344 Paul Foulkes, Gerard Docherty, Ghada Khattab, and Malcah Yaeger-Dror
the fact that listeners might only show a clear pattern of responses for stimuli
with extremely high or low f0 values. When we examined the f0 and ampli-
tude measures in more detail we found a much stronger positive correlation
when only those stimuli with extremely low f0 values were considered. We
arbitrarily tested for f0 and amplitude correlations with the 10, 12, 15, and 20
stimuli which had lowest f0. The correlation coef¿ cient reached signi¿ cance
for the 12, 15, and 20 stimuli with lowest f0. No signi¿ cant correlations were
found in a similar set of comparisons with the highest f0 stimuli. Thus, in
summary, stimuli with very low f0 may be eliciting high numbers of “girl”
responses because they are also quiet. This remains, however, an issue which
demands further exploration in future experiments.
5.5 Results by voice quality
Voice quality was not generally identi¿ ed as a signi¿ cant factor in the regres-
sion analyses, the only exception being for American listeners with pre-pausal
stimuli (Table 14.4). Nevertheless we present the main effects here for complete-
ness, and because a predictable trend did emerge in the responses. Figure 14.5
represents the proportion of “girl” responses for the three groups, with tokens
divided according to whether they were coded as modal, creaky, or breathy.
The number of stimuli in the creaky and breathy categories is small (6 creaky
and 4 breathy from the total of 67), but it is noteworthy that the breathy tokens
elicited the highest proportion of “girl” responses from all groups.
Figure 14.5 Percentage of “girl” responses by voice quality.