It was the goal of this experiment to answer the following questions:
* Is it possible to obtain information about voice quality (with respect to the linguistic layer of communication) from electroglottographic signals?
* How is linguistic information (especially intonation and stress) coded in voice quality?
The experiments give a positive answer to the first question. The EGG signal conveys information characterizing voice quality, which in turn is used for coding linguistic information. The parametrization of the EGG waveform as proposed in section 12 uncovers a complex relation between the signal and various laryngeal characteristics.
The dependence between the prominence of a vowel bearing word stress and the shape of the vocal fold contact waveform was investigated in the first experiment. The results of this investigation were extensively compared to the outcomes of the acoustic signal examination conducted by Claßen et al. (1996) for the same data. On this basis, the dependency between skewness of the glottal pulse (realized in the speech signal as a lowering of the spectral tilt) and the steepness of the EGG closing and opening phases was determined. This leads to the hypothesis that word stress in German is realized with a change in voice quality in the sense of changed laryngeal activity. The experimental evidence argues for a higher subglottal pressure for stressed vowels than for unstressed ones. A further result is that an increase of subglottal pressure manifests itself in the EGG domain in the form of steeper flanks of the contact pulse.
The outcome of the sentence intonation experiment points to another main correlate in laryngeal activity. The change in the duty ratio of the EGG is the most prominent effect of the increase in fundamental frequency. A rise of the steepness of the EGG slope is also observed (the intonation pattern was analyzed for lexically stressed vowels), although it is not as evident as for the word stress. This leads to the hypothesis that when F0 increases, the contact of the vocal folds is affected by the longitudinal tension of vocal folds which is related to cricothyroid muscle action.
These hyptheses will be tested on the computer model of the electroglottographic waveform. The analysis of this model follows in section 27.3.
In the first experiment, the tense/lax distinction was examined as an additional factor. The only characteristics of tenseness was an elongation of the rising contact phase, but the statistical significance of this effect was relatively weak. It was postulated that the effect is caused not by a change of the laryngeal settings but rather by the impact of the striction in the vocal tract (caused by the higher position of the tongue body in tense non-open vowels).
This results suggest that linguistic prominence on the word level in Germanis is linked to higher skewness of the glottal pulse, which also means a higher excitation strength of the vocal tract. Prominence on the sentence level is realized by a change of the longitudinal tension of the vocal folds.
Yet another outcome of the experiment is the selection of the introduced parameters of the EGG waveform. This analysis allows the formulation of a hierarchy of the parameters' relative importance: slope steepness and Open Quotient (measured as the relative duration of the complete opening of the folds) are most informative. Parameters like peak-to-peak amplitude, Speed Quotient, duration of particular EGG phases and crest factor do not exhibit any significant dependencies. However, some waveform features may be important for the description of non-linguistic behavior.
Concluding, the relation of the acoustic signal to the information retrieved from the electroglottographic signal is complicated, but it can be shown that the parametrization of the EGG waveform allows a precise view into the phonation behavior of non-pathological, modal voices. The whole procedure is robust and can be performed automatically.
A strong dependency of almost all parameters on individual voices was observed very early in this investigation. The EGG waveform strongly mirrors the speaker's gender. This allows the introduction of the next experiment in which the relation between the EGG and the para- and extralinguistic layers of communication will be examined.