The following article appeared in
Journal of the Acoustical Society of America 69: 1465–1475
and may be found at
http://scitation.aip.org/content/asa/journal/jasa/69/5/10.1121/1.385780.
Copyright (1981) Acoustical Society of America. This article may be downloaded for
personal use only. Any other use requires prior permission of the author and the Acoustical
Society of America.
Perceptual dimension of openness in vowels
Hartmut Traunmtlller
Institutionenfdr Lingvistik,StockholmsUniuersitet,
S-106 91 Stockholm,Sureden
{Received10 March 1980;acceptedfor publication10 November 1980}
The role of intrinsic factors determiningperceiveddegree of vowel opennesswas examined. In order to
determinethe role of F I and F0, one-formantvowels,coveringa wide range of fundamentaland formant
frequencies,were identifiedby 23 subjectswho were native speakersof a Bavarian dialect in which five
degreesof opennessoccurdistinctively.The significanceof intrinsic factorsother than F1 and F0 was also
studiedusingsyntheticversions
of naturalvowelswith F 1 and/or F0 systematically
displacedin frequency.It
was found that, generally,the tonality distancebetweenF 1 and F0 is decisivefor openness,
while the higher
formantscontributemarginally.It was further found that the distancebetweenwidely spacedformants,as
betweenF2 and FI in front vowels, is not crucial for vowel identification. The results are evaluated in terms
of a psychoacoustic
model of identificationby pattern matching. The model incorporatestwo basic
assumptions.
First, a certainpatternof excitationalongthe basilarmembraneis recognizedasa givenfeature
regardlessof positionalong the membrane.Second,there is an integrationband with a width of 3 Bark
effectivein spectrumenveloperecognition.
PACS numbers: 43.70.Dn, 43.70.Ve
INTRODUCTION
Most of the distinctive features of speech sounds ap-
parently constitute dimensions along which only a binary distinction is used. Along the dimension of
"openness" in vowels, however, more than two distinctions occur in many languages. In the present investigation, we attempt to examine how this dimension is
perceived. The investigation is limited to the role of
intrinsic factors,
i.e., to features that are acoustically
of the overlapping of formant frequency data of different
vowel phoneroes produced by men, women, and child,ren
(Peterson and Barney, 1952), it is evident that perceived openness is not solely related to Fl. It has been
shown that the frequency of the fundamental (F0) also
plays a certain role (Miller, 1953; Fujisaki and Kawa-
shima, 1968; Ainsworth, 1975).
Potter and Steinberg (1950, p. 812) first formulated
the hypothesis "that within limits, a certain spatial
present within the speech segment in question, as opposed to extrinsic contextual factors (cf. Ainsworth,
pattern of stimulation along the basilar membrane may
be identified as a given sound regardless of position
1975).
along the membrane."
"Openness" is an articulatory term, referring to the
degree of mouth opening or mandible depression. Its
use here to describe a perceptual dimension is justified by a close correlation between that dimension and
mandible depression (Kim and Fujisaki, 1974). This
eorretation is not compulsory, however, since subjects
can produce, apparently without difficulty, a closed
vowel with the mandible fixed in position for a very
open one (Lindblom and Sundberg, 1971; Lindblom et al.,
1979). Instead of "openness" some phonetieians use the
notion of vowel height, referring to the highest point
of the tongue, but this feature, if taken tirefatty, does
not admit some important phonotogicat generalizations,
since the steps in height of that point are not equal be-
tween corresponding front and back vowels (Ladefoged,
1967).
Anyway, phoneroeshave to be perceived in order to
be classified or distinguished,
ceived as an/a/,
and if something is per-
it is an /a/, regardless of its pro-
duction. Only the features perceptible to the listener
can play a distinctive role in the process of speech
communication; articulatory,
features
cannot.
Distinctive
or even proprioceptive
features,
then, should
This is suggested by the dis-
placement of the formant frequency pattern as observed
in a vowel spoken by a man, a woman, and a child. This
displacement is by and large uniform on a scale of
auditory critical bands, i.e., a scale representing criti-
cal bands with unit width (1 Bark), which also may be
considered as a tonality scale. Besides its proportionality with perceived tonality this scale is, in good
approximation, proportional to distances along the
basilar membrane (1 Bark • 1.3 mm • 150 haircells,
according to Zwicker and Feldtkeller, 1967). There
are, however, some deviationsin vowels spokenby
men, as compared to those spoken by women, which do
not directly fit this hypothesis. The female-male differences in tonality are shown in Table I for a sample
of vowels.
The values
in Table
I are
based
on a sum-
mary of investigations in several languages (Fant,
1975). Similar differences were found in the various
languages, but the spread was quite large. Therefore
deviations
less
than
0.5 Bark
in Table
I should
be dis-
regarded. Values for F0 and F4 were not published in
Fant (1975). The corresponding differences are roughly
1 Bark, equal to 100 Hz for F0 and to 18% of the frequency of F4.
An acoustic feature correlated with openness is the
In a previous perceptual study, it was found that the
tonality distance between the first formant and the
fundamental, rather than the position of the first formant alone, is decisive for the identification of oneformant vowels, as long as this distance remains within
frequency of the first formant (F1). However, because
the range found in natural vowels (Traunmfiller, 1976).
be described in perceptual terms. A description in
acoustic terms is possible, but complicated by the
transformations occurring in the organ of hearing.
1465
J. Acoust.Soc.Am. 69(5), May 1981
0001-4966/81/051465-11500.80
¸ 1981 Acoustical Society of America
1465
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
TABLE I. Mean female-male
tonality differences in vowels
of several languages. Values in Bark calculated from Fant
(1975). "*" means that the difference is estimated to be
roughly 1.0 Bark.
Number
Vowel
F0
i
F 1
;•
0.2
e, !
*
0.4
•e
*
1.3
F2
1.3
F3
F4
of
languages
0.8
*
1.4
1.1
*
5
1.1
1.1
*
6
TABLE III. Feature analysis of the non-nasal monophthongs
in Table II. 1 All the vowels are tense.
Vowel
i
e
•
a•
a
o
•
o
u y
•
Openness
1
2
3
4
5
4
3
2
1
1
Front/back(+/-)
+
+ +
+
+
Roundedness
.....
+
....
+
+
+
+
a•
•
2
3
4
+
+
+
+
+
+
6
o,o
*
0.8
0.7
0.9
*
5
o
*
0.3
0.3
1.1
*
6
u
*
0.2
0.0
1.4
*
6
y
*
0.0
1.2
1.1
*
4
p
*
0.2
1.0
1.0
*
4
This conclusion was, however, based on only two listeners, while the results obtained from two other listeners were not sufficiently clear.
The aim of the first of the present experiments is to
establish more reliably the decisive criteria for the
identification of one-formant vowels. The experiment
was planned to cover the whole range of variation in F0
the same feature will be collapsed.
Experiments 2, 3, and 4 are designed to find out
whether
the conclusions
drawn
from
the identifications
of one-formant vowels can be generalized to the perception of natural vowels. These experiments are intended to determine the possible role of the second and
higher formants, for the perception of openness. To
this end, F1 and F0 were systematically displaced in
frequency in synthetic versions of some natural vowels
in three ways: First, F1 and F0 simultaneously, keeping their tonality distance constant, second, F1 only,
and third, F0 only. All other characteristics were kept
at their original positions in frequency. The particular
natural
vowels
were
chosen
so that
some
information
(50 to 700 Hz) and in F1 (150 to 1480 Hz) that can be
about the perceptual distinction between front rounded
observed in natural speech.
and unrounded
It is an advantage to use listeners
whose native di-
alect employs many distinctive degrees of openness
without concomitant distinctions such as tense/lax or
diphthongization.
Such subjects are compelled by their
language to make careful distinctions along the dimension of openness. Moreover, interference from secondary features will be minimal in their responses. In
the eastern Central-Bavarian
dialect, used by the subjects in this experiment, as spoken in lower Austria
(Weigl, 1924/25), and by the older generation in Vienna
(Kranzmayer, 1953; Koekkoek, 1955), there appear to
be five distinctive
degrees of openness.
Table II shows the vowels occurring in stressed positions in the dialect of the subjects. All the vowels occur
in the sequences'V:(C) and 'VC: with durations as described by Banherr (1976)for westernCentral-Bavarian.
There is no perceptible difference in timbre associated
with a difference in duration, although formant frequen-
cies are slightly different (cf. Bannert, 1976). In syllables that cannot be stressed, there are a few lax (centralized) vowels and syllabic nasals and laterals. Table
HI shows the feature analysis • of the non-nasal monophthongs in terms of which the experimental results will
be presented. Identification responses of vowels sharing
vowels
could
also be obtained.
I. METHOD
A. Subjects
12 male and 12 female secondary school pupils from
Amstetten, Lower Austria, served as subjects. They
were between 16 and 18 years of age and had grown up
in the environment of the dialect spoken in the western
region of Lower Austria.
The subjects were divided
into four groups of six participants each. All 24 subjects participated in each of the four experiments. The
results
obtained
from
one of them
were
excluded
from
evaluation in all experiments because of several response omissions.
B. Stimuli
The stimuli were generated by digital simulation of
a terminal-analog
speech synthesizer.
Four repetitions
of each stimulus were recorded ontape with a high-frequency limit of 8 kHz and with due precautions for highquality reproduction. Stimulus durations and intervals
between repetitions and between different stimulus
quadruplets were as shown in Fig. 1. The fundamental
frequency as a function of time was as shown in Fig. 2
for
all
stimuli.
This
contour
and the
durations
were
based on isolated vowels spoken by the author, who was
familiar
with
an eastern
Central-Bavarian
dialect
TABLE II. Vowels occurring in stressed position in the eastern Central-Bavarian
dialect spoken in Amstetten, Austria.
Openness
Vowel set
5s
r
7s
T
,
5s
'1
FIG. 1. Intervals between four repetitions of each stimulus
and between different stimulus quadruplets in experiments
I through 4.
1466
J. Acoust.Soc. Am., Vol. 69, No. 5, May 1981
H. Traunm511er:Perceptionof vowelopenness
1466
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
located one frame on the answer sheets containing all
the different vowel symbols. The subjects were told
not to have any expectations concerning the frequency
of appearance of different sounds and that the same
vowel might be heard many times in sequence. The
vowels in experiment 1 were described to the subjects
as being incomplete in some sense, and those in the
1.2 FO
1.0FO
-1
other experiments as being artificial
some
300 ms
-
II. EXPERIMENT I: VARIATION OF F0 AND
FIG. 2. Voice source amplitude and frequency contour of all
stimuli. "F(•' here designates that value of the fundamental
frequency which was parametrically varied.
FORMANTFREQUENCYIN ONE-FORMANT
VOWELS
A. stimuli
The stimuli
closely related to that of the listeners. In experiments
2, 3, and 4, a fifth formant with a fixed position at
4.5 kHz and a 6 dB/oct emphasisabove 2.0 kHz were
were
Within each series,
13 series
of one-formant
vowels.
the tonality of the formant,
here
denotedF', was increased in steps of 0.5 Bark, beginning with a position 0.5 Bark above the nominal tonality
of F0, and endingat '/.0 Bark aboveF0. The tonality
used for an approximate correction for the missing
higher formants. The formant bandwidths B(F), defined
at 3 dB, obeyedthe function B(F)=0.05 f(F)+ 50 Hz,
where f(F) is the center frequency of the formant. The
voice source spectrum had a 100 Hz, 6 dB/oct lowpass contour, obtained by low-pass filtering a train of
pulses with a duration of 50 •s each.
C. Procedure
The stimuli were presented binaurally via headphones,
type Sennheiser HD 414, from a calibrated tape recorder, type Uher 4000. Loudness was set at a comfortable
listening level.
("verf•/lscht") in
sense.
The subjects were instructed to answer
scale was taken from Zwicker and Feldtkeller
(196'/,
p. '/4). The range of F' was restricted to 150 Hz (1.5
Bark) minimum and 1480 Hz (11.0 Bark) maximum. The
fundamental frequency was subsequently increased by
0.5 Bark for each series, beginning with 50 Hz (0.5
Bark), and ending at '/00 Hz (6.5 Bark). The number
of different stimuli thus obtained was 166, located in
13 series with at most 14 different stimulus quadruplets.
The sixth series, with F0 - 300 Hz, was presented twice,
the first time being in the beginning of the experiment,
before the series with F0 =50 Hz. This first presentation served to allow the subjects to get accustomed to
their
task.
by encircling the symbol of the perceived vowel on
answer sheets.
Each of the 13 vowels was given as a
permitted response to each stimulus in all experiments.
The subjects were allowed to encircle several vowel
symbols in cases of doubt. Processing the results,
polysymbolic responses were then fractionally attribu-
ted to the different symbol.
s. Before presentingthe
stimuli, roughly 10 min were used to instruct the subjects about the phonetic value of the symbols appearing
on the answer sheets by telling them and questioning
them about words containing the corresponding vowels.
Care was taken to check that the subjects could use cor-
rectly the symbolsfor [e]/[•] andfor [•]/[ mi. These
distinctions lack consistent correspondences in standard
High German and are mostly neglected in dialect writing. The symbols, their arrangement, and corresponding phonetic values were as shown in Table IV. The
symbol "j." was taken from dialect writing, the other
B. Results
The results, pooled over the 23 subjects and in
classes of the vowels with the same degree of openness,
are displayed in Fig. 3(a) for each series of stimuli
with the same F0. The initial presentation of the series
with F0 =300 Hz is not shown. Figure 3(b) shows the
corresponding results concerning perceived rounded-
nessand"place of articulation"(front/back).
Figure 4 shows the dominantly perceived degrees of
opennessand regions of at least 50% agreement among
subjectsfor the wholerangeof F 0 andF' usedin this
experiment, together with the tonality positions of the
first four partials.
The results were reasonably uniform across subjects
with respect to openness, and the pooling therefore
symbolsreflect standardHighGermancorrespondences seems
and letter names.
To each stimulus quadruplet was al-
TABLE IV. Symbols (left) and their phonetic values (right), in
arrangement as on employedanswer sheets. ("a" was represented twice with the objective to facilitate encircling in eases
of doubt.)
u
o
j
au
U
•
•
i•ul
i
e
•
ai
a
a
u
o
3
0
Y
d
m
(•
i
e
c
a•
a
a
justified. There were, however, systematic
differences between subjects with respect to place of
articulation.
One subject heard no front vowels at all,
and two other subjects heard no front vowels at F0 =630
and 700 Hz, and at 150 Hz, respectively. All other
subjects heard some front vowels in each F0-series.
It can be seen in Fig. 3(b) that the agreement among
subjects on perceived place of articulation of rounded
vowels was quite low.
In 90.3% of all cases, subjects responded by encircl-
ing the symbol of a single phoneme, 9.5% were two1467
J. Acoust. Soc. Am., Vol. 69, No. 5, May 1981
H. Traunm•ller: Perceptionof vowelopenness
1467
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
1
2
3
/.
5
6
2
?
3
/.
5
6
?
1
(a)100••i0
Hz'' ..i.'
3
2
/.
•
6
?
/,00 Hz
100 Hz
450 Hz
•
0
100
510 Hz
0
100
0
100
?00Hz
700 Hz
0
1
lOO
2
3
/,
1
2
3
/,
350 Hz
u, y, i
....
.....
(1)
back rounded
o, o, ß (2)
-•,a•,e
....
1
2
3
/.
5
6
?
..........
a
front rounded
(3)
......... o, a, a• (/,)
unrounded
(5 )
FIG. 3. (a)Perceived opennessof one-formant stimuli. Each frame represents results for a series of stimuli with the indicated
fundamentalfrequency. (b) Perceivedroundedness
and"place of articulation"of one-formantstimuli. Eachframe represents
results for a series of stimuli with the indicated fundamental frequency.
symbol, and 0.1% were three-symbol responses. There
were 0.2% response omissions.
C. Discussion
7 -
In Fig. 4, boundariesexclusively determined by the
tonalitydistancebetweenF' andF0 wouldbe parallel
to theF0 axis, while boundariesdeterminedby F' alone
wouldbe parallel to the lines restricting the range of
F' in the upper right and the lower left corner of that
6
5
figure. Figure 4, then, shows quite clearly that only
the distinction between openness 4 and 5 at fundamental
frequencies above 350 Hz is determined by the frequency
of the formant alone. This boundarybetween[a] and
3
less open vowels apparently cannot be pushed further up
than roughly 1.2 kHz.
The distinction between openness
1 and 2 is determinedby the distancebetweenthe for
mant and the fundamental.
2
The boundary remains at
roughly1.2 Bark for thewholerangeofF0. The
boundariesbetweenhigherdegreesof openness
roughly
follow the trend observed for the boundary between open-
1
0
ness 1 and2uptoaF0of
350Hz. BetweenF0=350Hz
and F0 =400 Hz, an abrupt change in response behavior
is observed. At F0's above 400 Hz, the distance between F' and F0 is again decisive, but vowels with
0
I
I
I
I
I
I
I
I
intermediate openness(degree 3 or 2) are not often
heard. The change in response behavior at F0 between
.02
.1
.2
.3
./,
.5
.6
.?
1
2
3
/,
5
6
7
tonality of fundamental (Bark)
fundamental frequency (kHz)
350 and 400 Hz can be explained in the light of a psycho-
FIG. 4.
acousticmodel, incorporatingthe conceptof a "second
integration band" which will be described in Sec. IV.
of openness(thick lines) and regions of at least 50%agreement
among subjects (dashedlines), showntogether with positions
of first 4 partials (thin lines). Encircled figures designate degree of openness(cf. Table II and Table III).
On the basis of the hypothesis that the stimuli were
1468
J. Acoust.Soc.Am., Vol. 69, No. 5, May 1981
One-forrnant stimuli:
Dominantly perceived degrees
H. Traunm[Jller:
Perception
of vowelopenness
1468
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
responses should have been larger.
identified by matching their total pattern of excitation
along the basilar membrane with stored reference
patterns for each vowel, it might be expectedthat oneformant vowels would be identified as back vowels,
III.
since
It was then suspected that
To test this possibility,
the stimuli were also
nevertheless the front
vowel
Thus
identifications
remained.
son to doubt the reliability
we have
of this result.
no rea-
4.5 kHz.
This is fur-
the stimuli were synthetic
Front
vowels
were
chosen
in order
to see
whether the tonality distance between F2 and F1 has
ther supported by the results of Agelfors and Gr•/slund
(1979) who found that when low-pass filtered at 470 Hz,
something to do with perceived
36 riB/oct, the vowels[i, y, a, u, e, •, o] were all iden-
roundedhess.
Figure 5 shows the kind of stimuli and their order
of presentation in the three experiments.
In each experiment the displaced characteristic was shifted suc-
tified predominantly as front vowels by Swedish listeners.
cessively in steps of 0.5 units on the modified tonality
scale described below and shown in Fig. 6.
In an investigationby Ainsworth and Millar (1972),
testing the effect of relative amplitude in two-formant
vowels, it was found that most but not all of their
English subjects gave predominantly back vowel responses when the amplitude of the second formant was
reduced to zero. It appears, then, that there are two
possible perceptual strategies for the distinction of
front versus back rounded vowels,
OF F0 AND
VOWELS
displaced. The purpose was mainly to test the perceptual saliency of the F l-F0 cue to openness against
imaginable conflicting cues based on the higher formants. The higher formantsF2, F3, andF4, as well
as the nondisplaced F1 or F 0 were fixed at the values
obtained spectrographically from two careful isolated
pronounciations of the vowels by the author. The formant positions and F0 values were as shown in Table
V. There was further a fifth formant, always fixed at
this might be due to a higher "formant" caused by disheard under low-pass filtering;
VARIATION
versions of natural vowels in which F1 and/or F0 were
have any prominent energy at high frequencies. One of
the subjects, in fact, never perceived any front vowels.
All the others heard, besides back vowels, also front
vowels, particularly front rounded ones. In the previous investigation (Traunmfiller, 1976), the subjects
tortion.
2 TO 4:
IN NATURAL
In these experiments,
they are dissimilar to front vowelq in that they do not
also heard front vowels.
EXPERIMENTS
F1 FREQUENCY
Several careful identifications, made by the author, of
one-formant vowels of essentially the same kind as
those used in experiment 1 led to the suggestion that the
dependence of perceived openness on the distance be-
tween F 1 and F 0 could be described uniformly if the
namely to rely on a
Bark scale of tonality was modified at its low-frequency
end. This modification is shown in Fig. 6. On the basis
of this modified scale, the comparatively small female-
second formant close to the third (present in front,
absent in front vowels), or to rely on prominent components at higher tonalities, naturally achieved by a
second formant close to the third (present in front,
male
differences
in F1
of the
same
vowels
can also
be
understood in terms of the hypothesis of movable spatial
patterns on the basilar membrane as mentioned in the
Introduction. At frequencies below 200 Hz, the dependence of this modified tonality measure on frequency is
weaker than at higher frequencies.
Therefore a formant
at, e.g., 350 Hz in a vowel with F0 at 200 Hz has to
be lowered by only 50 Hz if F0 is lowered by 100 Hz
and the tonality distance is to remain the same. Within
absent in back vowels). One-formant vowels, then, contain no direct cue to place of articulation since there are
no prominentcomponents
aboveF' at all, andthey contain conflicting indirect cues, the absence of a close
second formant indicating fronthess, while the absence
of higher formants indicates backhess. The lack of
agreement in perceived place of articulation can thus
be understood. The subjects, however, appear not to
the frame
of the
model
to be described
later
in Sec.
have been aware of missing information or conflicting
IV, the difference between the two scales is explained
cues, otherwise,
as an "end of scale
the number of multiple-phoneme
effect."
TABLE V. F0 and formant positions in the original natural vowels and their synthetic versions
used in experiments 2 through 4. Values in Hz (left) and in Bark (right).
Characteristic
Vowel
1469
F0
F1
200
150
140
130
frequencies
Characteristic
F2
F3
F4
F0
F1
F2
250
350
455
700
2300
2100
2000
1570
3400
2800
2600
2600?
3800
3400
3220
2950?
2.0
1.5
1.4
1.3
2.5
3.5
4.5
6.5
13.9
13.3
13.0
11.4
tonalities
F3
F4
16.5
15.3
14.8
14.77
17.2
16.5
16.1
15.67
190
275
2000
2400
3400
1.9
2.7
13.0
14.2
16.5
140
400
1860
2350
3400
1.4
4.0
12.5
14.1
16.5
130
130
470
600
1600
1425
2350
2520
3150
3050?
1.3
1.3
4.6
5.8
11.5
10.8
14.1
14.6
16.0
15.87
J. Acoust.Soc.Am., Vol. 69, No. 5, May 1981
H. Traunm511er:
Perception
of vowelopenness
1469
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
A. Experiment2
9
' i
I
i
i
i
i
i
i
i
8
In this experiment F0 and F 1 were shifted in the
vowels[i, e, •, •] from their original positions(Table
7
V) in such a way that the original distance between
F1 and F0 remained unchanged on the modified tonality
scale (Fig. 6).
5
6
The identification results are displayed in Fig. 7.
3
Responses were grouped in classes of the vowels with
the same degree of openness(left column) and in classes
with the same roundedness (right column). The original
2
-/
I
ß
0
vowel[•] wasapparentlya slightlytooopenvariant.
0
....modified
scale
i
i
I
.I
.2
.3
I
I
.4
.5
I
I
.6
.7
I
I
.8
.9
1.0
Several subjectsperceived it as an [m ].
frequency (kHz)
In all four vowel series the majority of judgments on
opennessremained the same at all positions of the F0
FIG. 6. Standardtonality (Bark) scale and modified scale
andF1 configuration.This fact impliesthat eitherthe
based on the hypothesis of movable patterns and on identifica-
subjects relied on the distance betweenF 1 andF0, or
they relied on information contained in the higher formants, or both. Experiments 3 and 4 were designed
to select among these possibilities. The results mitigate against the hypothesis that the distance between F2
i
ß
tions
of one-formant
vowels.
(or any higher formant) andF 1 (or F0) has anythingto
do with perceived openness in these vowels.
Although several subjects at some point jumped from
perceived openness 2 to openness 4 in the e and in the
•( series at increasi,.g F0, the sudden change in response
behavior observed at increasi,.g F0 in experiment 1 did
not appear in the • series in this experiment, where it
could be expected in analogy with the results of the oneform2nt vowel identifications.
This is likely due to in-
[
Exp. 2
ß0.1.2
.3
./,
.5
.6
F0,
F1
5O
5O
Exp. 3
=
o
o
1oo
1oo
50
5o
o
F1
0
lOO
lOO
5o
5o
e
Exp. 4
0 i •1
.0.! .2
i
I
i
.3
./,
.5
fundamental
F0
FIG.
5.
Stimuli in experiments
2, 3, and 4.
characteristics
as in Table
V.
Indicated
II I
•3.1 .2
I
.3
frequency
:•1 .5
,0
.6
(kHz)
Each small
frame represents one stimulus quadruplet. Phonetic symbols
are placed in frames representing original vowels with unshifted
.6
character-
istics were shifted in steps of 0.5 units on the modified
i
(i)
e,4
(2)
f, ,•e
(3)
,e,•
(,•)
(front)
ß
ß
rounded
originalvowel
tonality scale (Fig. 6). Order of presentation was left to right
FIG.
7.
Identifications
of vowels
in which F0 and F1 had been
and row by row, starting in the top left-hand frame. In the
first stimulus of experiment 2, i.e., in the "i" column, F0 and
F1 were lower than in the original vowel; in the other columns,
shifted uniformly on the modified tonality scale, while all
higher formants remained unshifted. Each frame represents
however,
results for stimuli with the higher formants at the positions
there was no vowel with lowered F0 and F1, because
F0 in the original vowels was too low to be lowered by 0.5
units. In experiment 3, F1 was lower than in the original
vowels in the beginuing of each series.
1470
J. Acoust.Soc.Am., Vol. 69, No. 5, May 1981
they had in the indicated original vowels. Left: Perceived
openness, right: Perceived roundedhess. In the "i" series,
no rounded vowel was perceived.
H. Traunm•ller: Perceptionof vowelopenness
1470
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
formation conveyed by the higher formants.
With respect to the hypothesis of vowel recognition by
pattern matching, at large shifts, the overall patterns
evoked by the vowels in this experiment would not match
internalized templates derived from natural vowels, no
matter how movable these templates were. Such an at-
tempt would inevitably result in a striking mismatch in
the pattern of either F0 and F1, or of the higher formants, if not of both. The observed front vowel responses to one-formant stimuli and to low-pass filtered
natural vowels also appear incompatible with such an
overall match.
These results are, however,
compatible
!'
with the hypothesis that relevant parts of the evoked
pattern are matched with stored reference patterns for
the F1-F0
cue alone.
ture-detection
The results thus suggest a fea-
process. We should, then, reformulate
.2
.3
.t,
.5
.6
.7
.8
the hypothesisof Potter and Steinberg (1950): Sounds
first
are not identified, but certain features of sounds are
recognized on the basis of spatial patterns of excitation along the basilar membrane regardless of position
along the membrane.
As for the perception of roundedness, we can conclude from the results in Fig. 7 that the distance be-
tween F2 andF 1 (or betweenF 2 andF 0) may play a
certain role, but it is not usually decisive,
otherwise,
we shouldexpect only /y•oe/ responseswhenF 1 (F0)
is increased by 3 Bark or more, andF2-F1 (F2-F0)
formant
i,y
e,e
(1)
(2)
• ,ee
ae,cz
a
(3)
frequency
(kHz)
(front) rounded
ß
&
origin(•!vowel
(5)
FIG. 8. Identifications of vowels in which oaly F1 had been
shifted. Each frame represents restfits for a series of stimuli
with F0 and higher formant frequencies as in the indicated
original vowels. Left: Perceived openness, right: Perceived
roundedhess.
consequently decreased by the same amount, in the
vowels based on/ice/.
There may be some differences
in the degree to which different
F2-F1
(F2-F 0) cue in roundedness recognition, but
observe
at all.
subjects relied on the
that in the i series
no rounded
vowel
was heard
Thus the distance bet{veen F2 and F1 is not
the major cue for the distinction of front rounded versus
front
unfounded
vowels.
F1 was shifted in the vowels
[e, •, •, ce] from Table V. The identificationresults
are displayed in Fig. 8. No systematic differences in
response behavior appeared across subjects in this experiment.
The boundaries between degrees of openness appear
to be located approximately at the same positions of F 1
in all four series of stimuli.
This conformity is particularly pronounced within the pairs of series based
either on rounded or on unfounded vowels only.
The regionofF 1 positionsfor which/•/responses
(openness3) were given in the series based on unrounded vowels is larger than in the series based on
rounded vowels and in experiment 1. Since the difference is accounted for by the higher formants, it may
these
within
narrow
limits
do have
an influence on perceived openness if there is a difference
in roundedness.
If there
is no difference
in roun-
dedness, however, the influence of the higher formants
appears to be negligible. If there were such an influence, it would be expected that the regions of F1 positions in which the original vowels were heard would be
enlarged in each series.
1471
fundamental is indeed the most important cue in perception of openness.
basedon [e] appearto indicatethatF 1 plays somekind
In this experiment
that
experiments 2 and 1, these results strongly suggest that
the tonality distance between the first formant and the
About the perception of roundedness, not much can
be said. Anyway, the results obtained with the stimuli
B. Experiment 3
be concluded
Since the possibility of F 1 to determine openness without reference to F0 has been ruled out by the results of
J.Acoust.
Soc.Am.,Vol.69,No.5, May1981
of role.
C. Experiment 4
In this experiment F0 was shifted upwards in the
vowels [e, •, ae] from Table V, up to the limit where
the distance
between
F1 and F0 was diminished
to less
'than 0.7 units on the modified tonality scale.
The identification results are displayed in Fig. 9.
Most subjects perceived the original vowel "ae" as an
[o•]. Apparentlythe positionsof F3 and/or F 4 in Table
V are not representativefor a typical /a•/. For this
reason, the results allow conclusionsaboutthe F0 dependence of perceived openness only.
Vowels with formant positions of the original vowels
[e] and [e] were perceivedas havinga lower degree of
openness at higher F0. This was also true for 10 subjects when listening to vowels with the formant positions of the original vowel" ae." They typically heard
the series with increasing F0 as /o•- ce-•-y/.
For
another 5 of the subjects there was no substantial F0
dependence in perceived openness. They perceived
predominantly /oe/ and a few vowels with openness3 or
5 at all positions of F0.
Another 5 of them followed the
H.TraunmiJller:
Perception
of vowelopenness 1471
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
-
"e"
--
1
IV. EXPLANATORY MODEL
2
A. General description
50
"e"•
50
Some of the major findings in the present experiments
can be explained via a general model of auditory pat-
tern matching and identification, shown in Fig. 10 (cf.
Traunm•iller, 1979). Some other findings can be incorporated into such a model.
I I T ,,oo
lOO
"as"
The acoustic stimulus reaching the ear, at point (1)
in the model, can be completely described by p(t) until
50
10 subj.
0
•
,
•1 1
I, I I I •oo
,
2
3
F1- FO
5 subj.
,
4
it reaches the tympanic membrane,
, 0
mension
5
I
I
I : •oo
(sc.u.)
front
back
............
5 subj.
I
!
•
E
1
i
2
F1-
i
•
FO
4
sc.u.
=
!
ß
it reaches
Bark
window.
a frequency-to-place
The resulting place di-
original vowel
(sc.u.}
in the 8th nerve and the psychoacoustical magnitude
i,y
(loudness) of a stimulus or parts of it. It has been
claimed (Traunm•iller, 1979), that magnitude for the
purpose of spectral matching and phonetic identification
(1}
aso(zoo (/,)
a
should be scaled on a linear scale of just noticeable
(5)
differences. The first step on such a scale of •rrom{n½nc½ is the threshold of masking.
FIG. 9. Identifications of vowels in which only F0 had been
shifted. F0 decreases from left to right in each frame representing results for stimuli with formants as in the indicated
original vowels. Left: Perceived openness, right: Perceived
roundedne
and no further di-
the oval
mension is largely maintained along the neural pathway
up to the auditory cortex (e.g., at the collicular level
according to Rose ½! al., 1963). Things are more complicated concerning magnitude. There is no close correspondence between the magnitude of neural activity
rounded
....
i 0
5
until
Within the cochlea, however,
transformation is performed.
0
I
is added
ss.
The informative
parts of a spectrum are those above that threshold. Accordingly, at point (2) in the model we have a prominence-density versus tonality representation of the
stimulus, incorporating the effects of frequency resolution
into
critical
bands.
After point (2) in the model, the signal is distributed
to several channels for the extraction of specific qualtrend of the first group up to F0 = 445 Hz, then changing
abruptly to the behavior of the second group. They
typicallyperceivedthe series as /o•-ce-•{-c•, •e/.
The remaining 3 subjects showed other combinations of
both trends.
-
The results demonstrate the dependence of perceived
phonetic quality on J70 even when all formants maintain
their positions. Moreover,
the results obtained with the
"•e"-based
stimuli demonstrate a limit of this F 0 de-
pendence.
that
the
This limit can be explained if we suppose
smaller
distances
between
characteristics
are
more important than the larger ones. In some sense,
then, characteristics
close in tonality are more apt to
fuse into a perceptual gestalt. This is compatible with
the results of experiment 2, where the large distances
between the higher formants and J71 proved to be of little
important. In those vowels, the partials in the region
ities. Beside the channel for spectral matching, there
is a similar channel for pitch extraction. In the box
labeled "specific quality extraction" certain transformations are performed which result in a representation (3)
that is largely freed from variation due to characteristics which are irrelevant to the particular specific
quality, e.g., intensity in pitch extraction and partial
resolution in spectral matching.
The representation at point (3), then, is entered into
short term memory and may be compared with a rep-
resentationpreviouslyentered into this memory and/or
with permanently stored, i.e., learned, templates
representing phonetically significant patterns. The
stimulus will then be categorized with the template(s)
most similar to the stimulus representation at point (3).
of F1 and F0 apparently fuse into one gestalt, whereas
those in the region of the higher formants fuse into
another gestalt quite independently of the first. In an
ordinary [ •e], on the other hand, the distancesbetween
F2 and F 1 and between F 1 and F0 are roughly the same
extraction
of
otherqualities
•
!
.
phoneticI
templatesI
I
categ½
r•zation
P'
'
cornpc
•Json ._
II st•e
term
J
ac--,stic
Iperipheral
IJI specific
IIj
st•r•ulus
•--•--H trans- •
quality •
'
'
. short
(5 Bark), and thus roughly equally important. Some of
the subjects, then, relied more on the F2-F1 cue,
and possibly on the higher formants, than on the Fi-F0
cue. Nevertheless, fewer than five of the subjects
showed no reliance
based
1472
at all on the F1-FO
cue in the "a•"-
stimuli.
J. Acoust.Soc.Am., Vol.69, No. 5, May1981
response
-
p•o..tio. -
E[G. 10.
tff•csUon
•sychoscousUc model of s•ct•sl
described
I
I
mAtchrig And 1den-
•n
H. Traunm•ller'Perception
of vowelopenness 1472
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
The results from the present experiments, indicating
that parts of vowel stimuli distant in tonality fuse into
distinct perceptual gestalten, demonstrate that these
phonetically significant patterns do not correspond to
whole phoneroes, but possibly to distinctive features.
The templates can be thought of as rigid but within
certain limits movable in tonality.
The box labeled "specific quality extraction" represents the location of the spectrum envelope reconstruction. If we accept the view that the difference between spectral representations of matched stimuli in
some way determines their perceived phonetic simil-
arity, it is necessary to explain how high-pitched vowels are matched to low-pitched ones. This leads to a
seemingly inevitable gross mismatch without spectrum
envelope reconstruction or an equivalent process of
spectral integration, since the lowest partials in highpitched vowels (F0> 150 Hz) are clearly resolved at a
peripheral
stage.
In matching experiments where one-formant
were
matched
to two-formant
stimuli
stimuli
it has been shown
(Chistovich et al., 1979) that the single formant will be
located
in the middle
between
both formants
in the two-
formant stimulus as long as the distance between the
formants
in the two-formant
stimulus
is less than the
critical distance of 3.0 to 3.5 Bark, provided that their
levels are approximately the same. If the formant separation is increased above the critical distance, the
location of the matched single formant will suddenly
approach one of the two formants. These results suggest the reality of spectral integration over a range of
roughly 3 Bark.
The results published so far do not allo/v a determina-
tion of the shape of the integration function. Tentatively,
a Gaussian shape can be assumed. This appears preferable to a rectangular integration window. The stimulus
representation at point (3), then, is determined by a
convolution of the prominence representation at point
(2) with a Gaussianshapedspreadingfunction with (y= 1.5
Bark.
If there is a second integ•rationban• of the described
kind, with a width of about 3 Bark, we can expect some
anomalies at both ends of the tonality scale, since the
convolution cannot reasonably be extended further than
the ends of that
scale.
the familiar
Bark. Then, if the prominence of the fundamental is
not as high as it is when the formant is very close to
it, a very open vowel will be heard for the following
reason: In a natural [ • ] or an [ce] with F0< 370 Hz
there Will be a peak at the fundamental, which is quite
prominent in those vowels, and another peak close to
the first formant, located at a distance larger than
3.0 Bark above the fundamental. If F0 is larger than
370 Hz, this kind of pattern will result even if F1-F0
is less than 3.0 Bark, since the position of the second
peak will then be determined by the second and third
partials rather than by the theoretical F 1. The model
predicts the sudden change in response behavior observed
between
critical
band--width
1
Bark; second, the range of spectral integration in
envelope reconstruction, i.e., a second integration
band--width 3 Bark; and further, with a limited range
of gestalt fusion.
Even if the distance between partials closest to a
formant is less than the critical distance, due to
the nonrectangular integration function we can expect
a slight tendency for peaks to be attracted to partials.
The
effect
be that the boundaries
between
ß
traced •n Fig. 4 for the boundary between openness 3
and 4 at F0>200
Hz.
Thequestion
ofperceptual
interaction
between
single
partials and F1 had been investigated previously
(Chistovich, 1971; Carlson et al., 1975; Florin, 1979),
but the results were difficult to interpret and did not
quite match the expectations. The concepts of movable
spatial patterns, i.e., of movable templates for certain
features, and of a second integration band appear to explain the consistent findings and leave some effects observed at small variations in F0 to be interpreted as
statistical or other noise. The concept of prominencedensity might explain the discrepancy between the func-
tions of equal loudness and equal phonetic significance
ßof partials in the region of F1 found by Mushnikov and
el, proportionality
pected.
Let us consider how the F1-FO cue, which was found
to be decisive for the perception of openness, is reflected in a blurred prominence-density representation
above.
If the distance between the two partials closest to the
J.Acoust.
Soc.Am.,Vol.69,No.5, May1981
between these functions is not ex-
If F1-F0 is less than 3 Bark, the pattern will contain only one peak, and it can be described by its width
or by the distance between the peak and the low-tonality
flank of the representation. For F0 = 150 Hz, this lower
flank will be located at the end of the scale.
B. Evaluation of predictions
1473
of this would
different degrees of openness would tend to follow lines
in the middle between partials. Such a tendency can be
decreased
as described
F0 = 350 and 400 Hz.
Chistovich (1972). Within the frame of the present mod-
We have, then, to reckon with at least two integrarive
processes of different range: First, peripheral frequency resolution,
formant is less than 3 Bark, there will be a peak in the
representation close to the tonality position that corresponds to the resonance frequency of the formant.
As can be seen in Fig. 4, at F0> 370 Hz, the distance
between the first two partials is larger than 3.0 Bark.
Then, there will be two peaks, one at the fundamental,
and one at the second partial or above. The third partial will contribute to this peak, since the distance between sequential higher partials always is less than 3
'
further
If F0 is
than to 150 Hz in such a vowel with
fixed F1, the width of the representation will not increase any more. The F0 dependenceof openness will
stop at a lower limit of 150 Hz. However, since the
shape of the secondintegration band is not rectangular,
there will be a transitional region of about 100 Hz. This
explains most of the difference between the Bark scale
and the modified scale shown in Fig. 6. To this extent,
H.Traunm•ller'
Perception
ofvowelopenness 1473
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
then, the modification is not a real modification of the
tonality scale. This end-of-scale effect also explains
the comparatively small female-male
differences in
F1 of vowels withF1-F0<3
Bark (see Table I). The
effect did not emerge very clearly in the results of experiment 1.
If the distance
between
two formants
is less
than 3
Bark, they will merge. In an [a] F 1 andF2 are roughly
equally prominent and just close enough to merge, but
they are not in [ae] and Joel. The results of Chistovich
et al. (1979) tell us that a single formant will be placed
betweenF 1 andF2 in an [a] in order to achievethe best
match. Therefore the boundary between openness 4 and
5 in Fig. 4 does not coincide with the position of F1 in
natural vowels. This boundary is shifted upwards by the
influence of the natural F2 on the internalized phonetic
template, and therefore the band of (F1-F0) values in
which openness 4 was perceived is so much wider than
those for openness 1, 2, and 3. The model assumes the
task of identification to be equivalent to that of matching,
in the sense that
in both cases
a stimulus
is matched
with a pattern in memory, although stored either permanently or just temporarily.
Although the present lack of a validated measure of
prominence restricts a quantitative treatment, the
model can be said to account satisfactorily for the experimental results. It remains, however, to be established whether any modifications are necessary in
order to explain phenomena that might appear in the
range above 1.5 kHz.
ACKNOWLEDGMENTS
I am grateful to civ. ing. Peter Branderud, who arranged the vowel generating program, and,to fil.kand.
Karin Holmgren for their assistance at the computer,
and to Hofrat Dr. Leo KlingenbSck and his staff at the
Bundesrealgymnasium in Amstetten for making localities available and for recruiting subjects. I am further
indebted to Professor Bj6rn Lindblom and Dr. James
Lubker for valuable comments on the manuscript.
l In a generarivephonological
description,the front rounded
series is derived from underlying front unfounded vowels
of the same degree of openness, followed by an /1/. This,
together with the symmetric occurrence or non-occurrence
of -e diphthongsand of nasalized phoneroesamong front and
back vowels (Table II), appears already to provide enough
support for the feature analysis given in Table III. If [a] is
considered neither front nor back, as in Table III, it can be
specified without use of a 5th degree of openness, but a description of its acoustic manifestation as the vowel with the
highest F1 appears to call for a 5th degree.
2The"secondintegrationband"shouldnot be confusedwith the
"second filter" invoked in order to explain the high peripheral
frequency selectivity of the ear (Zwicker, 1974; Duffuis, 1974,
1977).
The feature
in common to these notions is that to both
there is a corresponding, more peripheral, first thing,
namely the I Bark critical band and the mechanical frequency
selectivity of the basilar membrane, respectively.
Institute of Linguistics, University Stockholm, Rep. 1, 10-12.
Ainsworth, W. A. (1975). "Intrinsic and extrinsic factors in
vowel judgements," in Auditory Analysis and Perception of
Speech, edited by G. Fant and M.A.A. Tatham (Academic,
London), pp. 103-113.
Ainsworth, W. A., and Mfilar, J. B. (1972). "The effect of
relative formant amplitude on the perceived identity of synthetic vowels," Lang. Speech 15, 328-341.
Banheft, R. (1976). Mittelbairische Phonologieauf akustischer undperzeptorischer Gr•.ndlage, Travaux de l'institut
de linguistique de Lurid, edited by B. Maimberg and K. Had-
ding (CWK Gleerup, Lurid), Vol. X.
Carlson, R.,
Fant, G., and Granstr•m,
B. (1975). "Two-
formant models, pitch and vowel perception," in Auditory
Analysis and Perception of Speech, edited by G. Fant and
M. A. A. Tatham (Academic, London), pp. 55-82.
Chistovich, L. A. (1971). "Problems of speechperception,"
in Form and Substance, Phonet. Linguist. Pap. presented
to Eli Fischer-J•rgensen, edited by L. L. Haremerich,
R. Jakobson, and E. Zwirner (Akademisk Forlag, Copenhagen), pp. 83-93.
Chistovich, L. A., Sheikin, R. L., and Lublinskaja, V. V.
(1979)." 'Centers of gravity' and spectral peaksas the determinants of vowel quality," in Frontiers of Speech Communication Research, Festschrfft for Gunnar Fant, edited by
B. LindblomandS. •hman (Academic,London),pp. 143-157.
Duffhuis, H. (1974). "An alternative approachto the second
ffiter," in Facts end Models in Hearing, edited by E. Zwicker and E. Terhardt (Springer-Verlag, Berlin), pp. 100103.
Duffhuis, H. (1977). "Cochiear nonlinearity and second filter-a psychophysical evaluation," in Psychophysics and Physiology of Hearing, edited by E. F. Evans and J.P. Wilson
(Academic, London), pp. 153-163.
Fant, G. (1975). "Non-uniform vowel normalization," Speech
Transre. Lab. Quart. Prog. Stat. Rep., 2-3/]975, 1-19,
Royal Institute of Technology, Stockholm.
Flor6n,fk. (1979). "Whydoes[a] change
to [•] whenF0 is
increased?," phonetic experimental research at the Institute
of Linguistics, University Stockholm, Rep. 1, 13-23.
Fujisaki, H., and Kawashima, T. (1968). "The roles of pitch
and higher formants in the perception of vowels," IEEE Trans.
Audio Electroacoust. AU-16, No. 1, 73-77.
Kim, B., and Fujisaki, H. (1974). "Measurement of mandibular control in vowels and its relevance to the articulatory
description of vowel systems of Korean and Japanese," in
SpeechCommunication, Vol. 2 (Almqvist & Wiksell, Stockholm, and Wiley, New York), pp. 277-284.
Koekkoek, B. J. (1955), "Zur Phonologie der Wiener Mundart," Beitr. zur deutschen Philologie,
W. Mitzka (Giessen), Vol. 6.
herausgegeben yon
Kranzmayer, E. (1953). "Lautwandlungenand Lautverschiebungen im gegenw•rtigen Wienerischen," Z. f. Mundartforschung 21, 197-239.
Ladefoged, P. (1967). Three Areas of Experimental Phonetics (Oxford U. P., London), pp. 62-75.
Lindblom, B., Lubker, J., and Gay, T. (1979). "Formant
frequencies of some fixed-mandible vowels and a model of
speech motor programing by predictive simulation," J. Phonet. 7, 147-161.
Lindblom, B. E. F., and Sundberg, J. E. F. (1971). "Acoustical consequences of lip, tongue, jaw, and larynx movement," J. Acoust. Soc. Am. 50, 1166-1179.
Mushnikov, V. N., and Chistovich, L. A. (1972). "Method
for the experimental investigation of the role of component
loudnesses in the recognition of a vowel," Sov. Phys.Acoust. 17, 339-344. Translated from Akusticheskii Zhurnal
17, 405-411 (1971).
Peterson, G. E., and Barney, H. (1952). "Control methods
Agelfors, E., and Gr•islund, M. (1979). "Vowel identification
and spectral slope," phonetic experimental research at the
1474
J. Acoust.Soc.Am., Vol. 69, No. 5, May 1981
used in a study of the vowels," J. Acoust. Soc. Am. 24,
175-184.
H. Traunm•ller: Perceptionof vowelopenness
1474
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06
Potter, R. K., and Steinberg, J. C. (1950). "Toward the
specification of speech," J. Acoust. Soc. Am. 22, 807-820.
Rose, J. E., Greenwood, D. D., Goldberg, J. M., and Hind,
J. E. (1963). "Some discharge characteristics of single
neurons in the inferior colliculus of the cat. I. Tonotopical
organization, relation of spike-counts to tone intensity, and
firing patterns of single elements," J. Neurophys. 26, 294320.
Traunm'dller, H. (1976). "Einige Aspekte der Wahrnehmung
quasistation'•rer Vokale," Papers of the Institute of Linguistics, University Stockholm, 32, 8-23.
Traunm'dller, H. (1979). "Essentials of a psychoacoustic
1475
J. Acoust.
Soc.Am.,Vol.69, No.5, May1981
model of spectral matching," phonetic experimental research
at the Institute of Linguistics, University Stockholm, Rep.
1, 49-63.
Weigl, H. (1924/25). "Die nieder•sterreichische ui-Mundart,"
Teuthontsta 1, 149--186.
Zwicker, E. (1974). "A 'second filter' established within the
scala media," in Facts and Models in Hea•ing, edited by
E. Zwicker and E. Terhardt (Springer-Verlag, Berlin), pp.
95-99.
Zwicker, E., and Feldtkeller, R. (1967). Das Oh•
Nach•ichtenempfdnge•,zweite neubearbeiteteAuflage(S. Hirzel Verlag, Stuttgart).
H. Traunm[•ller:
Perception
of vowelopenness 1475
Redistribution subject to ASA license or copyright; see http://acousticalsociety.org/content/terms. Download to IP: 130.237.171.143 On: Wed, 19 Nov 2014 11:18:06