The discussion deals mainly with the comparison between the results of the acoustic and the EGG signal analysis, and with a general examination of the experiment's results.
In the EGG experiment the same recordings were used as in Claßen et al. (1996). This allows a detailed comparison of the values obtained in both research efforts, although the direct relationship between glottal flow (as measured indirectly by Claßen's group) and EGG is physically and physiologically complex. Apart from Titze (1990) there have not yet been any publications regarding this problem.
The analyses of the inversely filtered speech signal (Koreman, 1996) and the parameters induced from the acoustic signal (Sluijter, 1995; Stevens & Hanson, 1995) indicate that the growing skewness of the glottal pulse is one of the main correlates of an increased vocal effort and thus an important correlate of word stress.
Comparing the results of both the acoustic and the EGG analysis, one should take into account the diverse definitions of the Speed Quotient in both domains. The SQ as computed in the EGG is based only on the comparison of the durations of closing and opening phases. The Speed Quotient (skewness) of the acoustic signal, on the other hand, depends on the glottal volume velocity flow, so that the acoustic parametrization also takes into consideration the strength of the excitation. As the glottal pulse is more skewed (i.e. the closing is much shorter than the opening of the vocal folds), the high frequency region will be emphasized in relation to the low frequency region. The second way of raising the higher frequencies is changing the steepness of the closing portion while the duration remains almost unchanged. As the slope of the closing portion gets steeper, the amplitudes at the mid and high frequencies increase relative to the amplitudes at the low frequencies. From the results of the EGG analysis, it is quite obvious that the difference in spectral tilt between stressed and unstressed vowels corresponds to the changes in the abruptness of closure (the rapid increase of the contact between the folds) rather than to the changes in the duration of the opening and closing phases. The rising contact slope is systematically steeper for stressed vowels, which correlates well with the difference in the spectral tilt.
The relation between acoustic and EGG measurements of the tenseness effect should be viewed with great caution. The significant effect of tenseness is a relative elongation of the rising contact phase. The steepness of slopes also contributes to tenseness, however, the longer contacting phase is more obvious. Of course, the effect of this elongation on the glottal airflow pulse spectrum (while leaving other features unchanged) simply manifests itself in an increased spectral tilt (since the glottal pulse is more symmetrical). This fact was experimentally verified in measurements made by Claßen et al. (1996) and Kingston et al. (1997). This is also reflected in the steepness of the EGG's slopes.
The only supposition which may physiologically explain this is an increased acoustic load caused by a greater striction of the vocal tract above the glottis. The position of the tongue body during the articulation of tense vowels is higher (see section 15.2), which causes a higher acoustic load of the glottal source. This changes the movement of the vocal folds. The closing phase becomes longer, however, no laryngeal adjustments are made. Perhaps this is reflected in the EGG signal, as the only difference lies in the prolongation of the closing phase with proportionally slower changes in the contact area.
Comparing the results of both experiments, one should bear in mind that, in the acoustic domain, the information about laryngeal behavior was extracted only for one designated pitch period, whereas in the EGG the parameters were averaged for the whole vowel length. This may cause additional variation between the results. In the visual inspection of the EGG signals it was noted that especially the first periods of the vowels have a lower amplitude and that the first periods of fold vibration may be affected by aspiration (triggered by preceding [th]). The last EGG period of the vowel has a slightly rounded shape (in the transition to /l/). As mentioned previously, the two first periods and the last period of the vowel were excluded from the analysis. However, it is not guaranteed that all influences of surrounding segments have been eliminated.
The direct relations between both acoustic and EGG parametrizations can be summarized as follows[11] (the acoustic coefficients are described elsewhere (Slujiter, 1995:105; Jessen et al., 1996)):
Table 12: Relations between acoustic and EGG parameters regarding the effects of stress and
tenseness.
Acoustic EGG Physiological description
H1*-A2* steepness of rising skewness of the glottal
contact steepness of pulse, excitation strength
falling contact
H1*-A1 steepness of rising degree of glottal opening
contact steepness of
falling contact
H1*-H2* Open Quotient I Open duty ratio of the glottal
Quotient II (duration of pulse
full open phase)
H1*-A3* duration of rising duration of closing
contact segment portion of glottal pulse
In the physiological domain, the steeper slopes imply that the contact between the vocal folds decreases faster on the stressed vowels than on the unstressed. The faster increase in the contact between the vocal folds in the stressed tokens is equivalent to the more abrupt movement of the folds. The contact area increases and decreases faster, although the duration of the opening/closing gestures (measured relatively to F0) changes only slightly. This implies that the forces acting on the vocal folds folds increase with stress, causing the collision to be deeper along the full width of the fold and/or faster. If the collision is not complete (in the case of breathy voices) the effect should be identical (becuase the folds' motion is faster, although there is no full contact between them). The steeper slopes of the EGG pulses can also depend on the stronger tensions of the folds. The fundamental frequency of the stressed vowels is significantly higher for all speakers (see section 17.1). This means that the vibrating masses can be smaller (see equation (1)) and that the width of folds is also smaller, which is important for the contact area. This may cause a faster growth of contact, even though in that case the maximal amplitude of the EGG should be smaller (since the area becomes smaller). Such an effect was not observed in the experiment. The increase in forces acting on the folds' surfaces is thus the reason for the faster movement. We conclude from this experiment that the steepness of the rise and fall phases of the EGG waveform depends significantly on the stress factor and on the speaker's gender. The possible cause of this effect is an increased vocal effort of the speaker, which, in the laryngeal behavior, manifests itself as an increase in the forces acting on the folds. It can be further hypothesized that it is caused by an increased subglottal pressure rather than by an increased tension of the muscular parts of the folds (Koreman, 1996:156-157). This consideration is discussed in more detail in section 18.4.
The difference in the Open Quotient (of both measures used) between stressed and unstressed vowels although significant is quite small and varies depending on the context. The results of this investigation generally support Sluijter's (1995) conclusion that the Open Quotient is not a primary correlate of stress. It can be postulated that the difference is caused rather by change in intonation (pitch movement on a word spoken in isolation) than that by word stress itself.
A direct comparison of the effects on the EGG waveform using acoustic measurements is possible only for certain vowels, since the acoustic method is not suitable for many vowels. However, the results of the EGG analysis support the acoustic measurements of Claßen et al. (1996) who encountered some effects of stress for male speakers. Koreman (1996:155) concludes that the dependency of the OQ on stress is reflected rather weakly and negatively, i.e. the OQ even tends to decrease in stressed vowels (in our investigation OQ is dependent on tenseness and vowel type).
Assuming that the main mechanism of a more abrupt contact of the vocal folds is an increase in the subglottal pressure, it is to be expected that F0 increases and that the Open Quotient decreases (Ishizaka & Flanagan, 1972; Titze & Talkin, 1979; Strik & Boves, 1992). The only explanation for the observed growth of the Open Quotient is the change in muscle activity, caused particularly by an increased tension of the cricothyroid muscle, which leads to a simultaneous increase in the fundamental frequency and the Open Quotient (see Koreman, 1996:161-162 for the discussion and section 27.3 for the modelling of this effect).
There is only limited evidence for glottal correlates of the tense/lax distinction. The only easily identifiable effect was observed in the durations of the rising contact phases which became longer in almost all vowel groups. For some of the female speakers' vowel groups the slopes of lax vowels are steeper than those of their tense counterparts, which is fully supported by the results of Claßen et al. (1996).
Generally, it can be hypothesized that tenseness is not a distinctive feature of laryngeal behavior. The observed distinctions are caused by a changing acoustic load of the voice source due to tongue height and tongue root position[12] rather than by a changing tension of the laryngeal muscles or a different subglottal pressure.
In our experiment the strongest effects are due to individual differences as well as gender dependencies. Assuming a dependency between excitation strength and the steepness of the contacting slope, a stronger excitation for male speakers can be inferred. Due to differing signal amplification for the respective test subjects, no conclusions can be drawn about the amplitude of the EGG waveform. The differences in timing parameters were especially strong between genders.
One of the major outcomes of this experiment are the big differences between vowel groups. The amplitude as well as the time domain parameters differ noticeably between vowels. This suggests that the waveform of the relative contact area between the vocal folds is related to produced voiced sounds. It also implies that the EGG signal reflects the changes in the configuration of the vocal tract. In other words, the articulatory changes affect the contact between the vocal folds. This falsifies the common view that the EGG waveform does not change across phonemes. On the contrary, the exact parametrization of the signal, as achieved in this study, clearly shows differences between vowels.
11. our conclusions are drawn using also the correlations between mean values. Only the dependencies with r2 >0.25 and the significance a=0.05 were taken into account.
12. it is to be expected that more advanced position of tongue root for tense vowels causes, via hyoid bone, increased tension of the laryngeal muscles. As follows from the simulation experiment (section 27.3), higher tension of the vocal folds should be observed as an increase in Open Quotient, which however was not registered in this experiment.