Phonetic Detail in the Perception of Ethnic Varieties of US English 315and the relational measures of F3-F2 and F3-F1 in bark during the vowel steady
state.^18 The acoustic measures which are source oriented and which represent
measures for either the voiced portion of the word or the entire word included
(a) overall glottal airÀ ow characteristics (glottal period, open quotient, average
closure of glottis and period size, peak of glottal closure, maximum airÀ ow
declination rate), and (b) pitch characteristics (mean, minimum, maximum,
standard deviation, peak, initial, ¿ nal and difference in pitch). The last set of
acoustic measures—reÀ ecting a combination of both source and ¿ lter char-
acteristics—includes intensity characteristics such as the mean, minimum,
maximum, standard deviation, peak, initial, ¿ nal, and difference in root mean
square. Because pitch, intensity, and spectra capture overlapping information
of the source and ¿ lter (e.g., vowels have certain inherent pitch and intensity),
this data set has a potential to be driven by a latent factor or factors and is suit-
able for discriminant analysis. Results from a perceptual experiment using the
“hello” tokens were reported in Purnell et al. 1999. The model so far for this
data (Tables 13.4 and 13.5) has been based on the acoustic measures associ-
ated with the dialect that the speaker intended on producing. Instead of using
the dialect classi¿ cation of the tokens based on what the speaker intended to
produce, the next model relates the acoustic measures to the dialect the listen-
ers perceived the tokens as. The perceived dialect was assigned by a combined
frequency over 66%, e.g., AAE tokens were recategorized as SAE if subjects
responded that the AAE token was an SAE token 66% of the time.
4.3 Results
Results of a forward stepwise and canonical discriminant analyses are shown in
Tables 13.8 through 13.11. The forward stepwise analysis in Table 13.8 reveals
the importance of vowel space measures, primarily the expected backness dif-
ference of /͑/ and /o/ across the dialects (not between vowels). An ASCC nearing
1 for the acoustic data (speci¿ cally, 0.82) indicates a strong acoustic distinction
among the dialects by the following predictor measures: vowel backness (/ܭ/
F3-F2 Z, /o/ F2 Z), vowel intensity (RMS standard deviation, initial value),
glottal airÀ ow (maximum airÀ ow declination rate), and vowel height (/ܭ/ F1 Z).
A less strong overall ASCC value of 0.620 was found for the perception data.
The measures of note also include vowel backness (/o/ F2 Hz, /ܭ/ F3-F2 Z),
vowel intensity (change in RMS), and vowel height (/ܭ/ F1 Z, /o/ F3-F1 Z). Of
note is the identi¿ cation of one linear backness measure in hertz for the percep-
tion data (/o/ F2 Hz) and another non-linear backness measure in bark for the
acoustic data (/o/ F2 Z). Unlike hertz, which is an acoustic measure, bark is a
transformation of hertz along a psycho-acoustic dimension (Traunmüller 1990).