The mean values of the peak-to-peak amplitude show relevant relations to the speaker group (F(4,55)=3.4, p=0.01). The signals recorded for healthy subjects (normal and breathy groups) are significantly stronger (and with smaller standard deviation) than those of the other groups. The standard deviation was measured for the whole sustained vowel. The EGG signals of chordectomy and rp-without-c speakers are weaker and with greater standard deviation. The rp-with-comp patients also exhibit significantly greater variation than the control group. The reason for this is presumably the varying contact between the vocal folds: in one vibration cycle they close normally, but in the next cycle no or only weak contact is achieved. In the EGG signal this is registered as a fluctuation of the peak-to-peak amplitude from one period to the other (if the contact becomes weaker, then the peak amplitude of the EGG signal becomeslower; see section 11). Despite greater amplitude variation compared to the control group, the vowels spoken by the breathy group are consistently breathy, which causes the variation in the amplitude to be smaller than in the other groups. The results suggest that the "shimmer" of the EGG may be a valuable parameter in distinguishing groups of pathological speakers, which is in line with the results of Haji et al. (1986).
The analysis of the steepness of the rising and falling contact phases of the EGG periods was successful in establishing a significant relation, too. The strongest effect (F(4,55)=16.529, p<0.0005) was found with respect to the steepness of the rising contact phase (see Fig.37). The decrease in laryngeal impedance (increase in the EGG signal) is steepest for healthy voices (normal and breathy). The values for rp-with-comp differ only slightly, but there is a clear distinction with respect to the other two groups, their mean values being more than three times smaller. The impact of grouping is very strong, as about 74% of the data variance are explained by this effect (r=0.739)5.
The steepness of the falling contact phase (the start of opening) is not distinctive. The ANOVA measure does not exhibit any significant dependence, which is different from the results in the experiment with modal speakers. Moreover, we should recall the problems in determining the opening instant, which in turn caused problems in the proper estimation of the duration of the opening phase. This can also bias the results concering the steepness of the opening phase.
The steepness of the end of opening phase does not vary depending on the speaker group.
The parametrization of the discrepancies between original and modelled waveforms does not lead to new conclusions. The differences are significant for the contact and no-contact phases (F(4,55)=4.012, p=0.01 and F(4,55)=3.057, p=0.024, respectively). However, they resemble the already mentioned dependencies, especially those characterized by timing variables. This does not allow us to draw any conclusions about a different global shape of the waveform as a signifcant characteristic of a speaker group.
