Discussion

The results show a wide range of variation in the accuracy of correct identification of child’s gender by voice. What accounts for that variation? In other words, what are adults using to cue gender? We investigated whether the following aspects of the speech signal might be cued on by listeners:

• Pitch

• Breathy voice

• Nasality

• Time between words

• Pitch contour

Most of these were discarded quickly after auditory or visual inspection, and after measurements of the most extreme cases. For example, it is clear that pitch does not play a role, since the child with the highest pitch is a boy, and the child with the lowest pitch is a girl. Moreover, both of these children were among the most correctly identified (numbers 10 and 11).

We then turned to the vowel quality of /ai/ in “I” and /iy/ in “E.” We measured the centers f the first two formants for each vowel for each child, and plotted them as below. (Labels for each point refer to the child gender, how they were perceived, and their number, so GB-7 means that the child was a girl, usually mistaken for a boy, and is number 11. “R” for the second letter indicates the child was perceived as neither boy nor girl, but randomly one or the other.) Some children sang their alphabet and are labeled as such.

Click on each data point to hear the vowel that was measured.

For the “I” vowel, there are some clear clusters, with a few anomolies. The dashed circles mark children perceived to be girls, while the solid are either perceived to be boys or are random. The general trend is for boy-perceived speakers to have an /ay/ nucleus further back and higher than the girl-perceived speakers.

For “E,” we find height to be significant, with the boys tending to have a higher articulation than the girls. The solid line indicates that all speakers above that line were perceived as boys, except for speaker 7, who was a singer. Note that speaker 1, a girl who was perceived randomly, is in the girl range for /ay/ but the boy range for /iy/. This clash may account for the her being perceived neither as boy nor girl consistently.

The vowel articulation thus seems to provide the best predictions about whether a child will be perceived as girl or boy, even though it is still imperfect. Of course, it could be that the nucleus vowel height is a measure that indicates a different cue to which listeners are attending. For example, it could be that listeners perceive more change over a vowel (i.e. distance between vowel nucleus and glide) as a girl cue, and little change as a boy cue. Such a view would predict that perceived girls’ nuclei will be lower than boys, and this prediction is bone out in our findings.