As described in section 12 the proposed segmentation includes both the parametrization of slopes of the EGG waveform and the difference between the straight-modelled waveform and the original one.
The peak-to-peak amplitude, which is measured in all periods of the electroglottographic waveform, does not depend on gender, tenseness or stress factors, even after standardization of the data (which was necessary due to different settings of the recording equipment). The amplitude is related to the vowel group; the peak-to-peak amplitude for the types of /a/ is significantly smaller than for the /e/,/i/ and /u/ groups.
The shape of the EGG waveform was additionally described in terms of the difference between the original waveform and the straight line model. The area between the lines was computed and stored for each phase of the EGG period. The idea behind this was simply to describe the fluctuations as well as the rounded shape of the signal (for example in the case of tense voice). As expected, the results do not display the dependency on stress or the tense/lax distinction, but are specific to the speaker, as well as to the spoken vowel. This contradicts the expectation that the EGG signal does not change between different phonemes.
The results of the first step of the analysis (as described above) enable us to concentrate on only two parameters that describe the amplitude fluctuation of the EGG waveform: the steepness of closing (increasing contact) and the steepness of opening (decreasing contact). These phases are limited to the end of the closing phase and the start of the opening phase (as defined in section 12), respectively.
The first shape parameter describes the steepness of the closing phase (more precisely: the mean value of the steepness of the end of the closing phase). It has been shown that the energy contained in the acoustic signal is dependent upon the maximal rate of the glottal airflow decrease during the vocal cycle (Fant, 1980). The negative slope of the inversely filtered glottal flow is steepest during the EGG closing phase, and thus Orlikoff (1991) proposes the use of the EGG closing slope as the indicator of the source signal energy gain. This is of course a relative measure which depends on the current system settings. Orlikoff (ibid.) shows a strong correlation between this measure and the sound pressure level. Orlikoff's hypothesis is fully supported by the results of our experiment.
The steepness of the contact rise does not depend significantly on the speaker's gender, but indubitably depends on the stress factor. The effects of other factors interact with stress.
As mentioned before, the signal amplification settings were changed between recordings of speakers, so that a data normalization is necessary to get an unbiased result. After normalization the effect of stress is significant (with F(1,2344)=112.3, p<0.0005) for analyses of vowel type, stress, tenseness and measurement repetition.The significant difference is caused by the type of vowel (F(4,2344)=69.5, p< 0.0005) among other factors. The effect of tenseness is weak and interacts with other variables.
The influence of stress on the standardized steepness of closing is significant for all vowels (Fig.30). As follows from Fig.30 the rise slopes of the EGG pulses for stressed vowels are steeper than those of the unstressed ones. The physiological explanation for this phenomenon is the more abrupt closure of the vocal folds. If the folds are strongly adducted, then the increase in contact is faster because of the faster movement of the folds. The time delay between the motion of the upper and lower margins of the folds is also reduced, which in turn causes the contact to occur on the wider parts of the folds (Ishizaka & Flanagan, 1972:1260). Thus, the change in impedance registered by EGG is more abrupt and the cut-off of the airflow is also faster.
Considering Orlikoff's hypothesis, it is to be expected that the steepness of closing (ECA) is correlated with the acoustic measurements of the spectral tilt. Indeed, comparing the EGG measurements with the findings of Claßen et al. (1996), a strong correspondence was established.
The results in Claßen et al. (1996) are presented separately for stress and tense/lax tokens. A direct comparison with the EGG data can be made using the results given in App.I.
The ECA parameter corresponds best to the skewness of the glottal pulse. However, the acoustic correlates of the degree of the glottal opening and the rate of closure also show similar patterns.
The significant differences within normalized closing slope steepness can be summarized as follows:
a) female speakers:
- stressed vs. unstressed in lax tokens: /a/,//, /I/, //
- stressed vs. unstressed in tense tokens: /e/, /i/, /o/,/u/
- tense vs. lax on unstressed tokens: /a,a/,/o,/,/u,/
- tense vs. lax on stressed tokens: /i, I/,/o,/,/u,/
b) male speakers:
- stressed vs. unstressed on lax tokens: //,//
- stressed vs. unstressed on tense tokens: /u/
- tense vs. lax on unstressed tokens: /u, /
- tense vs. lax on stressed tokens: /e, /, /u,
| Figure 30. The effect of stress on the normalized mean values of the steepness of the end of the closing phase for vowel groups. The unstressed tokens are marked with empty circles, the stressed tokens with full ones. The standard deviations are plotted as whiskers. |
|
The effect of tenseness is much weaker and significant only for stressed // and // and unstressed // vowel groups and differs from vowel to vowel. This suggests that the effect of tenseness on the EGG amplitude parametrization is weak and that this relation does not permit to draw unequivocal conclusions.
In our experiment the steepness of the abduction phase (opening) is also used as a correlate of laryngeal behavior during the articulation of vowels in stress and tenseness conditions. The steepness is computed for the contact fall phase which lies between 90% of the peak-to-peak amplitude and the opening instant (this phase is called start of opening and its steepeness will be referred to as SOA from now on).
The opening slope shows tendencies similar to those of the closing slope. However, the SOA depends very strongly on the gender factor (F(1,5224)=1232.9, p<0.0005), which is not the case for the contact rise phase. The slopes for female speakers are approximately twice as steep as for male subjects (the mean value[9] for females is -252 compared to -132 for males).
The repeated measures ANOVA shows a significant dependency on stress, tenseness and vowel group factors (F(1,1547)=397.8, F(1,1547)=46.5, F(4,1547)=421.6, respectively, p < 0.0005) as far as the standardized values of the SOA for the female subjects are concerned. Nevertheless the effect depends also on repetition.
A strong effect of stress is also observed for males (F(1,797)=61.7, p<0.0005). The effect of tenseness is much weaker (F(1,797)=6.2, p=0.013)
As shown in Fig.31, and as confirmed by statistical analysis, the effects are significant in all vowels except /a/.
The significant differences within normalized opening slope steepness can be summarized as follows:
a) female speakers:
- stressed vs. unstressed in lax tokens: //, /I/, //
- stressed vs. unstressed in tense tokens: /e/, /i/, /o/,/u/
- tense vs. lax in unstressed tokens: /a,a/,/o,/, /u, /
- tense vs. lax in stressed tokens: /a,a/,/e,/, /i,I/,/o,/, /u, /
b) male speakers:
- stressed vs. unstressed in lax tokens: /I/,//
- stressed vs. unstressed in tense tokens: /a/,/e/, /i/,/o/,/u/
- tense vs. lax in unstressed tokens: /e,/
- tense vs. lax in stressed tokens: /a,a/,/e,/
| Figure 31. The effect of stress on the normalized mean steepness of the falling slope for vowel groups. The unstressed tokens are marked with empty circles, the stressed tokens with full ones. The standard deviations are plotted as whiskers. |
 |
As follows from Table 10 and Fig. 31, the effect of stress on the SOA is even stronger than it is on the steepenss of the rising slope. The effect of tenseness is not so clear-cut. For certain combinations of stress and vowels, the falling flanks of the EGG pulse are steeper, for others more gradual. It can be stated that SOA does not contribute reliably to the tenseness effect.
Comparing these results with the acoustic measurements of Claßen et al. (1996) it can be confirmed once again that they are in accordance with the glottal pulse skewness measure.
The analysis of the slopes of the remaining phases (i.e. the start of closing and the end of opening as depicted in Fig.22) does not contribute much to the general picture of the changes in the contact area. As was found in the duration analysis, both variables rely on the gender factor[10] (F(1,178)=5.12 and F(1,178)=41.38), but not on stress or tenseness. This supports our starting remark, that these short phases do not contribute seriously to the description of the EGG signal, and that the substantial information about laryngeal behavior is contained rather in the contact parts of the waveform than in the open phases of the folds' vibrations.
9. the slope values are relative and given as binary sample values. The values range from -32767 to 32767. Note that the values of the opening slope are negative, thus lower values describe steeper flanks.
10. these remarks are based on the comparison of mean values. The averages are computed for each vowel and speaker and then compared instead of the analyzing each pitch period.
