Summary & conclusions

summary conclusions

In summary, the distinction between groups of speakers is most appropriately expressed by the specific parameters of the EGG signal developed in this study, particularly by the steepness of the contact rise. The duration of the open phase (OT), as well as the durations of the closing and the opening phases, are also good indicators of the speaker category.

The EGG signal typical of chordectomy patients has the longest and most gradual transitions between open and contact phases. The steepness of the closing phase is on average less a quarter of that permit for normal subjects. This group also exhibits the shortest opening phase (about half as long as for the control group) and the averaged peak-to-peak amplitude is the smallest in all analyzed groups.
For the patients suffering from the recurrent nerve paralysis without compensation (rp-without-c) the increase in the EGG amplitude during vocal fold closure is also more gradual than for the non-pathological groups and comparable in magnitude to that of the chordectomy patients. The closure occurs slowly, as is the case in the chordectomy group. The Open Quotient (measured as the duration of the no-contact phase) is, however, quite normal. The mean value of the peak-to-peak amplitude is lower than for healthy speakers, but on the other hand higher than for the chordectomy group.
In almost all the measures used in this study, for subjects of rp-with-comp tend to yield results similar to those of the control group, thus exhibiting better results than the patients without compensation of the nerve paralysis. In the included figures the effect of the compensation can be observed in the form of a distinct tendency towards decreasing difference between the subjects with recurrent nerve paralysis and the control group. Thus, it can be said that these patients have a better voice quality than the group without the compensation effect. The significant differences compared to the control group manifest themselves in the Open Quotient and the amplitude measures. The mean value of the peak-to-peak amplitude is lower than for healthy speakers, especially the fluctuation of the amplitude is significantly higher than for the control group. The duration of the non-contact phase is also longer than for modal voices, i.e. the voices are slightly breathier than the modal voices.
As expected, the breathy group of speakers does not exhibit strong differences compared to the control group. The strongest difference was observed for the duty ratio measure which was significantly higher. This is in accordance with our expectations and corresponds to our description of voice quality given in section 11.

22. Conclusions

The description of the electroglottographic signal proposed in this study the was used to distinguish different voice qualities including pathological voice qualities. The data collected from the sustained voicing of vowels produced by five groups of speakers was characterized by phonetically and linguistically established parameters of the EGG waveform. This quantitative description of the data was compared with a qualitative grouping made a priori by a phoniatrician. The investigated groups include two groups of healthy speakers with modal and breathy voice qualities and three groups of speakers with a record of larynal pathologies: one group of chordectomy patients and two groups of patients with recurrent nerve paralysis (with and without compensation effect).

The results confirm the proposed method of parametrization as a base for the assessment of voice quality. The ANOVA analysis demonstrates the significant differences between groups of speakers in the domain of EGG parametrization. The most important parameters are the steepness of the rising contact phase and the relative durations of the open, opening and closing phases. Also, the peak-to-peak amplitude and its variations contribute to the discrimination of speaker groups. The automated rough sets classification based on the quantitative data is in full accordance with the data grouping provided by the physicians. The rules generated by the system (with roughness=0.3 and rule precision threshold=0.5) utilize mostly those parameters for which the group differences are statistically significant.

The following questions among others were asked at the beginning of this study:

How can voice quality be measured objectively and quantitatively ?

How do the pathological voices differ (and how are the differences expressed in the EGG domain)?

The results of the conducted experiment can, perhaps only partially, answer these questions.

The experiment offers experimental evidence to support the thesis that voice qualities (even pathological) can be distinguished using electroglottographic waveforms and their proposed description. The shape and timing of the single periods of the glottal waveform differ not only between different phonation types produced by the same speaker, but also between speakers and between groups of speakers. In this experiment the differences between groups of speakers were not only properly identified, but also the relative importance of the parameters was established. The results in some sense resemble those of the stress and intonation experiment. The steepness of the rising contact and the relative duration of the no-contact phase seem to be the most important cues to the phonatory process. They depend both on the linguistic context of a produced vowel and the speaker-dependent paralinguistic context (the physiological foundation of voice). This result is also supported by the preceding experiment, where the strongest dependency of the results was established for the speaker factor.