peak value of the glottal opening is larger than that for the one without a fricative. Moreover, the timing of first peak opening varies according to the property of the initial segments; early and relatively fixed for the fricative initial group , and late and more variable for the stop initial group. This finding is also consistent with our recent studies using American English (Yoshioka et al., 1979), Icelandic (Lofqvist & Yoshioka, in press) and Swedish (LOfqvist & Yoshioka, 1980), although the phonologies Of these languages differ among other things in the significance of stop aspiration. Therefore, we are inclined to conclude that at least the difference in the peak value between a voiceless fricative and a voiceless stop is universal. Furthermore, the plots of the velocity curves add another new dimension. Despite the clear difference of the peak value of the velocity between stop and fricative initial groups, the peak timing of the abduction velocity is almost fixed across the two groups. It should be mentioned that the line-up point was determined as the voicing offset of the preceding vowel regardless of the nature of the initial voiceless segment. In considering the fact that the glottis is usually slightly open at this moment, in particular when the initial segment is a voiceless fricative, peak velocity for the fricative initial group might occur a little later than for the stop group, if the beginning of the opening movement, defined as the inflection point in the movement curve, was chosen as the lineup point. These results in conjunction with other studies of ours using different languages may be interpreted in several ways. From a phonetic viewpoint, the faster and larger opening for a voiceless fricative may be related to the necessary supply of air during the voiceless fricative segment to produce adequate turbulent noise by a quick reduction of laryngeal resistance. On the other hand, in order to stop glottal vibrations at the implosion of a voiceless stop and to assist in the buildup of oral pressure, a slight opening gesture may be sufficient in combination with the closing gesture of the supraglottal articulators. As for the fixed timing of peak abduction velocity, regardless of the phonetic nature of the sequence, the interpretation seems open. From a physiological aspect, it is however possible that it reflects a basic nature of the voluntary movement control of the glottis particularly in relation to oral gestures. It could be the case that the velocity timing control is physiologically constrained in the temporal domain, while the adequate degree of velocity and displacement are attained within such a temporal framework tel meet different phonetic requirements. This work was supported by NINCDS Grants NS-13617 and NS-13870, and BRS Grant RR-05596 to Haskins Laboratories. We are grateful to David Zeichner and Richard Sharkany for technical assistance, to Agnes McKeon for preparing the illustrations, and to Ruiko Inoue for typing the manuscript. References Basmajian, J. & Stecko, G. (1962). A new bipolar indwelling electrode for electromyography. Journal of Applied Physiology, 17, 849.
10
H. Yoshioka, A. Lofqvist and H. Hirose
Fujimura, 0. (1971). Hatsuon no katei. Japan Journal of Logopedics and Phoniatrics, 12, 11- 21. Han, M.S. (1962). Unvoicing of vowels in Japanese. Study of Sound, 10,81- 100. Hattori, S. (1951). Onseigaku. Tokyo: Iwanamishoten. Hirano, M. & Ohala, J. (1969). Use of hooked-wire electrodes for electromyography of the intrinsic laryngeal muscles. Journal of Speech and Hearing Research, 12, 362- 373. Hirose, H. (1976). Posterior cricoarytenoid as a speech muscle. Annals of Otology, Rhinology and Laryngology, 88, 334-343. Hirose, H., Gay, T., Strome, M. & Sawashima, M. (1971). Electrode insertion technique for laryngeal electromyography. Journal of the Acoustical Society of America, 50, 1449-1450. Kewley·Port, D. (1977). EMG signal processing for speech research. Haskins Laboratories Status Report on Speech Research, SR-50, 123- 146. Lofqvist, A. & Yoshioka, H. (1980). Laryngeal activity in Swedish obstruent clusters. Journal of the Acoustical Society of America, 68 , 792-801. Lofqvist, A. & Yoshioka, H. (in press). Laryngeal activity in Icelandic obstruent production. The Nordic Languages and Modern Linguistics (Hovdhaugen, E. ed.) Vol. 4. Oslo : Universitetsforlaget. Sawashima, M. (1969). Devoiced syllables in Japanese : a preliminary study by photo-electric glottography. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 3, 35--41. Sawashima, M. (1971). Devoicing of vowels. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 5, 7- 13. Sawashima, M. (1973). Hatsuonji no koto chosetsu. Onseikagaku (Hiki, S. ed.). Tokyo: Tokyo University Press. Sawashima, M. (1976). Current instrumentation and techniques for observing Speech organs. Technocrat, 9--4, 19-26. Sawashima, M. & Hirose, H. (1968). A new laryngoscopic technique by use of fiberoptics. Journal of the Acoustical Society of America, 43, 168-169. Sawashima, M., Hirose, H. & Yoshioka, H. (1978). Abductor (PCA) and adductor (INT) muscles of the larynx in voiceless sound production. Research Institute of Logopedics and Phoniatrics, University of Tokyo, Annual Bulletin, 12, 5 3-60. Yoshioka, H. (1979). Laryngeal adjustments during Japanese fricative and devoiced vowel production. Haskins Laboratories Status Report on Speech Research, SR-58, 14 7-160. Yoshioka, H., Lofqvist, A. & Hirose, H. (1979). Laryngeal adjustments in the production of consonant clusters and geminates in American English. Haskins Laboratories Status Report on Speech Research, SR-59/60, 127-151.