No evidence of a dependency on the position of the vowel within the sentence or the sentence type (as defined in Fig.33) was found for any of the established parameters[14].
As expected, the variation in the results strongly depends on the speaker's gender and the vowel type. To provide a well-balanced analysis, the measurments were analyzed in four groups, dividing the data into male and female /i/'s and /a/'s. The data were then analyzed using repeated measure ANOVA with the factor "accent type" (four levels: H%, H*L, L% and L*H). An overview of the results is given in Table 13.
A dependency on gender was found for almost all EGG waveform shape parameters. This general relation is the same as the one described in section 17.
The most consistent effect was found for the fundamental frequency and the Open Quotient. F0 distinguishes between low and high tones, but the differences between boundary and pitch tones are statistically negligible (for both vowels and speaker groups). For all voices the fundamental frequency of high tones (H) was higher than that of low tones (L). The mean values of the F0 of boundary tones are more extreme than their pitch accent counterparts, i.e. the high boundary tones are higher and the low boundary tones are lower than the corresponding pitch accent tones. In other words, L*H tones are higher than L% tones and that H% is higher than H*L.
The standard deviation of the fundamental frequency across the analyzed vowels changes significantly, but, when related to the F0, it remains almost unchanged. The jitter of boundary tones exceeds that of pitch accents jitter, especially for males. This may be due either to a speaker's relaxed larynx adjustments at the end of an utterance or to faster pitch changes. The fundamental frequency does not depend on the vowel type.
For female speakers both Open Quotient measures (OQI and OQII) significantly vary between tones for both vowels. For males, however, only the second measure of the Open Quotient shows serious changes as far as tones are concerned. The Open Quotient is directly proportional to the fundamental frequency.
The OQI is about 8% lower for low tones than for high tones within the female test group. This difference is generally repeated in the second measure of the Open Quotient. For the male speakers the differences are significant only for OQII (the duration of the no-contact phase related to the duration of the whole pitch period) and resemble the tendency established for female speakers. Interestingly, the change in the duty ratio of the EGG waveform is much smaller between pitch accents than between boundary tones (for males the difference of the OQII between low and high pitch accents is as small as 2%).
The difference in the Open Quotient agrees with the expectations based on the modelling of the larynx. Titze and Talkin (1979) as well as Ishizaka and Flanagan (1972) explain this effect with greater vocal fold tension for H tones compared to L tones (cricothyroid contraction mechanism). Also, as expected (Holmberg et al., 1995), the OQ is smaller for males than for females. For female speakers the correlation between F0 and OQII is equal to 0.7, whereas for male speakers it amount to r=0.6. For male speakers the relation of the OQI to F0 is even stronger (r=0.74) as for the OQII, for females remains approximately unchanged (0.65). Of course, the OQ should not be seen as a simple consequence of changes in the fundamental frequency. We should for example recall the relationship between OQ and vocal effort already mentioned in section 17.2.
The relation between the Speed Quotient (defined as the skewness of the EGG pulse) and the accent type shows a complicated pattern. The value of the index depends on gender, accent type and even vowel type. For female speakers SQ decreases slightly from /a/ to /i/ (what corresponds well with word stress experiment, see Fig.27), but the dependence on accent type is significant only for /i/. For "female" /i/ the SQ is almost two times lower (stronger skewness) for low tones than for high tones. For males the SQ does not depend on the pitch level. However, it depends on the vowel type. In summary it can be said that the electroglottographic pulses do not exhibit a transformation of symmetry caused by a higher F0, but that the symmetry depends on the vowel type and the speaker's gender. In the preceding experiment (word stress) the skewness of the EGG pulse was inversely proportional to the word stress level, i.e. pulses were more symmetrical for stressed vowels. Stressed vowels had also higher F0, thus, it is to be expected that the glottal pulses of a higher F0 will exhibit higher SQ values also i this experiment, but it is not true. This discrepancy in the results between word stress and pitch accent gives a first hint of the different physiological mechanisms of word stress and intonation.
The peak-to-peak amplitude of male speakers is greater compared to that of females, but it does not change across vowels or tones.
Contrary to the results for word stress, the slopes of the glottal contact pulse do not correlate strongly with the pitch accent. For /i/ the slope of the "end of closing" phase is steeper for high tones than for low tones. However, for /a/ no effect was registered. This agrees with the tendency observed in the word stress experiment, where for /a/ the slopes did not show any trend, whereas for /i/ there was a strong correlation with vocal effort. The falling flank of the EGG pulse is steeper for higher tones in both vowels for female speakers, but does not change significantly for males.
The changes in the other parameters that are due to intonation are of a weak and/or random nature. Also the shape variation parameters which defined as the distance between the original waveform and the straight line model, do not have any significance.
One important aspect should be taken into consideration before comparing our results with the relevant studies of other authors. In this experiment all accented vowels were in the normal lexical stress position so that the effects of the two linguistic categories were superimposed, i.e. the effects of intonational pattern were analyzed for lexically stressed vowels.
However, the results of our intonation experiment are in good agreement with well-known studies. Publications by Pierrehumbert (1989), Koreman (1996) and Sluijter (1995) serve as a basis for comparison. In these studies, the relationship between glottal airflow and intonation in English, Dutch and logotomized speech was investigated. The influence of intonation on glottal behavior has not yet been systematically researched for German.
Pierrehumbert (1989) investigated the effects of varying intonation on the inversely filtered glottal airflow signal in English. The intonation patterns used in her experiment include similiar accent types as the EGG experiment, although no distinction between pitch accents and boundary tones was made. The analyzed accented vowels were located in positions of word stress. In Pierrehumbert's experiment five voice intensities were distinguished, whereas in this study only the normal, moderate voice level was examined. That is why a comparison is only possible for the voice level 3 of her work which corresponds to the normal voice level. There is also difference with respect to the number of subjects (only one male voice in the quoted study) and the number of repetitions (five per intonation contour). The only vowel investigated in her experiment was /a/. Comparable parameters determined in Pierrehumbert's work include the Open Quotient and the skewness of the glottal pulse.
The duty ratio values in our experiment correspond well with her results. As expected, the OQ's are greater for high than for low tones. The mean values of the OQI (determined in the EGG signal) for male speakers and /a/ reflect the same tendency. However, this dependency is not significant (a=0.05) for the difference between high and low tones. It may be possible that voice effort was lower than assumed. An overlapping of OQ ranges also occurrs in Pierrehumbert's examination, if a lower voice intensity is chosen for the comparison[15]. Nevertheless, the OQII, a no-contact ratio of the EGG, correlates perfectly with the acoustic results of Pierrehumbert. In our study, for all speakers and two vowels, the open phase of the vocal folds vibration is relatively longer for an increased fundamental frequency. The absolute values of the OQII are about twice as low as the original Open Quotient computed in the acoustic signal.
Pierrehumbert also detected a higher skewness of the glottal pulse for high tones in comparison to low tones. Such a dependency was not found for the Speed Quotient of EGG waveform. However, the conclusions of Pierrehumbert are fully supported here provided that it is assumed that there is a correlation between the skewness of the airflow waveform and the steepness slopes of the electroglottographic signal. Evidence of this correlation was found in the word stress experiment (section 17). The EGG pulses are significantly steeper in high tones than in the low tones.
The physiological mechanism for the increase in the fundamental frequency consists in the activity of the cricothyroid and thyroarytenoid muscles, their contraction and relaxation[16], which causes a higher longitudinal tension of the vocal folds (Titze & Talkin, 1979; Atkinson, 1978). Contrary to this explanation, Ishizaka & Flanagan (1972) postulate a higher tension of vocalis muscles only as a cause of the increase of F0. But both experimental data (Koreman, 1996:162-164; Slujiter, 1995; Pierrehumbert, 1989; Atkinson, 1978) and theoretical considerations (Titze, 1992; Pesak, 1990) suggest the correctness of the first hypothesis which also claims that there is a higher skewness of the glottal pulse at higher pitch.
Koreman (1996:133-168) investigates the influence of the F0 rise on the glottal source. His analysis is based on logotomized speech and his measurements are made in unstressed syllables. The results of inverse- filtered speech show (beside other parameters which are not comparable with the results of the EGG experiment) the simultaneous increase of the Open Quotient, the glottal airflow pulse skewness and of the maximum amplitude of the time derivative of the flow (excitation strength). These simultaneous increases are linked to the increase in fundamental frequency. In our experiment, the EGG-based parameters exhibit an increase in the Open Quotient and a weak dependency in the skewness and slopes steepness of the EGG pulse with higher pitch on stressed vowels.
In this sense, the study by Sluijter (1995:119-130) supports the EGG measurements. As far as the acoustic analysis is concerned (see section 17.6), the OQ is a correlate of accent rather than of stress. Spectral tilt, however, was found to be a reliable and independent correlate of stress, as well as accent, i.e. of vocal effort. Slujiter also detect an effect of accent on the amplitude of the glottal airflow signal.
In summary, the effect of pitch movement on the electroglottographic signal is statistically significant. The duty ratio of the EGG waveform (measured as the relative duration of the no-contact phase) constitutes the main correlate of pitch accent. The EGG's Open Quotient increases for high tones relative to low tones, whereas it does not change for stress versus unstressed syllables. The steepness of the EGG pulse slopes also depends on intonation and increases with higher F0. However, this dependency is not as obvious as the increase in the Open Quotient. These results are fully supported by acoustic investigations conducted for other languages.
Based on these results, the hypothesis can be formulated that the underlying physiological mechanism of sentence intonation is different from that of word stress. The laryngeal correlate of sentence intonation is the additional modulation of muscular tension (primarily the activity of the cricothyroid muscle), whereas word stress is realized mainly by the build-up of subglottal pressure. However, it should be kept in mind, that in these experiments word level prominece (word stress) and sentence level prominence (intonation) have not been investigated in a completely orthogonal manner. Therefore, it is necessary to interpret the results of the experiments in relation to the prosodic settings only.
14. not all possible combinations of sentence type, pitch accent, boundary tone and vowel were tested.
15. for example the voice level 2 in Pierrehumbert (1989).
16. the mechanism of the fundamental frequency changes is described in detail in section 27.