272 Erik R. Thomas, Norman J. Lass, and Jeannine Carpenter
1981, 1982; Jun and Foreman 1996), though this feature of African American
English is associated with speech acts that are not represented in the stimuli.
In addition, some studies (Loman 1975; Jun and Foreman 1996; Wolfram and
Thomas 2002) have reported that African Americans show constant alterna-
tions of pitch, or more intonational pitch accents, that European Americans
lack, and this feature might affect the standard deviation of F0. The degree
of breathiness, which the results of Thomas and Reaser (2004b) showed to
be an important cue, was measured using the Cepstral Peak Prominence
(CPP) method described in Hillenbrand, Cleveland, and Erickson (1994) and
Shrivastav and Sapienza (2003). Vowel quality was gauged in several steps.
First, measurements of the ¿ rst four formants were taken in the center of the
vowel for /æ/, 35 ms from the onset of the vowel for the nucleus of /o/, and 35
ms from the offset of the vowel for the glide of /o/. Then, formant values were
converted from Hz to Bark using the formula given in Traunmüller (1990).
Finally, the frontness of both the nucleus and glide of /o/ was indicated by
the value Z 4 -Z 2 (i.e., Bark-converted F4 minus Bark-converted F2), while
the height of /æ/ was indicated by the value Z 2 -Z 1. These Bark-difference
metrics serve to normalize vowel quality across speakers and are inspired
by Syrdal and Gopal (1986), though the Z 4 -Z 2 and Z 2 -Z 1 values yield more
consistent results than the metrics that Syrdal and Gopal proposed. Analysis
of jitter (local variation in F0) and shimmer (local variation in amplitude),
which Thomas and Reaser (2004b) found not to be correlated with ethnic
identi¿ cation, was omitted.
Multiple regression was applied to the stimuli, as shown in Table 12.3.
Table 12.3 is broken into three parts according to listener group, a, b, and
c. A factor was excluded from the analysis when it was neutralized by an
experimental treatment; i.e., measures of F0 were excluded from analyses
of monotonized stimuli and measures of vowel quality were excluded from
analyses of schwa-converted stimuli. Necessarily, measures of vowel quality
were also excluded from analyses of the control stimuli. Cells in the table are
shaded if the factor represented in that cell reached statistical signi¿ cance.
Several trends are evident in Table 12.3. One is that fewer cells reached
statistical signi¿ cance for the West Virginia listeners than for either of the
North Carolina groups. This result is undoubtedly related to the lower degree
of identi¿ cation accuracy that the West Virginians showed. Another is that
vowel quality factors were the ones that most often reached signi¿ cance. This
result may not be surprising, considering that the stimuli were designed to
highlight vowel quality. F0 measures showed some erratic patterns, espe-
cially for schwa-converted stimuli, but in most cases a higher mean F0 was
associated with African Americans and a higher maximum F0 with European