25.11.2014 Views

perilus v - Stockholms universitet

perilus v - Stockholms universitet

perilus v - Stockholms universitet

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Inc identally, let us note that, if it ex ists, acoust ic<br />

invar iance is a strange not ion since talkers can only<br />

mon itor it through their senses and listeners can only<br />

access it through the ir hear ing system. Why should sensory<br />

and aud itory transduct ion be assumed to have a transfer<br />

funct ion of one impos ing no transformat ion? Is it the case<br />

that what people really mean when they talk about acoustic<br />

invar iance is in fact Uauditory· invar iance? Let us look at<br />

some psycho-acoust ic results.<br />

IS PHONETIC INVARIANCE AUDITORY?<br />

We ment ioned earl ier a perceptual result that offers a<br />

rather cur ious parallel to Schulman's find ings. It is the<br />

uTraunmCller effect· wh ich is a demonstrat ion of the<br />

transforms requ ired to preserve the perceptual constancy of<br />

vowel quality under changes in (i) vocal effort and (i i)<br />

vocal tract size. It is also somewhat rem inscent of the<br />

find ings on FO-F1 interrelat ionsh ips in soprano vowels<br />

(Sundberg 197).<br />

Effort and vocal tract var iat ions can be dramat ically<br />

illustrated by synthetically modify ing a naturally spoken<br />

IiI. When all formants and FO are sh ifted equally along a<br />

Bark scale an IiI-l ike vowel is perce ived but the voice<br />

changes from an adult's to a ch ild's. When both F1 and FO<br />

are var ied in such a way that F1-FO is kept constant on a<br />

Bark scale - and the upper formant complex is left unchanged<br />

- an IiI-l ike vowel is perceived. Th is is remarkable in view<br />

of the fact that F1 reaches a value more typical of a lowpitched<br />

I /. One's impress ion is that the speaker rema ins<br />

the same but that she ·shouts·.<br />

Note the parallel between Schulman's and TraunmCller's<br />

results. Are the find ings causally related? Do we expla in<br />

the lack of formant compensat ion in loud speech in terms of<br />

the TraunmCller effect? Or do we account for the vowel<br />

qual ity results in terms of the ·Schulmanu effect?<br />

Of importance for the present discuss ion is the fact<br />

that behav ioral constanc ies have been demonstrated and that<br />

they imply that at least in this case phonet iC invar iance<br />

must be def ined at a level of auditory representat ion.<br />

Let us return for a moment to the alleged invar iance of<br />

the release spectra of stop consonants. Diana Krull<br />

collected perceptual responses from Swed ish listeners to<br />

burst fragments obtained from V1C:VZ words (Krull 1987). One<br />

hundred test words were generated by constructing all<br />

poss ible combinat ions of V1 or Vz = short Ii e a 0 ul with<br />

C: · Ib: d: rd: g:/. Confus ion matr ices for the burst<br />

st imul i demonstrate the drastic coart iculat ion effects. By<br />

and large, listener responses can be accounted for in terms<br />

of the acoust ic propert ies of the st imul i. Th is is shown in<br />

her attempts to predict the confus ions from aud itor ily based<br />

·perceptual distance- computat ions.<br />

A related study has been carr ied out by Lacerda (1986).<br />

W. can characterize one part of his research as var ia tions<br />

on the theme struck by Flanagan in his .arly -difference<br />

limen· exper iments on vowel formant frequenc ies (Flanagan<br />

19). Lac.rda's quest ion was: How well can listeners<br />

discr im inate four-formant st imuli that differ solely in<br />

terms of the frequency of F2. His work perm its us to compare<br />

7

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!