The values of the EGG parameters were statistically analysed by a multiple ANOVA procedure. The following factors were considered: speaker, speaker's sex, vowel type, stress and tenseness. The experiment has a full factorial design with measurement repetitions for all factors (Hays, 1988; Rietveld & van Hout, 1993). The statistical analysis was performed using MGHL module by SYSTAT. In some cases t-tests as well as Kolmogorov-Smirnov non-parametric tests were made (Berk, 1994). The significance level was set at 5% (a = 0.05).
For all analysis factors, the strongest effect was definitely caused by the speaker factor. Also, interaction of the speaker factor with other factors was significant. In the case of the parameters reflecting amplitude and amplitude derivatives, this strong effect may be caused by different amplifications of the EGG recordings[5]. As a consequence, it is necessary to normalize the results in order to compensate for inter-speaker variability. The effects of stress and tenseness were significant for certain EGG parameters. Based on these dependences, the following parameters were chosen for the second step of the analysis:
- Open Quotient I and Open Quotient II (relative duration of open phase)
- Speed Quotient
- steepness of closing and opening phases (rising and falling contact)
- relative durations of closing, closed and opening phases
For these parameters results are collected for all EGG waveform periods (starting from the third period of the vowel and ending in its penultimate period) of all vowels for all speakers, This yields ca. 5200 samples for each parameter. The results of the analysis for sex, stress and tenseness factors are summarized in Table 10. The significance of the differences in group means was tested using ANOVA's post-hoc Tukey's HSD test (Berk, 1994).
To reduce the effect of the individual speaker, the data was normalized, as was done for the acoustic data (Claßen et al., 1996). The data was standardized for a given speaker, i.e. instead of the original data x the z-scores were used:
(19)
where x^ is the sample's mean and S is the sample's standard deviation (Hays, 1988:180). The standardization of results converts values in such a way that they have a mean of 0 and a standard deviation to be 1. However, the shape of the probability distribution of the data remains unchanged, allowing a correct comparison of effects regardless of the speaker.
Such prepared data are made subject to the ANOVA analysis which investigates the distribution of the variance regarding gender, vowel, stress and tenseness factors , as well as the dependency on measurement repetition. Table 11 displays the percentage of variance which reflects the effects of the gender, stress and tenseness factors.
Additionally, Appendix I contains the group means of the standardized parameters that are subdivided into vowels and tenseness and stress groups for females and males. The summarized results for both genders are also given in Appendix I. The significant differences (of the t-tests) in group means are marked (bold) and are determined using the significance level of a<0.001. This validates the conclusions drawn using t-tests, while also using Bonfferoni probabilities for multiple comparisons (Berk, 1994:223-257;Hays, 1988:410-414). The non-normalized group means of the parameters are also given for reference.
Table 10: The effects of stress and tenseness on the EGG parameters for male and female groups. The group means
and standard deviations (inparenthesis) are given. Thesignificant differences between vowel
groups are listed in the significance column.
females males
1.lax 2. 3.tens 4.tens signif 1.lax 2. 3.tens 4.tens signif
unstre lax e e icance unstre lax e e icance
ssed stress unstre stress ssed stress unstre stress
ed ssed ed ed ssed ed
fundam 219.9 131.2
ental (22) (17)
freque
ncy
[Hz]
209.7 239.1 213.6 239.0 1-2,4 127.9 135.4 124.7 136.9 3-4
(11) (21.6) (9.7) (22) 2-3 (20.1) (15.9) (15.9) (14.7)
3-4
Open 56.0 53.3(5
Quotie (5.9) .9)
nt I
[%]
54.6 57.6 54.3 55.2 1-2 53.7 54.4 53.2 52.7 1-4
(6.4) (5.6) (4.7) (6.1) 2-3,4 (4.8) (6.2) (4.6) (5.4) 2-3,4
3-4
Open 37.3 32.4
Quotie (8.7) (7.1)
nt II
[%]
34.9 38.8 36.2 38.0 1-2,4 33.3 33.5 32.7 33.8
(7.9) (9.8) (7.8) (8.5) 2-3 (9.2) (9.2) (9.1) (8.3)
3-4
Speed 20.1 13.5
Quotie (6.1) (6.7)
nt [%]
18.7 20.0 20.9 20.5 1-2,3, 11.1 13.0 12.6 14.7 1-2,4
(5.5) (6.5) (5.8) (6.0) 4 2-3 (3.9) (9.0) (4.3) (6.5) 2-4
3-4
closin 673.7 788.3 688.3 865.3 1-2,4 785 846 741 781
g (225) (279) (237.9 (431)) 2-3,4 (604) (637) (515) (664)
slope ) 3-4
[%]
openin -252.2 -131.8
g (141) (84.1)
slope )
[%]
-206.9 -252 -235.3 -276.6 1-2,3, -112 -140 -117 -137 1-2,4
(87) (115) (114) (173) 4 2-4 (74) (96) (68) (83) 2-3
3-4 3-4
full 15.1 18.7
closur (2.9) (2.8)
e
durati
on [%]
14.9 14.5 15.3 15.5 1-2,4 18.3 18.1 18.2 19.2 1-4
(1.4) (2.8) (1.6) (3.7) 2-3,4 (2.7) (3.1) (2.3) (2.6) 2-4
3-4
openin 22.8 23.7
g (5.4) (4.8)
durati
on [%]
24.0 21.8 23.4 22.8 1-2,4 24.6 23.2 24.5 23.4 1-2,4
(5.9) (5.1) (5.1) (5.4) 2-3,4 (4.0) (4.9) (3.7) (5.2) 2-3 3-4
closin 7.3 4.8
g (1.4) (2.1)
durati
on [%]
7.3 7.1 7.8 7.3 1-3 3.9 4.4 4.6 5.3 1-2,3,
(1.5) (1.2) (1.4) (1.4) 2-3,4 (1.1) (1.7) (1.7) (2.4) 4 2-4
3-4 3-4
Table 11:
Overview of the percentage of explained variance in the standardized EGG parametrization measured for the
entire vowel durations (N=5244). The effects of gender, vowel type, stress and tenseness are
investigated for Open Quotient I(OQI), Open Quotient II(OQII), Speed Quotient (SQ),closing (ECA)
and opening (SOA) slopes and duration closed phase duration (CT), opening (SOT) and closing duration (ECT). Significant
effects are bold.
Parameter OQI OQII SQ ECA SOA CT SOT ECT
total explained 9 11 16.7 41 53.3 23.7 10.9 14.9
main effects
gender 0.028 0.001 0.142 0.454 0.001 0.098 0.04 0.346
vowel 0.194 1.984 3.535 5.944 19.44 2.656 0.482 2.514
tenseness 1.26 0.028 0.686 0.27 0.041 1.54 0.009 1.719
stress 0.186 0.675 1.12 3.014 7.894 0.064 1.649 0.261
interactions
gender *vowel 0.12 1.279 0.908 8.274 1.811 3.047 0.048 0.941
gender*tenseness 0.002 0.03 0 1.488 0.558 0.023 0.475 0.015
gender*stress 0.414 0.243 0.099 0.946 0.001 0.63 0.084 0.908
vowel*tenseness 1.008 0.512 0.576 0.478 1.423 1.675 0.437
vowel*stress 0.725 0.607 2.06 1.665 5.181 1.001 1.421
tenseness*stress 0.473 0.245 0.081 0.002 0 0.821 0.062 0.005
gender*vowel*tens 0.554 0.516 0.705 1.643 0.854 0.847 0.226 0.892
e
gender*vowel*stre 0.904 0.772 0.295 0.866 0.187 0.159 0.319 0.192
ss
gender*tense*stre 0.072 0.158 0.123 0.209 0 0.028 0.039 0.108
ss
vowel*tense*stres 0.506 0.199 0.469 0.388 0.928 0.18 0.474 0.993
s
gender*vowel*tens 0.125 0.107 0.036 0.714 0.298 0.926 0.244 0.249
e* stress
In the following sections these results will be discussed for each variable and the specific impact of the individual factors will be analysed as well.
5. but this does not clarify why the timing parameters also strongly depend on the speaker. This results addresses the use of the EGG and, in the broader sense, voice quality factors in forensic phonetics (see Jessen et al., 1996), but is outside of the scope of this work.
