Identi¿ cation of African American Speech 277Fo u r e x a m ple s of e a c h s e nt e n c e r e a d by A f r ic a n A m e r ic a n m a le s , by Eu r o p e a n
American males, by African American females, and by European American
females were selected. Unlike in Experiment A, the same speakers were not
used for each sentence. Instead, examples of each sentence were selected that
showed features of vowel quality and intonation typical of the ethnicity of the
speaker, as judged by Thomas. Then, for each sentence, utterances by African
American males and European American males were paired with each other,
as were utterances by African American females and European American
females. The same two speakers were not paired with each other more than
once; i.e., if speaker x was paired with speaker y for the ¿ rst sentence, he or
she was not paired with speaker y for any of the other three sentences.
Next, a Praat script was prepared in order to perform synthesis. The
script was designed so that the F0 contours and segmental durations could
be swapped for each of the paired utterances, syllable by syllable. The result
for each pair of speakers was four stimuli: the unmodi¿ ed sentence uttered
by the African American, the unmodi¿ ed sentence uttered by the European
American, the sentence uttered by the African American but with the prosody
of the European American substituted, and the sentence uttered by the Euro-
pean American but with the prosody of the African American substituted. On
a larger scale, the experiment involved a 2x2x2x2x4 design in that it included
male/female x African American/European American x presence/absence of
diagnostic vowels x unmodi¿ ed/prosody-swapped x 4 speakers.
As in Experiment A, the order of the stimuli was randomized and the
stimuli were presented in groups of ¿ ve, with a voice announcing “Set 1,”
“Set 2,” etc. at the beginning of each grouping. However, there was no need
to group the stimuli into sections for Experiment B. As before, several stimuli
at the beginning of the experimental recording served to acclimate listeners
to the task and responses to them were excluded from analysis. Once again,
African Americans at North Carolina State University, European Americans
at North Carolina State University, and European Americans at West Virginia
University served as subjects for the experiment and were asked to circle
either African American or European American for each stimulus. Results
from listeners who reported hearing impairments, whose ¿ rst language was
not English, or who were not African American or European American were
excluded from analysis.
Accuracy of identi¿ cation for each of the independent factors is plotted in
Figure 12.1. Stimuli with diagnostic vowels were identi¿ ed more accurately
than those without diagnostic vowels, unmodi¿ ed stimuli were identi¿ ed
more accurately than prosody-swapped stimuli, stimuli uttered by males were
identi¿ ed more accurately than those uttered by females, and stimuli uttered