Speech Comprehension Difficulties in Older Adults: Cognitive Slowing or Age-Related Changes in Hearing?

Psychology and Aging
Copyright 2005 by the American Psychological Association
2005, Vol. 20, No. 2, 261–271
0882-7974/05/$12.00
DOI: 10.1037/0882-7974.20.2.261
Speech Comprehension Difficulties in Older Adults: Cognitive Slowing or
Age-Related Changes in Hearing?
Bruce A. Schneider, Meredyth Daneman, and Dana R. Murphy
University of Toronto at Mississauga
Speech comprehension declines more rapidly in older adults than in younger adults as speech rate
increases. This effect is usually attributed to a slowing of brain function with age. Alternatively, this
Age
Speed interaction could reflect the inability of the older adult’s auditory system to cope with
speed-induced stimulus degradation. When the authors speeded speech in a way that produced minimal
degradation, both age groups were equally affected. However, when speech was speeded using other
methods, word identification declined more in older than in younger adults. Hence, auditory decline
rather than cognitive slowing may be responsible for older adults’ poorer performance in speeded
conditions.
Keywords: speed of processing, speech recognition, aging, hearing, sensory– cognitive interactions
One of the most prevalent theories in aging research is that a
comprehension declines more rapidly for older adults than for
generalized slowing in brain function with age is responsible for
younger adults, a result that is consistent with a generalized slow-
most, if not all, of the age-related declines in problem solving,
ing hypothesis (Wingfield, 1996; Wingfield et al., 1985). How-
reasoning, memory, and language (Cerella, 1990; Lindenberger &
ever, there is another possible explanation as to why older adults
Baltes, 1994; Salthouse, 1985, 1993, 1996). According to this
find it more difficult to handle rapid rates of speech. Speeding
theory, slowing in brain functioning is thought to reduce the speed
speech, in addition to increasing the rate of flow of information,
at which various cognitive operations can be performed. For ex-
also tends to degrade and/or distort the speech signal (Gordon-
ample, it is generally assumed that the reason why older adults
Salant & Fitzgibbons, 1999). Therefore, it is possible that the
often find it difficult to understand someone who is talking rapidly,
reason why older adults are more affected by speeding is that the
or fail to follow a conversation when there are multiple speakers,
auditory systems of older adults are less able to handle these
is that the rate of flow of information approaches or exceeds the
distortions than are the auditory systems of younger adults. Age-
maximum rate that can be accommodated by the cognitive pro-
related declines in hearing have been well documented (see
cesses involved in language comprehension (Wingfield, 1996;
Schneider, 1997; Schneider & Pichora-Fuller, 2000, for recent
Wingfield, Poon, Lombardi, & Lowe, 1985).
reviews), and it has been shown that such declines contribute
substantially to older adults’ poorer comprehension and memory
Cognitive Slowing or Perceptual Decline?
for connected discourse when listening is difficult (Humes, 1996;
Schneider, Daneman, Murphy, & Kwong-See, 2000).
The contribution of speed of processing in language compre-
Thus, although there is convincing evidence that speeding
hension has been studied by comparing the performance of
speech has a more deleterious effect on older than on younger
younger and older adults when speech is artificially speeded
adults, the effect could be due to age-related declines in hearing or
(Fitzgibbons & Gordon-Salant, 1996; Gordon-Salant & Fitzgib-
to a generalized slowing in the cognitive and linguistic functions
bons, 1993, 1999, 2001; Vaughan & Letowski, 1997; Wingfield,
1996; Wingfield et al., 1985), and the typical finding has been that
that are required for good speech comprehension. Because older
adults are experiencing declines in a number of auditory process-
ing abilities, the stimulus degradation produced by speeding
speech might be expected to produce delays or mistakes in pho-
Bruce A. Schneider, Meredyth Daneman, and Dana R. Murphy, Depart-
neme or word recognition that are more severe in older than in
ment of Psychology, Biological Communication Systems, University of
younger adults (Pichora-Fuller, Schneider, & Daneman, 1995;
Toronto at Mississauga, Mississauga, Ontario, Canada.
Schneider, Daneman, & Pichora-Fuller, 2002). These early (sen-
Dana R. Murphy is now at the Department of Psychology, Nipissing
University, North Bay, Ontario, Canada.
sory) problems may then cascade upward and lead to poorer
This research was supported by grants from the Canadian Institutes of
comprehension and/or memory for the information being con-
Health Research and the Natural Sciences and Engineering Research
veyed. Alternatively, it could be that the degree of auditory decline
Council of Canada. We thank Jane Carey for assistance in conducting these
with age is not severe enough to impair the reception of speeded
experiments, Brenda Hannon and Michael Gordon for help on statistical
speech but that there is a slowing in semantic and/or linguistic
analyses, and Frank Walshe, Andrea Proctor, and Leslie Garbett for help in
processing.
preparing the stimuli.
How then can we determine whether the Age
Speed interac-
Correspondence concerning this article should be addressed to Bruce A.
tion results from age-related declines in the speed of cognitive
Schneider, Department of Psychology, Biological Communication Sys-
tems, University of Toronto at Mississauga, Mississauga, Ontario L5L
processing or from age-related declines in the ability to process
1C6, Canada. E-mail: bschneid@utm.utoronto.ca
auditory signals that have been distorted by speeding? First, we
261

---

262
SCHNEIDER, DANEMAN, AND MURPHY
note that if older adults have a lower limit than younger adults for
because of its extensive use in the lab and in the clinic. Sentences,
the rate at which information can be processed, we will always
spoken by a male speaker, were presented in a background babble
observe an Age
Speed interaction. For example, if the upper rate
consisting of 12 talkers speaking simultaneously. The listeners’
limit at which younger adults can process speech information
task was to identify the last word of each sentence immediately
without error is x units of information per second, whereas the
after they heard it. Sentences were of two types: those in which the
upper limit for older adults is y units per second (x
y), then the
last word was predictable from the sentence context (e.g., “The
performance of older adults will begin to decline when the rate of
witness took a solemn oath”) and those in which the last word
information flow exceeds y, whereas the performance of younger
could not be predicted from sentential context (e.g., “John hadn’t
adults will not decline until the rate of information flow exceeds x.
thought about the oath”). There were 25 high- and 25 low-context
Thus, when speech is speeded, comprehension will begin to de-
sentences in each list. On the basis of the results of previous
cline in older adults before it does in younger adults, leading to an
studies, we selected signal-to-noise ratios (SNRs) that we expected
Age
Speed interaction.
would produce approximately 80% correct identification of the
We also note that if it is more difficult for an older adult’s
low-context words in both younger and older adults when the
deteriorating auditory system to compensate for the distortions
R-SPIN sentences were unspeeded. This meant that younger adults
introduced by speeding than it is for a younger adult’s auditory
listened to the R-SPIN sentences at an SNR that was lower than
system, comprehension will begin to decline at a lower rate of
that presented to older adults.
speed for older adults than it does for younger adults. The inability
Given that younger and older adults are equated with respect to
of an aging auditory system to handle speed-induced distortions
their ability to hear individual words in the unspeeded condition,
means that mistakes in phoneme or word recognition would begin
we would expect the performance of older adults to decline more
to occur in older adults at a lower speed than they would in
rapidly than that of younger adults when the R-SPIN sentences are
younger adults. Hence, we would expect an Age
Speed inter-
speeded if (a) there are age-related declines in cognitive processing
action if there were age-related declines in the ability to process
speed and/or (b) if older adults are more sensitive to speed-induced
speed-distorted signals.
acoustic distortion than are younger adults. To distinguish between
Because both age-related changes in cognitive processing speed
these two possibilities, we explored different methods of speeding
and age-related changes in the ability to handle speed-induced
speech to produce different degrees of speed-induced distortion.
acoustic distortion can lead to an Age
Speed interaction, it
Consider first what we would expect if older adults are more
becomes difficult to determine the source(s) of this interaction.
sensitive than younger adults to speed-induced acoustic distortion
The only conditions under which we would not observe an Age
and if their cognitive processing speed is slower than that of
Speed interaction are when there are no age-related differences in
younger adults (Hypothesis 1). We first note that we should always
cognitive processing speed and no age-related differences in the
observe an Age
Speed interaction because of cognitive slowing.
ability to compensate for the acoustic distortions resulting from
Moreover, if we speed speech in such a way that it is much more
speeding speech.
difficult for older adults to hear the speed-distorted words than it
The obvious way to approach this problem is to find older adults
is for younger adults, the Age
Speed interaction should be
with auditory systems equivalent to those of younger adults. How-
larger. If, on the other hand, we can speed speech in such a way
ever, it is very difficult, if not impossible, to do this given that
that it is equally detrimental to both younger and older adults to
there are age-related declines in virtually every auditory ability
hear the speed-distorted words, the differential effect of speed on
(Schneider, 1997). Hence, even when older adults are matched to
the two age groups should be substantially reduced. In other
younger adults with respect to hearing thresholds (e.g., Tun, 1998),
they may not be matched with respect to other auditory abilities.
words, the extent of the Age
Speed interaction should be
Often, the best that can be done in comparing performances across
modulated by the method of speeding.
age groups is to make it equally difficult for older and younger
Of course, it could be the case that the contribution of age-
adults to correctly perceive individual words when these words are
related sensory deficits to the speeded-speech effect are negligible
unsupported by context. In previous work using unspeeded mate-
and that the Age
Speed interaction is mediated primarily by
rials, we (Schneider et al., 2000) have found that, when we equate
cognitive slowing (Hypothesis 2). If this were the case, then
younger and older adults with respect to perceptual difficulty,
changing the degree of speed-induced distortion should have a
age-related differences in comprehension and memory for con-
negligible effect on the Age
Speed interaction. Nevertheless we
nected discourse tend to disappear. Hence, before speeding the
should still observe an interaction because of cognitive slowing.
speech material, we first made it equally difficult for younger and
A third possibility is that there are no differences in cognitive
older adults to hear individual words in the unspeeded condition.
processing speed between younger and older adults and that the
To equate younger and older adults with respect to perceptual
Age
Speed interaction is due solely to speed-induced acoustic
difficulty in the present experiment, we presented the speech signal
distortion (Hypothesis 3). If this hypothesis were true and one of
in a noise background whose level was adjusted to make it equally
the methods of speeding successfully eliminated age differences in
difficult for younger and older adults to identify individual words
the effects of speed-induced acoustic distortion, we would expect
in the unspeeded condition when there was no contextual support
to find younger and older adults to be equally affected by speeding.
to aid in identifying these words. The speech materials used were
In other words, we should not observe an Age
Speed interaction.
the sentences from three of the lists in the Revised Speech Per-
Finally, if there are no age differences in cognitive processing
ception In Noise (R-SPIN) Test (Bilger, Nuetzel, Rabinowitz, &
speed and no age differences in sensitivity to speed-induced acous-
Rzeczkowski, 1984). This test was chosen because it has been
tic distortion, we should not observe an Age
Speed interaction
standardized so that performance is comparable across lists and
in any of the speeded conditions (Hypothesis 4).

---

SPEECH COMPREHENSION DIFFICULTIES IN OLDER ADULTS
263
Methods of Speeding
stressed at all. Hence, when speech is speeded in a quiet back-
ground, the auditory systems of younger adults may have excess
The first method that we used to increase speed eliminated every
capacity that would allow them to handle the inevitable distortions
nth amplitude sample in a digitized version of the speech signal.1
introduced by speeding, whereas the distortions introduced by
In general, speeding speech by eliminating every nth amplitude
speeding may overwhelm the auditory systems of older adults. To
sample (a) shifts the energy into a higher frequency range (remov-
equalize performance on the baseline (unspeeded) condition, and
ing every nth amplitude value shortens the period of every sinusoid
to place younger and older adults on an equal perceptual footing,
in the speech signal, thereby translating all frequencies upward),
we tested both groups in a background babble whose level was
(b) speeds up all transitions, and (c) shortens all gaps (periods of
adjusted to produce equivalent performance on low-context sen-
silence or relative silence in the speech signal). These three con-
tences for unspeeded speech. Experiment 2 was the same as
sequences might prove to be particularly difficult for older listen-
Experiment 1 except that the condition in which the speech was
ers to handle given their loss of high-frequency sensitivity and
speeded by deleting every third amplitude sample was replaced by
their declines in temporal resolution (Fitzgibbons & Gordon-
a condition in which the speech was speeded by deleting every
Salant, 1996; Schneider & Pichora-Fuller, 2001).
third 10-ms segment of speech. In Experiment 3, each listener
In the second method for speeding speech, we speeded the rate
heard one of the lists presented at normal speed and the other two
of flow of information by segmenting the speech signal into 10-ms
lists speeded by deleting steady-state portions of the signal to
segments and eliminating every third segment. Note that this
produce speeds one and a half and two times the normal rate.
method, which is the one most frequently used in the literature
(Wingfield et al., 1985), increases speed without a frequency shift
Method
but removes speech segments without regard to their informational
content. For example, if 10-ms segments are removed during a
Participants
formant glide, discontinuities are introduced into the glide, and
In each of three experiments, independent groups of 12 younger and 12
glide rate is increased. If segments are removed during a stop
older adults, whose first language was English, were tested. The older
consonant, the duration of the stop is shortened. Such removals
adults were volunteers from the local community; the younger participants
may affect which phoneme is heard.
were recruited from students and staff at the University of Toronto at
In the third method, we increased speed by the same amount
Mississauga. None of the participants had any history of hearing disorders,
without distorting the transitional information. We did this by
and none used hearing aids. All participants were paid $10 an hour for their
examining the speech signal to locate steady-state portions. These
participation and had pure-tone, air-conduction thresholds less than or
could be pauses or gaps between words or syllables, or portions of
equal to 25 dB HL (the dB elevation in the listener’s threshold above what
a steady-state vowel. Whenever possible, we removed or shortened
is normal for healthy young adults) between 0.25 and 3.00 kHz (American
pauses between words and portions of steady-state vowels. We
National Standards Institute S3.6-1989) in the right ear (exceptions are
also avoided excising material where there were transitions. How-
listed in Table 1), with interaural differences less than or equal to 15 dB at
each frequency. Although listeners with hearing in this range are usually
ever, sometimes it was necessary, after pauses and portions of
referred to as having normal hearing, the hearing of older adults is by no
steady-state vowels were removed, to shorten bursts and excise
means equivalent to that of younger adults. Older adults’ average thresh-
initial and terminal portions of transitions, especially at the higher
olds are 8 –10 dB poorer than those of younger adults for frequencies less
speeds. In this way, transitions were by and large preserved, and
than or equal to 2 kHz. For frequencies greater than 2 kHz, threshold
there were no frequency shifts. Gordon-Salant and Fitzgibbons
differences increased and differed by as much as 40 dB at the highest
(1999) have argued that there is convergent evidence that “older
frequency tested (8 kHz). This audiometric pattern is typical of older adults
listeners have difficulty following the rapidly changing acoustic
whose hearing is considered to be clinically normal. In general, their
elements in a speech sequence” (p. 301), and they have shown that
thresholds are significantly increased at higher frequencies, and they are
older adults find it especially difficult to deal with selective time
suffering from a number of anomalies with respect to temporal processing
compression of consonants (Gordon-Salant & Fitzgibbons, 2001).
(Fitzgibbons & Gordon-Salant, 1996; Schneider & Pichora-Fuller, 2001;
Wingfield, 1996; Wingfield et al., 1985). The average age, years of formal
By leaving the transitions relatively intact while speeding the
education, Mill Hill Vocabulary scores (which assessed general language
speech, we thought to minimize the effects of speeding on the
functioning; Raven, 1965), and number of exceptions to the hearing criteria
aging auditory system.
of each of the participant groups are given in Table 1.
In Experiment 1, each listener heard one of the lists presented at
normal speed, one that was speeded to one and a half times the
Apparatus and Stimuli
normal rate by deleting every third amplitude value and one that
was speeded to one and a half times the normal rate by deleting
The unspeeded stimuli were digitized versions (sampling rate
20 kHz)
steady-state portions of the signal. On the basis of previous data
of Lists 1, 3, and 5 of the revised R-SPIN Test (Bilger et al., 1984). The
(Pichora-Fuller et al., 1995), we selected SNRs of 3 dB for
average speech rate across the three lists was 216.4 words per minute. Five
younger adults and 8 dB for older adults to equalize the perfor-
different versions of each list were created. Version 1 was the unadulter-
mance of the younger and older participants at a value that was
ated list. In Version 2, each R-SPIN sentence was speeded by removing
significantly less than perfect performance for the low-context
every third amplitude sample (time compression ratio of 33%, speech rate
of 324.6 words per minute). To accomplish this, we first extracted digitized
items in the unspeeded condition. Under quiet conditions, and at
normal speech rates, both younger and older adults with good
hearing perform at or near ceiling. However, if there are subclin-
1 This method of speeding has not been used to study how age affects the
ical age-related declines in auditory processing, the auditory sys-
recognition of speeded speech. It is introduced here to illustrate how the
tems of older adults may be functioning at or near capacity
differential sensitivity of younger and older adults to speed-induced dis-
whereas the auditory systems of younger adults are not being
tortions affects their ability to recognize words.

---

264
SCHNEIDER, DANEMAN, AND MURPHY
Table 1
Mean Age, Years of Education, Mill Hill Vocabulary Scores, and Number of Participants Who
Had Hearing Levels at One Frequency Only That Were Greater Than 25 dB HL But Less Than
or Equal to 35 dB HL in the Right Ear
Mill Hill
Age
Education
vocabulary scores
Hearing
criterion
Experiment
M
Range
M
SD
M
SD
exceptions
1
Young
20.67
19–23
15.42
1.16
14.50
1.45
0
Young SNR
8a
21.00
20–23
16.17
1.27
12.92
2.43
0
Old
71.17
65–79
14.33
2.50
15.17
2.12
3
2
Young
22.00
20–24
16.67
1.87
13.50
1.83
1
Old
68.25
65–74
14.17
4.32
14.25
2.30
1
3
Young
20.67
19–23
15.25
1.76
14.08
1.78
1
Old
72.25
65–84
13.50*
2.32
16.17**
2.04
1
Note.
HL
the dB elevation in the listener’s threshold above what is normal for healthy young adults.
a The control group of young participants in Experiment 1 tested at a signal-to-noise ratio (SNR) of
8.
* significant age difference at p
.05.
** significant age difference at p
.01.
sentences from the list. Then every third amplitude value from the digitized
Technologies (Alachua, FL) A-D converter under the control of a personal
sentences was discarded. Before the shortened sentence was replaced in the
computer and presented to the right ear over a Telephonics (Huntington,
digitized list, zero amplitude values were added to the beginning and end
NY) TDH-49 headphone to the participant who was seated in a double-
of the sentence (zero padding by equal amounts at both ends) to restore the
walled sound-attenuating chamber. The lists were presented at a level that
extracted segment to its original length. The speeded extract was then
was 50 dB above each participant’s threshold for detecting speech babble.
reinserted in the R-SPIN list at the same position from which it was
(The standard procedure in the R-SPIN Test is to present the sentences 50
extracted. Note that neither the position of the speeded sentence in the list
dB above the listener’s babble threshold to ensure equal audibility across
nor the time between sentence midpoints was affected.
listeners.)
In Version 3, the digitized sentence was divided up into adjacent 10-ms
segments (10 ms
200 amplitude samples). Then every third 10-ms
Procedure
segment was deleted. After zero padding, the shortened sentences were put
back into the list. Again, neither the position of the sentence in the list nor
During a session, each participant read and signed the consent form, took
the time between sentence midpoints was affected by the shortening.
the Mill Hill Vocabulary Test, and had their audiometric thresholds deter-
In Version 4, each R-SPIN sentence was speeded by removing steady-
mined. Following this step, each person’s babble threshold was determined
state segments from each sentence. In constructing the Version 4 sentences,
using an adaptive two-interval forced-choice procedure. In this procedure,
we first displayed the sound spectrogram of an extracted sentence. The
a babble segment was presented in one of two randomly chosen intervals
technician then identified steady-state portions of the sentence. These
(the other interval was empty). The two intervals, which began 1.5 s after
typically were parts of the sentence where sound was absent or where there
the listener pressed a button, were each 1.5 s long and were separated by
was relatively little change in the sound’s spectral content. The technician
a 1.5-s silent period. Lights on the button box indicated the occurrence of
then shortened these steady-state components by removing material from
each interval, and the listener’s task was to identify the interval containing
their centers, taking care to always make the cuts at zero crossings (points
the babble segment by pressing one of two buttons. Immediate feedback
on the amplitude waveform where the amplitude switched from positive to
was provided. An adaptive staircase procedure (Levitt, 1971) was used to
negative or vice versa) to minimize transient effects. After removing a
determine babble threshold (the sound pressure level corresponding to the
segment, the technician then listened to the shortened sentence to ensure
79% point on the psychometric function). This babble threshold was used
that the removal had minimal effects on phoneme recognition. Approxi-
to determine the sound pressure level of the R-SPIN sentences.
mately the same portion of sound was removed from each part of the
Following a brief break, each participant listened to each of three lists in
sentence. In Version 4, as in Versions 1 and 2, the sentence was shortened
a different random order, with the order in which each participant experi-
by one third. After zero padding, the shortened sentences were then
enced the three lists being completely counterbalanced across participants.
reinserted into the list. Again, neither the position of the sentence in the list
In Experiment 1, one of the three lists was unspeeded (Version 1), one was
nor the time between sentence midpoints was affected by the shortening.
speeded by a factor of one and a half by deleting every third amplitude
Version 5 was constructed in the same fashion as Version 4 but with half
value (Version 2), and one was speeded by a factor of one and a half by
of the sentence removed, so that the duration of the sentence was shortened
deleting steady-state segments (Version 4). In Experiment 2, one list was
by one half (time compression ratio of 50%, speech rate
432.8 words per
unspeeded (Version 1), one was speeded by a factor of one and a half by
minute).
deleting every third 10-ms segment (Version 3), and one was speeded by
The background babble that accompanied each list was also sampled at
a factor of one and a half by deleting steady-state segments (Version 4). In
a rate of 20 kHz and was left unmodified. All versions of the list were
Experiment 3, one list was unspeeded (Version 1) and the other two were
presented at a signal-to-babble ratio of 8 dB for older adults (the standard-
speeded by factors of one and a half and two by deleting steady-state
ized procedure for the administration of R-SPIN) and 3 dB for younger
segments (Versions 4 and 5, respectively).
adults in an attempt to produce equal performance levels in the unspeeded
Participants were encouraged to take a short break between lists. Occa-
condition with respect to the low-context words. The sentences and babble
sionally, older adults asked to break the session into 2 days. When that was
were then converted to an analog signal by means of the Tucker-Davis
done, the three lists were presented in the second session. There were no

---

SPEECH COMPREHENSION DIFFICULTIES IN OLDER ADULTS
265
observable differences in performance between those participants tested on
1 day versus over 2 days in any of the experiments. Participants were asked
to identify the word immediately after hearing the sentence and specify if
the last word was predictable or not from the sentence context.
Statistical Analyses
In Experiments 1 and 2, three preplanned t tests were conducted. The
first evaluated whether we were successful in matching performance across
age groups in the unspeeded condition. The second determined whether
there was a significantly greater reduction in performance in the older than
in the younger participants due to speeding by removing every third
amplitude segment, whereas the third determined whether there was a
greater reduction in performance in the older than in the younger partici-
pants due to speeding by deleting steady-state segments. Previous experi-
ments (Gordon-Salant & Fitzgibbons, 1995, 1999, 2001; Stine & Wing-
field, 1987; Tun, 1998; Tun, Wingfield, Stine, & Mecsas, 1992; Vaughan
& Letowski, 1997; Wingfield et al., 1985; Wingfield, Tun, Koh, & Rosen,
1999) involving Age
Speed interactions indicated an average effect size2
in the intermediate range (Cohen’s f 2
0.08). Given an effect size of this
magnitude, the a priori power of the t test for interaction was estimated to
be .59 (
.05, one-tailed). Hence, given the average effect size found in
previous experiments investigating Age
Speed interactions, the likeli-
hood of detecting an age difference in the extent of the reduction due to
Figure 1.
Percentage correct identification of the final word in a sentence
speeding by eliminating every third amplitude value or by eliminating
as a function of the degree of speeding for younger (circles) and older
steady-state segments is .59. However, because Experiments 1, 2, and 3
(squares) adults. The sentences in the left panels were speeded by deleting
included three independent tests of whether older adults were more sus-
every third amplitude sample. The sentences in the right panels were
ceptible than younger adults to speeding by a factor of one and a half, the
speeded by deleting steady-state segments. The results for low- and high-
power obtained by combining the results of these three experiments for this
context sentences are shown in the top and bottom panels, respectively.
speeding condition rises to .95. Because Experiment 3 was a three-factor,
Error bars represent standard errors.
completely crossed design, with speed and context as within-subject factors
and age as a between-subjects factor, the data were subjected to a three-
factor analysis of variance (ANOVA). Assuming an effect size of 0.08
to speeding was significantly greater than the decline in perfor-
(Cohen’s f2) for the interaction between age and speed and a Type I error
mance for younger adults, t(22)
4.46, p
.001, one-tailed.
of .05, the a priori power of detecting an Age
Speed interaction is .83 in
Hence, for both low- and high-context sentences, speeding speech
this design. The probability of observing a main effect of age when
by deleting every third amplitude sample has a much more dele-
Cohen’s f2 equals 0.08 is 0.89.
terious effect on older than on younger adults.
When the speech was speeded by the same amount by removing
Experiment 1: Results and Discussion
steady-state portions of the speech signal, younger and older adults
Figure 1 (top left) shows that when the low-context sentences
were equally affected by this method of speeding for low-context
were presented at a normal rate of speed, younger adults performed
sentences (see Figure 1, top right), t(22)
0.17, p
.568,
slightly (72.33%) but not significantly worse (77.67%) than older
one-tailed. However, for high-context sentences (see Figure 1,
adults, t(22)
1.53, p
.139, two-tailed. Hence, our attempt to
bottom right), the same method of speeding seems to have a lesser
match the two age groups with respect to performance on low-
effect on older than on younger adults, t(22)
2.64, p
.015,
context items by presenting the sentences to younger adults at a
two-tailed.4 Thus, although older adults find it more difficult than
lower SNR (3 dB) than the same sentences presented to older
younger adults to process the signal when it is speeded by deleting
adults (8 dB) can be considered as roughly successful. Figure 1
every third amplitude sample for both low- and high-context
(top left) also shows that when the low-context sentences were
sentences, older adults are not especially disadvantaged vis-a`-vis
speeded by deleting every third amplitude sample, older adults
younger adults on low-context sentences when the signal is
experienced a much greater drop in performance (34 percentage
speeded by deleting steady-state portions of the signal, and they
points) than did younger adults (18 percentage points). This was
appear to be significantly less affected by speeding when there is
confirmed by a t test which showed that the decline in performance
contextual support for identifying the sentence final word. This
in older adults due to speeding was significantly greater than the
latter result, however, could be due to ceiling effects because both
decline in performance in younger adults due to speeding, t(22)
3.44, p
.001, one-tailed.3
2 For each experiment, Levin’s (1997) method was used to estimate 2.
Figure 1 (bottom left) shows that both younger and older adults
These 2 values were then converted to Cohen’s (1988) f 2 and averaged to
were correctly identifying the last word in the sentence when there
obtain an estimate of the average effect size.
was contextual support nearly 100% of the time. However, when
3 A one-tailed test was used here because the research hypothesis was
the high-context sentences were speeded by deleting every third
that the decrement in performance should be greater in older than in
amplitude sample, older adults recognized significantly fewer
younger adults.
words than younger adults. This was also confirmed by a t test
4 A two-tailed test was used in this instance because the result is in the
which showed that the decline in performance in older adults due
opposite direction to our prediction.

---

266
SCHNEIDER, DANEMAN, AND MURPHY
younger and older adults are at ceiling when high-context sen-
Figure 2 (bottom panels) also compares the performance of the
tences are presented at a normal speed. Hence, when sentences are
two young groups on high-context sentences. Note that increasing
speeded by deleting steady-state portions, a procedure that pre-
the SNR cannot increase performance in the unspeeded condition
serves pitch and formant transitions, there is no evidence that older
for high-context sentences because performance is already at ceil-
adults are more affected by speeding than younger adults.
ing. However, it can and does improve performance in the two
However, before we can conclude that the Age
Speed inter-
speeded conditions. The result is that the decline in performance
action is eliminated if we speed speech by deleting steady-state
when the sentences are speeded is smaller for both types of
portions of the sentence, we must consider a possible alternative
speeding and significantly so when speech is speeded by deleting
explanation. The fact that the performance of older adults was
steady-state segments, t(22)
4.76, p
.001, two-tailed. Nev-
slightly but not significantly better than that of younger adults in
ertheless, caution must be exercised in interpreting Age
Speed
the normal speed condition for low-context items (see Figure 1, top
interactions when one of the speed levels produces asymptotic
left) means that older adults had a greater potential range of
performance.
decline than younger adults. To see whether this factor could
account, in part, for the pattern of results in Experiment 1, we
tested an additional group of 12 younger adults (see Table 1) at an
Experiment 2: Results and Discussion
SNR of 8 dB. The performance of these two young groups (one
tested at an SNR of 3 dB, and the other tested at SNR of 8 dB) on
Experiment 2 compared the effect of speeding speech by delet-
low-context sentences is shown in Figure 2 (top panels). Increasing
ing every third 10-ms sample (the customary way that speech is
the SNR by 5 dB boosted performance of younger adults by 13
speeded in the literature) with the effect of speeding speech by
percentage points in the unspeeded low-context condition, t(22)
deleting steady-state portions of the signal. Figure 3 (top left)
4.48, p
.001, two-tailed, to the point where they were outper-
shows that younger adults performed less well (71.33%) than older
forming the older adults who were also tested at an SNR of 8 dB,
adults (82.67%) in the unspeeded condition in Experiment 2 for
t(22)
2.53, p
.018, two-tailed. However, there was no indi-
low-context sentences, t(22)
3.32, p
.003, two-tailed.
cation in younger adults that SNR affected the extent of decline for
Hence, the signal-to-noise difference between younger and older
low-context sentences when speech was speeded by deletion of
adults was not successful in equating the two age groups with
every third amplitude value, t(22)
0.66, p
.516, two-tailed, or
respect to performance on low-context sentences in the unspeeded
when it was speeded by deleting steady-state segments, t(22)
condition of Experiment 2. However, as we saw in the ancillary
0.16, p
.874, two-tailed. Hence, the relative position of older and
condition to Experiment 1 (see Figure 2), the extent of the decline
younger adults in the unspeeded condition for low-context sen-
in younger adults due to the speeding of low-context sentences
tences had no effect on the extent of decline due to speeding.
does not appear to depend on whether they are tested at an SNR of
8 dB (where they outperform older adults for low-context items) or
at an SNR of 3 dB (where older adults may outperform them).
Figure 3 (top left) shows that when the low-context sentences
were speeded by deleting every third 10-ms sample, older adults
experienced a greater drop in performance than did younger adults.
This was confirmed by a t test which showed that the decline in
performance in older adults due to speeding was significantly
greater than the decline in performance in younger adults due to
speeding, t(22)
1.99, p
.030, one-tailed. Hence, an Age
Speed interaction is observed even when the speeding method
preserves pitch. However, the size of the interaction is smaller than
that in Experiment 1 when the method of speeding also produced
an upward pitch shift.
Figure 3 (bottom left) shows that both younger and older adults
were correctly identifying the last word in the sentence when there
was contextual support nearly 100% of the time. However, when
the high-context sentences were speeded by deleting every third
10-ms segment, the performance of older and younger adults
declined by about the same amount, t(22)
1.00, p
.327,
two-tailed. Hence for low- but not for high-context sentences,
speeding speech by deleting every third 10-ms segment had a more
deleterious effect on older than on younger adults.
As in Experiment 1, when the speech was speeded by the same
Figure 2.
Percentage correct identification of the final word in a sentence
amount by removing steady-state portions of the speech signal,
as a function of the degree of speeding for younger adults tested at
younger and older adults were equally affected by this method of
signal-to-noise (S/N) ratios of 3 dB (filled circles) and 8 dB (unfilled
speeding for low-context sentences (see Figure 3, top right),
circles). The sentences in the left panels were speeded by deleting every
t(22)
1.13, p
.136, one-tailed. However, for high-context
third amplitude sample. The sentences in the right panels were speeded by
sentences (see Figure 3, bottom right), this method of speeding
deleting steady-state segments. The results for low- and high-context
sentences are shown in the top and bottom panels, respectively. Error bars
seems to affect older adults less than it does younger adults,
represent standard errors.
t(22)
2.00, p
.057, two-tailed. Again, caution must be

---

SPEECH COMPREHENSION DIFFICULTIES IN OLDER ADULTS
267
that found no effect due to age, F(1, 22)
0.53, MSE
132.23,
p
.476, a significant effect of speed, F(2, 44)
315.35, MSE
138.26, p
.001, and a significant effect of context, F(1, 22)
225.68, MSE
2049.56, p
.001. However, none of the two-way
interactions with age were significant: speed by age, F(2, 44)
0.02, MSE
138.26, p
.985; context by age, F(1, 22)
2.00,
MSE
93.16, p
.171. The Speed
Context interaction,
however, was significant, F(2, 44)
9.23, MSE
58.77, p
.0001, as was the three-way interaction between speed, age, and
context, F(2, 44)
3.98, MSE
58.77, p
.026. Because there
is no evidence in Experiment 3 of any performance differences for
low-context sentences between younger and older adults at any of
the three speeds used (see Figure 4, top), we can explore this
three-way interaction by comparing the beneficial effect of con-
textual support when the two age groups are equated with respect
to their performance on low-context items. The beneficial effect of
context is calculated by subtracting at each speed and age group
performance on low-context sentences from performance on high-
context sentences. Figure 5 shows that the beneficial effect of
context declines more rapidly as a function of speed in younger
adults than it does in older adults. A 2 (age)
3 (speed) ANOVA
Figure 3.
Percentage correct identification of the final word in a sentence
on the difference scores between high- and low-context items
as a function of the degree of speeding for younger (circles) and older
indicates a significant effect of speed, F(2, 44)
9.23, MSE
(squares) adults. The sentences in the left panels were speeded by deleting
117.54, p
.001, no age effect, F(1, 22)
2.00, MSE
186.32,
every third 10-ms segment. The sentences in the right panels were speeded
p
.171, but a significant Age
Speed interaction, F(2, 44)
by deleting steady-state segments. The results for low- and high-context
3.98, MSE
117.54, p
.026. It appears that this interaction is
sentences are shown in the top and bottom panels, respectively. Error bars
due to the fact that the difference in contextual benefit between
represent standard errors.
younger and older adults increases with speed. However, none of
the pairwise age contrasts (contextual benefit for older adults
exercised in interpreting interaction effects in the high-context
minus contextual benefit for younger adults) reached the .05 level
conditions because of the presence of ceiling effects.
of significance (Newman–Kuels test). These results confirm pre-
Experiments 1 and 2 taken together show that when speech is
vious findings (Pichora-Fuller et al., 1995) that older adults, who
speeded by deleting segments without regard to the critical fea-
tures of the speech signal, word recognition in older adults declines
more than it does in younger adults in low-context conditions.
However, when speech is speeded by deleting steady-state portions
of the signal, losses in comprehension are no more severe in older
adults than in younger adults when SNRs are adjusted so that the
performance of younger and older adults is nearly comparable in
unspeeded conditions on low-context sentences.
Experiment 3: Results and Discussion
In Experiment 3, the rate of speech was increased to two times
its normal value to see whether age differences would begin to
show up at higher speeds when speech was speeded by deleting
steady-state portions of the signal. In this experiment, performance
was compared at three speeds: normal (Version 1), one and a half
times normal (Version 4), and twice normal (Version 5). Figure 4
(top panel) shows that the performance of both younger and older
adults did not differ on low-context sentences in the unspeeded
condition (73.7% and 77.7%, respectively). Hence, testing younger
and older adults at different SNRs successfully balanced the two
groups with respect to perceptual listening stress in the unspeeded
condition, t(22)
1.27, p
.217, two-tailed. Figure 4 also
suggests that performance for the younger and older adults did not
Figure 4.
Percentage correct identification of the final word in a sentence
differ at any of the three speech rates for either low- or high-
as a function of the degree of speeding for younger (circles) and older
context sentences, although the pattern of decline with speed
(squares) adults when speech is speeded by deleting steady-state segments.
appears to be dependent on sentential context. These conclusions
The results for low- and high-context sentences are shown in the top and
were supported by a 2 (age)
2 (context)
3 (speed) ANOVA
bottom panels, respectively. Error bars represent standard errors.

---

268
SCHNEIDER, DANEMAN, AND MURPHY
However, when speech is speeded by deleting steady-state por-
tions of the signal in Experiments 1–3, losses in word recognition
are no more severe in older adults than in younger adults in
low-context sentences, and there is some evidence that older adults
are less sensitive to speeding than younger adults when listening to
high-context sentences. Recall that when we speeded speech in this
way, we attempted to remove speech segments in such a way that
cues to phoneme recognition were preserved. Hence, there were no
frequency shifts, formant transitions were left alone, and phone-
mically relevant gaps were by and large left unaltered. Thus, three
of the factors that we know have a larger perceptual impact on
older than on younger adults were essentially unaltered when the
speech material was speeded. It is interesting that there was no
evidence of an Age
Speed interaction in any of the three
experiments when speech was speeded by deleting steady-state
segments for low-context sentences.
Figure 5.
Contextual benefit as a function of speed in Experiment 3.
Evaluating the Perceptual Hypothesis
Contextual benefit is obtained by subtracting percentage correct for the
The perceptual hypothesis is based on the notion that older
low-context sentences from percentage correct for the high-context sen-
adults find it more difficult to handle speed-induced acoustic
tences. Note that the contextual benefit at the highest speed is significantly
greater for older adults (squares) than for younger adults (circles). Error
distortions than do younger adults. A corollary of this hypothesis
bars represent standard errors.
is that older and younger adults should be differentially sensitive to
the different methods of speeding because they introduce different
types of acoustic distortion. In other words, older adults should
may be characterized as being in the early stages of presbycusis,
show a different pattern of responding to the three methods of
are better able to use context than younger adults when both are
speeding than will younger adults. To determine the extent to
tested under difficult listening situations. This ability of older
which this was true, we compared the response of younger and
adults in the early stages of hearing loss to make better use of
older adults with an increase in speed by a factor of one and a half
context may be a consequence of the fact that they are forced to
across three different methods of speeding. Recall that in Experi-
depend on context more often than younger adults. Some sugges-
ment 1, speed was increased by a factor of one and a half by
tion that this might be the case comes from a study by Dubno,
deleting every third amplitude value. In Experiment 2, the same
Ahlstrom, and Horwitz (2000) who found that older adults with
increase in speed was obtained by deleting every third 10-ms
exceptionally good hearing did not benefit more from context than
segment, whereas in Experiment 3, speed was increased by the
did younger adults.
same amount by deleting steady-state segments. For each of the
participants in each of these experiments, we determined the loss
in percentage of word recognition due to speeding (percentage
General Discussion
correct at normal speed minus percentage correct at one and a half
The results of Experiments 1–3 indicate that when listening
times the normal speed).
conditions in unspeeded conditions are adjusted so that younger
To determine whether the method of speeding differentially
and older adults are approximately equally accurate in identifying
affected younger and older adults, we computed the reduction in
individual words unsupported by context, the magnitude of the
accuracy due to an increase in speech rate for each method of
Age
Speed interaction depends on the manner in which speech
speeding. The data from Experiment 1 were used to evaluate the
is speeded. When speech is speeded by deleting every third am-
reduction in accuracy occasioned by deleting every third amplitude
plitude value, the Age
Speed interaction is relatively large for
value (Method 1 speed factor
1.5). Specifically, for each par-
both low- and high-context sentences. Speeding speech by deleting
ticipant in Experiment 1, we subtracted the percentage of sentence-
every third amplitude segment both shifts the energy in the speech
final words correctly identified when the sentences were speeded
signal into a higher frequency region (all frequencies are translated
by deleting every third amplitude value from the percentage of
upward by a factor of one and a half), speeds up formant transi-
words correctly identified when speech was unspeeded. The com-
tions, and reduces the gap in stop consonants. Experiment 2 shows
parable data from Experiment 2 were used to compute the effect of
that when speech is speeded by deleting every third 10-ms seg-
speeding by deletion of every third 10-ms segment (Method 2
ment, older adults are also more affected than younger adults for
speed factor
1.5), and the data from the unspeeded and one and
low-context sentences. Speeding speech in this way, in addition to
a half times the normal speed condition of Experiment 3 were used
haphazardly shortening vowel duration and pauses between words,
to compute the effect of speeding by deletion of steady-state
will speed up formant transitions in an uneven fashion and shorten
segments (Method 3, speed factor
1.5). A 2 (age)
2 (con-
gaps in stop consonants in an inconsistent way. However, speeding
text)
3 (method of speeding) ANOVA on these speed-induced
speech by deleting every third 10-ms segment does not translate
accuracy decrements revealed a significant main effect of age, F(1,
frequencies upward. Hence, this type of speeding might be ex-
66)
12.02, MSE
150.32, p
.001 (speed-induced accuracy
pected to have less of a perceptual effect on word recognition than
decrements were larger, on average, in older than in younger
speeding by deleting every third amplitude value.
adults), a significant main effect of context, F(1, 66)
52.37,

---

SPEECH COMPREHENSION DIFFICULTIES IN OLDER ADULTS
269
MSE
89.59, p
.001 (the accuracy reduction was larger for
pronounced decrement in performance. A 2 (context)
3 (method
low-context than for high-context sentences), but no clear main
of speeding) ANOVA on the performance decrements in young
effect due to method of speeding, F(2, 66)
2.84, MSE
150.32,
participants indicated main effects due to method of speeding, F(2,
p
.066. However, there was a clearly significant interaction
33)
3.90, MSE
134.20, p
.030, and context, F(1, 33)
between age and method of speeding, F(2, 66)
6.12, MSE
11.41, MSE
76.10, p
.002, but not for the interaction between
150.32, p
.004, supporting the conclusion that the method of
these two factors, F(2, 33)
1.41, MSE
76.10, p
.258.
speeding affected the two ages differentially. Neither the interac-
Pairwise comparisons (Newman–Kuels test,
.05) showed that
tion between context and method of speeding, F(2, 66)
1.14,
the main effect of method of speeding was primarily due to a
MSE
89.59, p
.326, nor the three-way interaction between
significant difference between speeding by deleting every third
age, context, and method of speeding, F(2, 66)
1.57, MSE
amplitude segment and speeding by deleting every third 10-ms
89.59, p
.217, was significant. Hence, the differential effect of
segment. That speeding by deleting every third amplitude sample
method of speeding on younger and older adults appears to be the
produced the smallest decrement suggests that younger adults are
same for both high- and low-context sentences. Finally, there was
relatively immune to the frequency shifts induced by this method.
a significant interaction between age and context (a switch from
However, they do appear to be affected when the method of
low- to high-context sentences mitigated the effect of speeding on
speeding inconsistently speeds up formant transitions and shortens
performance more in older than in younger adults), F(1, 66)
phonemic gaps.
8.04, MSE
89.59, p
.006.
Figure 6 also shows that older adults are more susceptible to
The significant interaction between age and method of speeding
speeding when speech is speeded by deleting every third amplitude
in the overall ANOVA indicates that the method of speeding has
sample, or when it is speeded by deleting every third 10-ms
a different effect in older adults than it does in younger adults.
segment, than when it is speeded by deleting steady-state seg-
Figure 6 plots decrements in performance due to speeding sepa-
ments. A 2 (context)
3 (method of speeding) ANOVA on the
rately for younger and older adults. (In this figure, decrements in
performance decrements in older participants confirmed main ef-
performance were averaged across high- and low-context sen-
fects due to method of speeding, F(2, 33)
4.94, MSE
166.44,
tences because there was no statistical justification to support the
p
.013, and context, F(1, 33)
44.08, MSE
103.09, p
.001,
notion that the effect of method of speeding differed as a function
but not for the interaction between these two factors, F(2, 33)
of context). For younger adults, speeding by removal of every third
1.31, MSE
103.09, p
.284. Pairwise comparisons (Newman–
amplitude sample had the least effect on word recognition,
Kuels test,
.05) indicated that speeding by deleting every third
whereas speeding by deleting 10-ms segments produced the most
amplitude value and speeding by deleting every third 10-ms seg-
Figure 6.
Reductions in accuracy due to speeding by a factor of one and a half as a function of method of
speeding for younger and older adults. Error bars represent standard errors.

---

270
SCHNEIDER, DANEMAN, AND MURPHY
ment produced more of a decrement in performance than speeding
However, one cannot rule out the possibility that age-related
by deletion of steady-state segments. That speeding by deleting
cognitive differences might account for a larger portion of the
every third amplitude value proved difficult for older adults is not
speeded-speech effect in more cognitively complex listening situ-
surprising because this method of speeding translates all frequen-
ations. If, for example, younger and older adults were required to
cies upward by a factor of one and a half. In addition, speeding by
answer questions about a story they just heard (a task that is more
deleting every third 10-ms segment was nearly as difficult for
cognitively complex than identifying the last word of a sentence),
older adults to process as was speeding by deletion of every third
age-related cognitive declines might make older adults more sus-
amplitude value. Apparently, both younger and older adults find it
ceptible to speeding than younger adults even when speech is
difficult to process speeded speech when the method of speeding
speeded by deleting steady-state segments. If this were to happen,
inconsistently speeds up formant transitions and shortens phone-
then we would be forced to conclude that age-related differences in
mic gaps.
linguistic and cognitive processing were also contributing to
Why, then, do some methods of speeding differentially affect
Age
Speed interactions. The answer to this question awaits
older adults more than they do younger adults? Recent studies by
further experimentation.
Gordon-Salant and Fitzgibbons (1993, 2001) suggested that older
Finally, one could also argue that our failure to find an Age
adults find it difficult to handle the spectral distortion and time
Speed interaction when speech is speeded by deleting steady-state
compression of consonants that often occur when speech is
segments was due to a lack of power. To investigate this possibil-
speeded. Further evidence that older adults might be more sensi-
ity, we performed two additional analyses on the low-context
tive to speed-induced acoustic distortions comes from a study by
sentences where we can be sure that the ability to measure an
Wingfield et al. (1999) in which spoken passages were speeded by
Age
Speed interaction is not compromised by ceiling effects.
deleting steady-state segments without regard to their location.
First, we conducted an ANOVA across all three experiments for
After speeding these passages, these investigators also created a
sentences presented at a normal rate and for sentences speeded to
condition in which the passages were restored to their normal
one and a half times the normal rate by deleting steady-state
length by inserting pauses at syntactic boundaries. This latter
segments. A 3 (experiment)
2 (age)
2 (speed) ANOVA found
manipulation preserves speed-induced acoustic distortions but re-
no effect of experiment, F(2, 66)
1, a significant age effect, F(1,
stores the time available for cognitive processing of the material to
66)
8.48, MSE
120.69, p
.005, a significant speed effect,
what was available before the passages were speeded. Restoring
F(1, 66)
195.16, MSE
75.43, p
.001, but with none of the
processing time for the speeded materials eliminated all speeding
two- and three-way interactions approaching significance (all ps
effects for younger adults, suggesting that reduced processing time
.25). Thus, there is no evidence across experiments of any Age
was responsible for the speeded speech effect in younger adults.
Speed interaction when low-context sentences are speeded by a
However, restoring processing time for older adults only partially
method that minimizes stimulus degradation. Second, a power
reduced the effects of speeding, even when the inserted pauses
analysis for the low-context sentences5 in Experiment 3 showed
increased the time available for processing the speeded material
that the proportion of variance accounted for by the Age
Speed
beyond what was available when the unspeeded passages were
interaction was only
2
.01, with Cohen’s f for this interaction
presented. These results are consistent with the hypothesis that the
term being 0.08, an effect size that Cohen (1988) considered small
acoustic distortions induced by this method of speeding have a
( f
0.10). By way of contrast, the effect size for speed ( 2
.88,
greater effect on older adults than on younger adults.
Cohen’s f
2.67) was much larger than an effect size that Cohen
considered large ( 2
.14, Cohen’s f
0.40) and 137 times larger
Evaluating the Cognitive Hypothesis
than the effect due to the interaction betw