Our first analysis concerns the OQ computed according to the first method (OQI) described in section 12. The OQI is defined as the relative duration between the opening and closing instant. The moment of opening is determined as a minimum of the first derivative of the EGG waveform or, if the minimum is weak, as the instant of intersection of the waveform with the line connecting two neighboring closing instants (Fig.22). The mean of the Open Quotient (computed for the whole vowel length) depends primarily on the speaker's gender. The mean value for females is about 5% higher (10% in relative terms) than for males (repeated measures ANOVA with gender factor only: F(1,5242)=178.527, p < 0.0005). This is in line with the findings of Holmberg et al. (1988), Sluijter (1995:App.B1) and Claßen et al. (1996). However, based on the measurements of the acoustic signal Slujiter estimates the difference in the OQ to be about 25%, much more than found here (ca. 5% of relative difference). The differences in the mean values of acoustic measurements found by Claßen et al. (1996) are much smaller (2.05 dB for H1* - H2* difference[7] for males compared to 5.51 dB for females). The 3.5 dB difference is equivalent to a ca. 12% change in the OQ (linearly interpolating the results of Klatt & Klatt, 1990), which makes the difference between the EGG and the acoustic measurments more realistic. Holmberg et al. (1988), by means of the inversely filtered acoustic signal, investigate the difference in the OQ between males and females with respect to loudness. In their examination the difference between male and female speakers was about 13% for normal loudness. It was already mentioned (section 13) that the OQ found in the EGG domain is almost always smaller than the one found in the acoustic signal. For example, the values given by Higgins and Saxman (1991) for young males and females (which is also the subject of this analysis) are about 6% smaller for the EGG duty cycle, which corresponds well to the our results.
The effect of stress (and tenseness) on the OQ was investigated on standardized data. The results, however significant, are not deterministic for the pooled data; the direction of the changes depends on the vowel type.
Among females the interaction effects between vowel group, tenseness and stress factors are significant. The results depend also on the repetition of the spoken word. As far as the t-test results in App. I are concerned, no common tendency can be reported. If the vowel groups /a/, /e/, /i/ are stressed, the OQI grows, whereas, when /o/, /u/ are stressed, it decreases.
Unfortunately, the results of Claßen et al. (1996) are unavailable for all these tokens (due to the limitations of the acoustic technique; see section 15.1). However, the differences for /a/ were also not found to be significant.
Generally speaking, the Open Quotient cannot be regarded as a good predictor of stress for female speakers. The effects are significant but depend primarily on the vowel type.
Effects of tenseness were significant for female speakers, but again there is interaction with other factors and no common tendency can be observed.
For male subjects the effects on the OQI are easier to recognize. The Open Quotient depends only on tenseness (F(1,797)=19.255, p < 0.0005). However, this effect interacts with the vowel factor (F(4,797)=9.4, p < 0.0005). Thus, once again, no common tendency can be found.
The dependence of the standardized OQI on stress is depicted in Fig. 25 (for all speakers). And for the third time, there is no recognizable common tendency - in stressed tokens the OQ lowers for back closed vowels (o,u) and grows for all other. There is also no definitive dependency on tenseness - generally, OQ was found to be lower for tense vowels within stressed tokens, whereas in unstressed tokens it varies from vowel to vowel. In summary, it can be said that OQI does not show a reliable dependency on the stress and tenseness factors.
In the acoustic measure of the OQ, Claßen et al. (1996) found the difference between stressed and unstressed vowels to be insignificant. Their results also vary depending on the vowel type. No effect of tenseness was detected, as was the case in the EGG domain.
The results of Sluijter (1995:119) for stress within the [+F] class (for /e/,/i/,/a:/) show a difference of ca. 8% which is also in accordance with the results given here. However, due to the limitations of the acoustic method other vowels she does not analyse other vowels.
Contrary to Kingston et al. (1997) tense and lax vowels do not exhibit a reliable difference in the Open Quotient, neither acoustically nor in the EGG domain. In several languages, tense vowels are more breathy than lax ones, and therefore we would expect to find a higher OQ for tense vowels.
| Figure 25. The effect of stress on the standardized mean values of the Open Quotient measured by method I (OQI) 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 mentioned previously, the Open Quotient measure in the EGG waveform is, due to problems with the opening instant estimation, prone to errors. In order to compensate for this, another correlate of the Open Quotient has been proposed in section 12. The duration of the non-contact phase (OQII) was investigated along the same variables as the previous measure of a glottal duty cycle. Generally, the results are quite similar to the previous; however, some differences should be noted.
The dependency of gender is still very strong (repeated measures ANOVA with sex factor only, F(1,5242)=293.9, p < 0.0005) and the mean for females is also greater than for males (37% against 33%). The effects on standardized values also depend on gender.
The effect of stress is still significant for females (repeated measures ANOVA, stress, tenseness and vowel factors, F(1,1547)=50.722, p<0.0005), albeit the results depend on the measurments repetition and interact with a vowel group and tenseness. For lax tokens (without /a/) OQII grows under increased stress, however, on tense tokens the dependency of /o/ and /u/ does is not significant. For female speakers OQII is independent of tenseness.
For male subjects on the other hand there are weak dependences that are significant only for a limited number of tokens (see Appendix I).
The effects of stress on both genders are summarized in Fig. 26. The effect is the strongest for the vowel /i/, which is in line with acoustic measurements by Claßen et al. (1997). The effect of tenseness is statistically weaker and an interaction between stress and tenseness factors occurs.
| Figure 26. The effect of stress on the standardized values of the Open Quotient II (OQII) for vowel groups. The unstressed tokens are marked with empty circles, the stressed tokens with full ones. The standard deviations are plotted as whiskers. |
 |
On the whole it can be said that the duty ratio of the EGG (the duration of vowel fold separation) depends primarily on the speaker's gender. There is a slight dependency on stress concerning the analyzed vowels and their linguistic context. However, this effect interacts with vowel and tenseness factors. Comparing these results with other experiments, a far-reaching agreement with the results of Claßen et al. (1996) and Sluijter (1995) for OQI can be stated. It has to be noted, though that only a limited set of vowels was used in their examinations. Also, the influence of tenseness does not exhibit a clear pattern, as this effect interacts with other factors. Vowel type does not influence the adduction factor, either.
