Fairbanks, Grant. Experimental Phonetics – T11

The Influence of Consonant Environment upon the Secondary Acoustical
Characteristics of Vowels *1

Arthur S. House and Grant Fairbanks
Speech Research Laboratory, University of Illinois, Urbana, Illinois
(Received September 24, 1952)

The consonant environments of vowels were varied by forming nonmeaningful stimulus syllables consisting
of 72 combinations of six vowels and 12 consonants. The syllables were spoken by subjects, and the
duration, fundamental frequency, and relative power of the vowels were measured. All three factors varied
significantly in response to changes of the consonant environment. The variations were systematically related
to the attributes of the consonants, the most powerful attribute being the presence or absence of vocal fold
vibration, followed by manner of articulation and place of articulation, in that order.

Among the acoustical investigations of vowels,
experiments employing wave analysis have
naturally been most numerous, while studies of the
secondary characteristics — duration, fundamental frequency,
and intensity — have been relatively few.
Especially has this been true of variation in these
secondary characteristics which may be systematically
related to the widely varying consonantal environments
of vowels in words.

The purpose of the present experiment has been to
pursue the problem raised by such findings with a more
extensive and systematic phonetic design. Its general
plan was to place vowels in various consonant environments,
to cause them to be spoken by subjects, and to
make the appropriate physical measurements. Representative
vowels were used, and the following articulatory
characteristics of the consonants were controlled:
the presence or absence of vocal-fold vibration; variations
in the manner of production (fricative, stop-plosive,
etc.); variations in the characteristic place of
articulation (bilabial, velar, etc.).

I. Procedure

Materials

After considerable study of the various alternatives,
it was decided to construct stimulus materials in which
only one consonant influence was present in each item.
Syllables in which the vowel is both preceded and followed
by the same consonant, as in the word cease,
appeared to be suitable. Study of all such symmetrical
syllables led to the conclusion that appropriate materials
would be provided by restricting these items to 72,
involving 12 consonants in combination with six vowels.

The following 12 consonants were selected: [p], [b],
[t], [d], [k], [g], [f], [v], [s], [z], [m], [n]. It will
be seen that the first 10 sounds are voiceless-voiced
cognate pairs, providing direct contrast for the voicing
factor. In manner of production, six stop-plosives, four
fricatives, and two nasals are available, in that order.
Differences in characteristic place of articulation are
provided by the bilabials [p], [b], and [m]; the labiodentals
[f] and [v]; the post-dentals [t], [d], [s], [z],
and [n]; and the velars [k] and [g].

The six vowels chosen were [i], [e], [æ], [ɑ], [o],
and [u]. These vowels span the range of tongue,
mandible, and lip positions. They also vary in certain of
their secondary acoustic characteristics as reported by
Crandall, 1112 Parmenter and Treviño, 1213 Heffner and others
and Black 1314 for duration, by Crandall, Black, and
Taylor 1415 for fundamental frequency, and by Fairbanks,
House, and Stevens, Black, and Sacia and Beck 1516 for
relative intensity.

Combinations of these consonants and vowels resulted
in a mixed list of words and nonsense syllables,
which appeared to be unsuitable, since control over
semantic influences was regarded as important. After
various attempts, it was decided to prefix each item by
an unstressed syllable, creating bisyllabic nonsense items
with iambic stress patterns, and to select orthographic
forms that would yield least meaning. Eventually,
[hə] was selected as an appropriate initial syllable,
since its component sounds are easily pronounced,
usually neutral, and seemed likely to have minimal
effect upon adjacent sounds. Table I shows the 72
stimulus items in the form used in presentation to the
subjects.

Table I may also be regarded as depicting the essentials
of the statistical design. With all items produced by
all subjects, the triple-classification scheme for analysis
of variance was employed to test the significance of the
variances attributable to vowels, consonants, subjects,
and their first-order interactions, and to enable the other
statistical manipulations. 1617

Subjects

Ten male students enrolled in elementary speech
courses at the University of Illinois served as subjects.
The mean age was 20 years, six month., and the individuals
ranged from 18 years, seven months to 26 years,
six months. Thirty-five potential subjects were interviewed
by two judges trained in phonetics and speech
pathology, who screened their speech for aberrations in
pitch, loudness, duration, voice quality, articulation,
and pronunciation, and questioned them concerning
their hearing. The ultimate subjects were without
speech disorders, had no history of hearing pathology,
and spoke some form of general American dialect. In
the case of the intensity data only, 10 additional subjects,
selected in the same manner, were added to the
basic group.

Apparatus

The procedure involved phonograph recording of the
subjects' responses to the stimulus items. Equipment
was arranged conventionally in a two-room laboratory,
consisting of a sound-treated room and a control room.
A General Radio type 759-B sound level meter, with
its associated Brush 9898 crystal microphone, was set
85to flat response and 60 db attenuation, and its output
fed to a Presto 92-A recording amplifier, with equalizer
setting out. The recordings were made at 78.26 rpm on
a Presto 8D-G turntable with a Presto 1-B cutting head.

The recordings were played for graphic recording of
relative power and for simultaneous verification of the
presence of the complete set of responses for each subject.
The system used included a Presto 64-A turntable
with a Lear PA-200 pick-up, a Goodell PRA-1 preamplifier,
a Sound Apparatus Company HPL-E highspeed
level recorder with 0-50 db potentiometer, and
a Goodell ATB-3 power amplifier for the monitor
speaker. The level recorder was set for fast stylus speed
and 10 mm/sec paper speed. Measurements of duration
were made on sound spectrograms recorded on a Kay
Electric Company sona-graph. The above reproducing
system was employed. Oscillograms used in the. measurement
of fundamental frequency were obtained
through the use of a locally constructed instrument
similar to that described by Cowan. 1718 The instrument is
essentially a special type of oscillograph for the photographing
of signals stored on phonograph records. A
turntable is mounted on a large drum around which
photographic paper is wrapped. During the playing of
the record, the turntable and the drum mo/e synchronously,
and an optical lever, activated by a crystal pickup
and amplifier, is manually lowered by means of a
helical screw mechanism to spiral the trace about the
drum.

Experimental Procedures

The phonograph recordings were made with the
subject seated comfortably in the sound-treated room
and with the microphone, mounted on a boom stand,
approximately 12 inches before his mouth in the horizontal
plane. An experimenter was seated immediately
in front of the subject and presented the stimulus items
visually at approximately four-second intervals. Each
item was typed in lower case on a 3 x 5 card. Two judges
trained in phonetics independently evaluated the acceptability
of each response, using reasonably liberal standards
of articulation and stress pattern. When either
judge did not accept a response, its stimulus card was
displaced backwards in the order of presentation by at
least eight items in order to present it to the subject a
second time. An auxiliary list was used at the beginning
of the recording session to familiarize the subject with
the situation and to permit adjustment of equipment.
The same auxiliary list was used to displace backwards
in time stimulus items so dose to the end of the list that
they could not otherwise be displaced by at least eight
items. The list of 72 items was randomized anew for
each subject.

Before actual recordings were made each subject was
given a short training period during which he read aloud
all items from a standard list arranged as in Table I.
He was assured that no premium would be placed on
accuracy of response as such. Instructions pertaining to
vocal behavior were kept to a minimum, and no attempt
was made to regulate pitch, duration, or intensity. The
subject was simply instructed to pronounce each word
as if talking to the experimenter.

II. Results

Influence upon Duration of Vowels

The identification of the beginning and end of a vowel
surrounded by consonants is an arbitrary act that is
both difficult and artificial. Location of these points was
aided by the relative clarity with which they are shown
in sound spectrograms. In this study, the spectrograms
were produced with a narrow, 45-cps, band response at
their base and a wide, 300-cps, response in the higher
frequencies, and covered a 0-3500-cps range. 1819 This
procedure simplified the identification of the voice bar
without weakening the vowel resonance areas. Identification
of specific sounds was facilitated by information
presented by Potter, Kopp, and Green 1920 and Joos. 2021
The preceding and following consonants were used to
find the general area of each vowel, but in all instances
the vowel limits were established in terms of “vowelness,”
based on the presence of voice and of resonance
areas associated with particular vowels.

The effect of the consonant environments on the
duration of the vowels is summarized in Table II and
Fig. 1. The left half of Table II shows the mean duration
of all vowels spoken by the 10 subjects in each of the
12 consonant environments. These means vary over a
range of 0.134 sec, and analysis of variance reveals
statistically significant differences between means to
exist at the 1 percent level of confidence. By use of the
t statistic, differences between the various means may
be evaluated with a requirement of 0.011 sec. Of the
total of 66 such intercomparisons, 59 exceed the minimum.

Further study of the duration means shows that they
vary systematically with certain characteristics of
consonant production. A comparison of voiceless environments
with their voiced cognate environments,
for example, reveals larger values for the voiced environments
in every case. All voiced environments,
furthermore, produced vowels that differed significantly
from all those produced in voiceless environments. When
all responses are pooled with respect to this characteristic,
as in the upper right portion of Table II, there is a
statistically significant difference of 0.079 sec between
the two means. The marked effect of voicing on duration
is demonstrated in Fig. 1. The baseline shows the
12 consonant environments and is arranged with stop-plosive
consonants to the left of the figure and fricatives
to the right. Cognates and nasal correlatives are shown
in the same vertical plane. Within each manner of
production category sounds are arranged from left to
right according to place of articulation along the antero-posterior
dimension of the oral cavity (e.g., the three
stop-plosives are bilabial, post-dental, and velar in
order).

Attempts to interpret the effect of voicing of the
consonant upon vowel duration have thus far been fruitless.
It may be, for example, that the voicing of a vowel
in a voiceless environment, in contrast to a voiced
environment, is withheld until the physiological vowel
“target” is more nearly approximated, and terminated
sooner in the transition to the following consonant The
problem seems to require additional experimentation
with the transition intervals, particularly of a type
employing simultaneous acoustical and physiological
measurements.

Table II also shows that the values for vowels surrounded
by stop-plosive, fricative, and nasal classes,
according to manner of production, vary over a 0.036sec
range, and demonstrate means that differ significantly.
The close similarity between the fricative and
nasal means is of interest, since nasal sounds are related
to stop-plosives physiologically and to fricatives dynamically.
The values suggest a stop-continuant dichotomy,
but as the nasal class is composed only of
voiced sounds, which have been shown to increase
the duration of contiguous vowels, this problem cannot
be analyzed definitively with these data.

Reinspection of the 12 individual environment means
in Table II indicates that, voicing constant, consonants
that differ in manner of production produce vowel
durations that usually differ significantly. Figure 1 also
shows, in both voiced and voiceless lines, the trend for
fricative sounds to prolong vowels more than do stop-plosives.
Apparently, the gradual, controlled movements
of continuant consonants favor longer vowel durations
more than do the abrupt, ballistic movements of the
stop-plosives.

The three interactions between consonant environment,
vowels and subjects were significant at the 1
percent level.

Influence upon Fundamental Frequency of Vowels

The oscillograms described above were synchronized
with the spectrograms used in the measurement of duration,
and analogous points in the vowel wave forms
nearest to the limits of the duration interval were
identified. Measurement of this distance was made in
centimeters rounded to the nearest quarter of a millimeter
and divided by the integral number of cycles
which it subtended, yielding mean period in cm. This
value, divided into the film speed, 249.3 cm/sec, 2324 gave
the mean fundamental frequency of the vowel.

The means for the different consonant environments
shown in the left column of Table III vary significantly
over a range of 7.28 cps. Comparison of these mean
values to requirements of 2.40 cps and 1.82 cps at the
1-percent and 5-percent levels, respectively, shows that
41 of the 66 possible differences exceed the minimum at
the 5-percent level, and that 35 of these exceed the 1percent
requirement.

Inspection of these means reveals that the fundamental
frequencies of vowels in voiceless environments
are invariably higher than those in voiced environments.
With the exception of [f] compared to [m] and [g],
all of these differences are significant at the 5-percent
level or better. The means are graphed in Fig. 2, the
arrangement being the same as for Fig. 1. Voiceless and
voiced groups were formed, with means as shown in the
upper right of Table III, and the difference was tested
with the expected result. The conclusion is reached that
the presence or absence of vocal-fold vibration during
consonants has a real effect upon the fundamental frequency
of adjacent vowels in the direction mentioned.

The effects of varying the characteristic place of
articulation are generally similar. A test of significance
of the lower right means in Table III allows the rejection
of the hypothesis of no difference at the 5-percent
level of confidence, but a close inspection of the variables
suggests that most of the differences probably may be
attributed to chance.

The interactions involving subjects reached statistical
significance, while the vowel-by-consonant interaction
failed to reach the 5-percent level. The relative
variation of the individual vowels in the differing consonant
environments was strikingly similar.

General study of the data presented in this section
indicates that the effects of consonants upon fundamental
frequency, although significant, are probably
less than the variations in fundamental frequency
natural to the vowels themselves when consonant environments
are constant (see Table V).

Influence upon Relative Power of Vowels

The intensity curves produced by the high speed
level recorder showed typical bimodal forms with the
second and greater mode corresponding to the stressed
syllable. For each such syllable the maximum level was
measured. The phonetic structure of the material and
the characteristics of the records allow the assumptions
that this point was reached during the production of
the vowel, that it occurred within the time interval
measured for duration and fundamental frequency, and
that it furnished valid data concerning the intensity of
the vowel. The measurements for each subject were
expressed in db above the lowest value for that subject
and in turn converted to relative power to facilitate
arithmetic treatment. 2426 These manipulations tended to
minimize variation in intensity from subject to subject.
It will be recalled that this vocal characteristic was not
controlled in the original procedure, where each subject
was permitted to establish his own general level. The
existence of individual differences in vocal output is
a well-known phenomenon and was not of interest in
this study. It should be noted further that results of
analysis of variance indicate that subject variation was
not completely obliterated.

Study of the consonant environment means in the
left column of Table IV shows them to range from 5.43
to 23.28, and a test of this variation allows the rejection
of the null hypothesis at the 1-percent level. If the 66
possible differences between these means are evaluated
against a requirement of 4.04, 42 are seen to be significant
at the 1-percent level. Further study of these
means shows that voicing of the consonant environment
was almost uniformly productive of greater mean
power. When voiceless and voiced categories are formed,
as in the right of the table, this difference is seen to be
large on the average and is significant. Statistical
evaluation of the differences between the individual
consonant means shows that all voiced environments
produced significantly greater power than all voiceless
environments with the single exception of [s]. The
nature of these differences may be visualized in the
graphs of these means in Fig. 3. This finding would be
an expected one in that the continuation of phonation
throughout the consonants as well as the vowel would
be likely to favor greater maximum intensity.

The lower right section of Table IV shows that the
effect of the place of production of adjacent consonants
is small. Although this effect is statistically significant,
since the low intensity velar environments are all voiceless,
and the post-dental environments at the other
extreme are weighted in favor of voicing, the evidence
for differences caused by variation in place of articulation
is regarded as inconclusive. Nevertheless, the semi-parallel
curves of Fig. 3 indicate similarities of consonant
effects within a given voicing class.

When relative power is considered, the interaction
between vowels and consonants reaches significance
barely at the 5-percent level. The interactions involving
subjects are significant.

Differences between Vowels

Table V presents the data for the six specific vowels
listed in order of the conventional physiological vowel
diagram. The general behavior of the three secondary
characteristics may be observed in the top row of each
section of Table V where the various environments
have been pooled, and these data are also graphed in
Fig. 4. Analyses of the variances attributable to vowels
showed them to be significant beyond the 1-percent level
for all three variables.

The group data for relative power are shown in the
lowest portion of Fig. 4. With this arrangement of
vowels, regularity of progression, as in the case of
duration and fundamental frequency, is absent. The
curve also differs in certain respects from previous
results. While the vowels [æ] and [ɑ] are here seen to
be lowest in mean power, Sacia and Beck, 1540 Black, 1341
and Fairbanks, House and Stevens 142 report reverse
findings. These vowels, furthermore, are known to have
the largest anterior diameters of the vocal conduit,
an aspect which Fairbanks 2643 has shown to be closely related
to vowel intensity. The vowels fall into two groups
of greater and lesser power, and within each group the
range is very small. While the general analysis of variance
revealed significant differences among the vowels,
comparison of the individual means within each of the
two groups of vowels mentioned yielded no statistically
significant differences.

It would appear that this atypical vowel curve, if not
resulting from chance, might be a product of the present
experimental design, which differs from those of previous
investigations. For one thing, consonant environments
were less restricted in the present experiment,
which might be important in view of the substantial
variance for consonants mentioned above. Investigation
of this factor by regrouping according to the three main
characteristics of consonant environment, however,
showed that similar curves were found under all these
conditions. Another difference from previous experiments,
although it seems an unlikely source, is that the
stimuli in this instance were bisyllabic with the syllable
studied being preceded by a common unstressed syllable.
A third difference that should be mentioned was
that all stimuli were nonmeaningful, although this
factor would appear to operate, if at all, in the opposite
direction. A more plausible factor is the phonetic symmetry
of the present syllables, which required a subject
to begin from and return to the same consonant position.
It seems reasonable to suggest that this condition might
restrict the extent of movement to the vowel position,
and that the restriction might be greatest for vowels of
normally large mouth opening such as [æ] and [ɑ].
An additional condition, involving the spelling of stimulus
items, should be mentioned. It will be seen in Table
I that the vowels [æ] and [ɑ] were represented for the
subjects by spellings of one letter while the other
vowels were spelled with two letters. Also, it would
seem likely that the spoken vowels intended by the four
two-letter spellings would, in general, be more dear to
the naive subject, and during the course of the experiment
it was observed that subjects conditioned more
swiftly to them than to the one-letter spellings. It will
be recalled that the original recordings included those
responses which necessitated repetition because of mis-articulation
of the vowels. A count of these showed that
[æ] and [ɑ] were involved approximately equally in
more than two-thirds of the total, or double the chance
expectancy. A confident explanation of the atypical
findings for relative power cannot be advanced, but
both of the latter two possibilities present problems
which are themselves worthy of investigation.

The nature of these interactions may best be appreciated
by reference to Fig. 5 which displays the values
for duration in Table V and is exemplary of the general
findings. In this three-dimensional figure, the ordinate
is duration while consonant classes and vowels are
shown along the horizontal axes. The figure is split and
spread to show both sides. The systematic variation of
vowel duration in response to changes in both consonant
environment and in vowel is clearly seen. It is
of considerable interest that the influence of neither
factor is obscured by the other, which is a finding of
obvious implication for experimental design.

Interrelationships between Acoustical
Characteristics

In the above discussion, duration, fundamental frequency,
and power have been considered separately.
In this section they are brought together for purposes
of illustrating their covariation.

Figure 6 shows the effects of voicing and manner of
production of consonants on vowels. The figure depicts
frequency-duration areas for voiced and voiceless stop-plosives
and fricatives. Each area boundary connects
the most divergent coordinate values, i.e., the maximally
varying vowel means, in each type of consonant
environment shown. The characteristic influences of the
voiced and voiceless groups upon both duration and frequency,
the contrasting effects of cognate environments,
and the distinct differences in duration between stop-plosives
and fricatives are readily apparent.

A concise illustration of the major findings of the
experiment is found in Fig. 7 which shows the means
of the pooled vowels for the 12 consonant environments.
The ordinate is frequency, the abscissa is duration, and
the diameter of the dots is proportional to relative
power. Substantial intercorrelations are evident between
all dimensions. It will be observed that vowels in
voiced environments are, in general, longer in duration,
lower in fundamental frequency, and greater in power
than are the same vowels when in voiceless environments.
Within voicing groups, clusters corresponding
to manner of production may also be found.

III. Summary and Conclusions

1. Consonant environment significantly influenced all
three acoustical characteristics of the vowels. Of the
types of consonant influences studied, the effects of
voicing were greatest. In the comparisons of voiced and
voiceless consonant environments, vowels in voiced
environments, with few exceptions, were longer in duration,
lower in fundamental frequency, and greater in
relative power.

2. Manner of production was the second most influential
consonant characteristic Its effect upon the
duration and relative power of vowels was more consistent
than upon fundamental frequency, although all
three varied significantly.

3. Place of articulation appeared to be the least important
of the consonant characteristics, but its influence
may have been obscured by the conditions of the
experiment.

4. When all consonant environments were pooled,
significant differences between vowels were found in all
three acoustical characteristics. From vowel to vowel,
duration and fundamental frequency varied in a manner
systematically related to the usual conceptions of vowel
physiology, while variations in relative power were
atypical.

5. When changes in the acoustical characteristics of
the vowels were examined in relation to variations of
both, consonant environment and vowel, the influence
of neither factor obliterated that of the other.

6. Variations of the three acoustical characteristics in
response to changing consonant environments were
substantially intercorrelated.92