schwa-raising - formants

[Mark J. Jones]

----------------------------Original message----------------------------
Formant measurements were made of single FFT and LPC spectra at the vowel
midpoint (128 sample window at 11025 sample rate - lowpass filtered at 5000
Hz). The words used were 'Saddam' with coronal consonants either side of
the unstressed schwa, 'Copernicus' with non-coronal consonants flanking the
unstressed schwa, and 'Sidney, with coronals flanking a stressed /I/ for
comparison. Three tokens each recorded in a carrier sentence "Say X again"
by me - 31 yr old male speakers of Southern British English. The vowels
were all of extremely short duration and in the case of 'Copernicus' showed
extensive F2 transitions (the vowel in 'communicate' was investigated but
was nasalised and so no reliable measures could be made of F2 and F3). No
F2 transitions were seen in 'Saddam' or 'Sidney'.

The unstressed vowel of 'Saddam' had an average duration of 49.3 ms (range
47-53 ms) and of 'Copernicus' had an average duration of 31.3 ms (range
28-36 ms). The stressed vowel of 'Sidney' had an average duration of 56.7
ms (range 49-69 ms). Note the similarity of this to the unstressed schwa
durations in 'Saddam' even though it is stressed.

F1 values for 'Saddam' ranged between 330 Hz and 400 Hz (average 373.3 Hz)
whereas F1 for 'Copernicus' showed values between 435 Hz and 460 Hz
(average 446 Hz) - F1 was therefore lower for the 'Saddam' vowel. For
'Sidney', F1 was 384 Hz on average (range 360-408 Hz).

F2 values were average of 1601 Hz for 'Saddam' (1541-1641 Hz range) and
1621.3 Hz for 'Copernicus' (range 1600 - 1643 Hz). F2 was therefore also
lower on average, if very slightly, though the range for 'Saddam' was
greater, and F2 was moving in the case of 'Copernicus'. The fall in F2
throughout the vowel in 'Copernicus' was 394.7 Hz on average (range 172 -
603, Hz values between 1809 and 1206). This obviously makes the idea of an
F2 'value' for the 'Copernicus' vowel more or less problematic depending on
your view of how vowel quality is perceived. For 'Sidney', F2 moved little
and was 1865.7 Hz on average (range 1813-1910 Hz).

F3 values were more or less static throughout. The average for 'Saddam' was
2577.7 Hz (range 2541 - 2571), and for 'Copernicus' 2400.3 Hz (range 2400 -
2401). F3 was therefore higher for the 'Saddam' vowel. In 'Sidney', F3 was
also static and at 2546 Hz on average (range 2498-2594 Hz).

These results are only from 3 repetitions of 1 speaker, but they do
indicate that there are quality variations in the static F1 and F3 targets
of schwa vowels. Quality differences are also likely on the basis of F2,
though it is unclear how a general F2 target should be identified.

A higher jaw position in 'Saddam' and relatively little lingual movement
away from a high front position for the coronals may cause the lowering of
F1 and possibly also raise F3 according to perturbation theory.

In a standard vowel plot, the 'Saddam' vowel will be higher than the
'Copernicus' vowel, but the problem of identifying a good F2 value for
'Copernicus' schwa and the lack of representation of F3 means that it is
not possible to locate this latter vowel accurately. Relative to stressed
/I/ of 'Sidney', the 'Saddam' vowel shows a greater similarity in terms of
F1 and F3 values, but F2 is much higher for /I/, suggesting that the
'Saddam' vowel is more located more centrally than /I/, and higher then
'Copernicus' schwa within my vowel space. It is not the case that all
schwas are subject to raising (at least in my British English speech), and
the effect appears to be a phonetic one due to coronal consonants.

Mark

Mark J. Jones
Department of Linguistics
University of Cambridge