Key Words
acceptable noise level - background noise level - most comfortable listening level
- speech rate - temporal processing
INTRODUCTION
Auditory temporal processing has been described as the perception of temporal aspects
of a sound or the alteration of the duration of a sound within a defined time interval
([Musiek et al, 2005]). The ability to process temporal aspects of speech is critical for everyday listening
activities because it serves as the foundation for many auditory processing capabilities.
Accurate processing of temporal structures of sound allows individuals to discriminate
voicing of consonants as well as discriminate similar words.
Temporal processing abilities have been studied using different methodological approaches
and measurements such as detection of gaps or accurate perception of sound patterns.
Temporal processing abilities have also been measured using temporally distorted speech.
Speech can be temporally distorted through rate-alteration of speech (time compressed
or expanded) or the addition of reverberation. Speech perception and speech understanding
(in quiet and in noise) have been measured using rate-altered speech in younger and
older adults ([Gordon-Salant and Fitzgibbons, 1993]; Vaughn and Letowski, 1997; [Gordon-Salant and Fitzgibbons, 1999]; [2004]; [Adams and Moore, 2009]; [Adams et al, 2012]). Several studies have demonstrated that older adults struggle with understanding
rapid speech ([Wingfield et al, 1985]; [Gordon-Salant and Fitzgibbons, 1993]; [1999]; [Wingfield et al, 1994]; [Wingfield, 1996]; [Vaughan and Letwoski, 1997]; [Gordon-Salant et al, 2007]); however, slowed speech has been demonstrated to benefit older adults ([Gordon-Salant et al, 2007]; [Adams et al, 2012]). The reported difficulties older adults encounter with rapid speech may be due
to an underlying auditory temporal processing deficit or a result of age-related cognitive
slowing ([Wingfield et al, 1999]).
[Gordon-Salant and Fitzgibbons (1993)] studied speech recognition performance and the effect of altered temporal factors
in younger and older adults with and without hearing loss. To assess speech recognition,
the authors used temporally degraded speech measures [time compression (TC), reverberation,
and interrupted speech]. Independent of hearing sensitivity, significant differences
were found across groups for TC, reverberation, and interrupted speech. The older
adults performed more poorly than the younger adults on the temporally distorted speech
measures. Thus, the authors concluded that age-related factors other than hearing
loss contributed to the speech recognition difficulties found for older adults.
Further research was conducted to determine if the difficulties encountered by older
adults when listening to rapid speech were due to speed of information processing
or a decline in processing brief acoustic cues. [Gordon-Salant and Fitzgibbons (2001)] examined younger and older adults with and without hearing loss. The stimuli were
sentences, phrases, and random word strings that were either undistorted, time-compressed
uniformly by 50%, or were selectively time compressed (i.e., reduction in time of
pauses, vowels, or consonants). A significant age effect was found for time-compressed
sentences and phrases. Older listeners had the most difficulty with the uniform time-compression
condition and the selective time-compression of consonants. Significant effects of
hearing loss were also found for most of the listening conditions; however, these
findings were once again independent of the aging effects. The authors concluded that
the difficulties older listeners encounter when listening to rapid speech is due,
in part, to a decline in processing brief acoustic cues (consonants) that is common
to rapid speech.
Naturally fast speech has been shown to be more difficult to understand than digitally
time-compressed speech. [Gordon-Salant et al (2014)] investigated whether recognition of time-compressed speech could predict speech
recognition for naturally fast speech. A group of young normal hearing adults ranging
in age from 19 to 22 yr (n = 13) and a group of older normal hearing adults ranging
in age from 66 to 76 yr (n = 12) participated. Twelve test lists were recorded for
this experiment with three rates (natural normal rate, natural fast rate, and 40%
TC). The authors chose to use 40% TC because the natural fast rate was roughly the
equivalent to a 40% TC speech rate. The twelve test lists contained six high predictability
sentences and six low predictability sentences from the Revised Speech Perception
in Noise (R-SPIN) test ([Bilger et al, 1984]). Each participant was tested under a total of 12 conditions that varied by speech
rate (natural normal, natural fast, and 40% TC), environmental condition (quiet, noise),
and context (high predictability, low predictability). Regardless of group, in noise,
the natural normal speech rate produced the best scores of speech understanding, and
the naturally fast speech condition produced the poorest scores of speech understanding.
Both age groups performed similarly for the natural normal rate speech condition;
however, the younger adults had better scores of speech understanding for the 40%
TC and naturally fast speech conditions. In the quiet condition, scores of speech
understanding were better for both groups and had a similar pattern of findings compared
with the noise conditions. Because of these findings, the authors concluded that digitally
rate-altered speech may not fully represent the difficulties encountered in everyday
listening situations.
In addition to time-compressed speech, some researchers have examined speech perception
and understanding using slowed speech or time-expansion. [Adams et al (2012)] examined older adults’ (with and without hearing loss) performance on speech tasks
that were time-expanded and compressed. Quick Speech-in-Noise Test (QuickSIN; [Etymotic Research, 2001]) sentences were rate altered to be slow, average, and fast. Both older adult groups
performed poorly under the fast speech rate condition; however, when the speech rate
was slowed the normal hearing older adults showed improved speech perception in noise.
The older adults with hearing loss were not able to take advantage of additional processing
time and performed the same in the slowed speech condition as the average speech condition.
The authors concluded that regardless of hearing status, rapid speech is detrimental
to older adults and decreasing from a fast speaking rate slowing to a more normal
rate of speech is beneficial especially when listening in noise.
Conversational speech rates vary across speakers. Average speaking rates have been
documented to occur between 160 and 200 words per minute (wpm; [Yorkston and Beukelman, 1981]; [Picheny et al, 1986]) with everyday listening situations and conversations often exceeding 200 wpm ([Wingfield et al, 1985]). Because of the difficulties older adults experience with time-compressed speech,
it is reasonable that listening in an environment with rapid speech may be particularly
challenging for older adults. Research has indicated that older adults struggle with
speech understanding when the speech signal is temporally degraded, but does temporal
degradation affect the amount of background noise older adults are willing to accept?
Acceptance of background noise, or the acceptable noise level (ANL), was initially
described by [Nabelek et al (1991)]. The ANL quantifies an individual’s willingness to accept background noise while
listening to speech. The ANL is measured by introducing a primary stimulus continuous
discourse at the individual’s most comfortable listening (MCL) level and adding a
competing background noise. The listener then is asked to set their background noise
level (BNL) by adjusting the noise in an up and down manner until it is set at the
highest level they are willing to accept while listening to the primary stimulus (i.e.,
discourse). The ANL is a calculated measure and is the difference between MCL and
BNL (ANL = MCL-BNL).
[Nabelek et al (2004)] investigated speech perception in noise and the acceptance of background noise for
listeners in aided and unaided conditions. Forty-one full-time hearing aid users and
nine part-time hearing aid users (mean age of 71 yr) participated in this study. Findings
of the study revealed that full-time hearing aid users accepted more background noise
than part-time hearing aid users. The authors determined that this was due to lower
BNLs for the full time hearing aid users. The MCL was found to decrease in the aided
condition regardless of group. The BNL was found to decrease in the aided condition
with a greater decrease for full-time hearing aid users. While both MCL and BNL decreased
in the aided condition, BNL decreased more for the full-time hearing users. The authors
stated that this caused the decrease in ANL for full-time hearing aid users. The authors
found no significant difference in ANL between aided and unaided conditions. For R-SPIN
scores, listening condition had a statistically significant effect. In the unaided
condition, mean scores were 72.1% and in the aided condition mean scores were 83.1%.
No correlation was found between the ANL and R-SPIN scores. The ANL and R-SPIN scores
were found to be reliable and stable over a 3 mo period in aided and unaided conditions.
However, ANL and R-SPIN scores were not related to one another. Hearing aid gain did
not change listeners’ ANLs (i.e., aided versus unaided ANLs are not different); however,
hearing aid use (full-time user versus part-time user) impacts ANL. The authors suggested
that ANL may be measured before hearing aid fitting. In addition, the authors concluded
that ANL and speech understanding were independent of one another.
The influence of intelligibility of the primary speech signal on ANL was assessed
by [Gordon-Hickey and Moore (2008)]. Thirty young adult females with normal hearing participated. The stimulus conditions
were intelligible and unintelligible (reversed speech and unknown language). The Arizona
Travelogue ([Frye Electronics, Inc].) was used for the intelligible stimulus condition with a reversed recording of
the Arizona Travelogue and a recording of conversational Chinese serving as the unfamiliar
or unintelligible stimulus condition. The MCL and BNL measurements were then conducted
for each condition. The ANL was calculated by subtracting BNL from MCL. A repeated
measures analysis of variance (ANOVA) was performed to analyze the differences among
the MCL, BNL, and ANL. For MCL, no significant differences were found. For BNL and
ANL, significant differences were found for primary discourse type and BNL, and significant
differences were found for discourse type and ANL. Post hoc testing revealed significant
differences for intelligible and unintelligible conditions (reversed and unfamiliar)
for BNL as well as significant differences between intelligible and unintelligible
conditions for ANL. For BNL, the intelligible condition and reversed condition were
significantly different; however, the intelligible and unfamiliar conditions were
not significantly different. The researchers suggested that listeners accepted more
background noise when it was intelligible rather than unintelligible. For ANL, each
condition was significantly different from one another with the lowest ANLs found
for the intelligible condition. Overall, intelligibility of the primary discourse
did not affect MCL, but affected BNL thus resulting in a change in ANL. The authors
concluded that as speech intelligibility changes, ANL may change, further resulting
in a potential change for a patient’s predicted hearing aid success rate.
When [Nabelek et al (1991)] first described the ANL, they found that the ANL did not correlate with MCL, pure-tone
average, or age. [Nabelek et al (2006)] further confirmed these findings when they evaluated the use of ANL as a predictor
of hearing aid use with nonusers of hearing aids, part-time hearing aid users, and
full-time hearing aid users. A statistically significant, but weak, correlation for
age and aided ANL was found. The authors contributed this to the large sample size
of the study and deemed the significant correlation for age and aided ANL as clinically
insignificant. This finding has been questioned by [Walravens et al (2014)]; however, there are substantial methodologic differences between the two studies.
For this reason, the predictive value of ANL before hearing aid fitting should continue
to be evaluated.
Studies have demonstrated that ANL is a predictor of hearing aid success and is related
to hearing aid use ([Nabelek et al, 2006]). The ANL is quite variable with few known factors influencing a listener’s ANL.
Listeners with more self-control have lower ANLs, and listeners with less self-control
have higher ANLs ([Nichols and Gordon-Hickey, 2012]). The ANL is also influenced by the intelligibility of the primary stimulus ([Gordon-Hickey and Moore, 2008]). The ANL improves with spatial separation of the primary and background stimuli
([Freyaldenhoven et al, 2005]; [Ahlstrom et al, 2009]). There are conflicting reports in the literature regarding several factors. [Gordon-Hickey et al (2012)] reported that ANL is influenced by the primary talker gender and the number of background
talkers. However, [Plyler et al (2011)] reported that ANL did not differ because of the primary talker gender. While many
researchers report that ANL is not related to speech perception in noise ([Nabelek et al, 2004]; [2006]), some report that it is ([Ahlstrom et al, 2009]). There are also conflicting reports regarding the influence of hearing sensitivity
on ANL ([Nabelek et al, 1991]; [Walravens et al, 2014]). ANL is not related to age ([Nabelek et al, 1991]; [2006]; [Crowley and Nabelek, 1996]) and listener gender ([Crowley and Nabelek, 1996]; [Rogers et al, 2003]).
The purpose of the current study was to evaluate the influence of age and speech rate
on acceptance of background noise. Previous research has demonstrated that compared
with their younger adult counterparts, older adults perform more poorly on speech
understanding and speech perception tasks when increases in speech rate occur. The
specific goal of this study was to determine if the difficulty with speech understanding
experienced by older adults when speech is rate-altered would translate to a difference
in the amount of background noise this population is willing to accept. A slowed speech
rate, normal speech rate, and a fast speech rate were used to evaluate background
noise acceptance. The following questions were posed: Does listener age affect the
ANL? Does speech rate affect the ANL? Do speech rate and listener age interact to
influence the ANL?
METHODOLOGY
Participants
Forty-four adults were recruited from the University and surrounding community. Fourteen
participants did not meet inclusionary criteria because ofpoor auditory thresholds,
poor scores on the cognitive screening, or a combination of the two. Participants
included in the study were grouped by age with fifteen younger adults ranging in age
from 19 to 27 yr (mean = 21.93) and fifteen older adults ranging in age from 55 to
73 yr (mean = 61.8). Participant numbers were identified with power analysis software
(G. Power version 3.1; [Erdfelder et al, 1996]). All participants had Type A tympanograms, bilaterally. All young adult participants
were screened for normal auditory thresholds (i.e., 25 dB HL) and passed for the octave
frequencies ranging from 500 to 8000 Hz, bilaterally. The older adult participants
were also screened for normal auditory thresholds (i.e., 25 dB HL). All older adult
participants passed the hearing screening except for four participants who had hearing
loss that was within age normative data based on [Cruickshanks et al (1998)]. Participants with significant hearing loss were referred for further audiological
testing. All participants were native speakers of English and no participant had a
significant history for receptive speech or language disorder. All participants passed
the Montreal Cognitive Assessment (MoCA) with a score of 26 or better (Younger adults
mean = 29; Older adults mean = 29; MoCA; [Nasreddine et al, 2005]). Case history information was obtained before inclusion in the study. All participants
recruited for this study read and signed a Statement of Informed Consent approved
by the University of South Alabama Internal Review Board.
Apparatus
Stimulus recordings, audiometric testing, and ANL testing were completed in an Industrial
Acoustics Company double-walled sound-treated room meeting American National Standards
Institute specifications for maximum allowable ambient noise levels for audiometric
test rooms ([ANSI, 2008]). Stimulus recordings were completed using the Computerized Speech Laboratory (CSL)
Model 4300B (Kay Elemetrics Corporation, Lincoln Park, NJ) and a head-mounted microphone.
Audiometric screening and ANL testing were conducted with a computer-based audiometer
(Madsen Astera Otometrics) calibrated in accordance with [ANSI (2010)] specifications for a Type 2 audiometer. The audiometric testing was completed using
TDH-39P earphones mounted in supra-aural cushions. The ANL stimuli were delivered
in a sound field routed through the audiometer to an Insignia loudspeaker. Participants
were seated 1.5 m from the loudspeaker at 0° azimuth.
Stimuli Recordings
Primary and background stimuli recordings were created for the present study. Previous
research indicated that when only one stimulus (primary or background) speech rate
was altered, listeners easily distinguished between the primary and background stimuli
as different sound sources ([Gordon-Salant and Fitzgibbons, 2004]). Because of this, both primary and background speech stimuli were recorded for
each rate. In addition, recordings were created for the present study to achieve natural
fast speech and natural slowed speech for the connected discourse. When attempting
to digitally rate-alter the original connected discourse, distortion was created because
of the length of the passage and desired compression and expansion percentages. Therefore,
six volunteers (three female, three male) were recruited from the University and surrounding
community to serve as talkers. All talkers were native speakers of American English
and had no history of expressive speech and/or language disorder. Each talker recruited
for the present study read two different passages, one for the primary stimulus and
one for the background stimulus. For the primary stimulus recordings, each talker
read the Arizona Travelogue script (ANL CD, [Frye Electronics, Inc].). For the background stimulus recordings, each talker read a different passage
from a novel about the history of each state quarter in the United States ([Noles, 2008]).
All vocal recordings were completed using the CSL and Adobe Audition (version 1.5)
software. During the recording session, the talker was seated in front of a computer.
The talker wore a head mounted microphone with the microphone placed 5 cm from the
talker’s mouth. The recordings included a 6-secsustained vowel, /a/, and six different
one minute recordings of running discourse. The sustained vowel was recorded to complete
objective vocal analysis through the CSL’s Multi-Dimensional Voice Program-Advanced
software. The running discourse was recorded for the creation of the primary and background
stimuli and for the evaluation of each talker’s vocal characteristics. For the running
discourse recordings, the two passages (i.e., Arizona Travelogue and novel passage)
were read with three separate instructions. Each participant was given the passage
and asked to read it at a normal rate. The researcher then played an example file
of fast speech (compressed) or slowed speech (expanded). The talker was then instructed
to either slow or speed their speech rate similarly to the recording. A visual indicator
of rate was provided via a metronome displayed on a computer monitor in view of the
talker. The metronome was used to help maintain consistency of rate throughout the
recording. At the completion of each successful recording, the researcher analyzed
the speech rate in wpm to determine if the speech rate was slowed or speeded as needed.
The researcher then provided feedback to the talker and requested additional recordings
until the talker was able to approximate the goal for speeding or slowing the passage.
The goal was to slow the speech by approximately 25% and speed the speech by approximately
50% from the individual’s normal rate. For example, if the normal rate was 180 wpm
for a given talker, the slowed speech rate of 135 wpm and speeded speech rate of 270
wpm served as goals for that talker. Once each recording was completed, the researcher
then repeated the process with the talker for the remaining speech rate (i.e., slowed
or speeded) and for the novel passage in all three conditions. All talkers were provided
breaks as needed, and talkers were allowed multiple sessions to complete the requested
recordings.
The sustained vowel recordings were trimmed to include only the center 4 sec. The
trimmed and sustained vowel recordings were objectively analyzed via the CSL’s Multi-Dimensional
Voice Program-Advanced function. All talkers were within normal limits for jitter,
shimmer, noise to harmonic ratio, and fundamental frequency for their gender. Root
mean square (RMS) power for each of the discourse recordings was evaluated and adjusted
with the use of Adobe Audition software so that all the recordings had the same RMS
values.
Ten second excerpts of all primary stimulus recordings for the six talkers for each
condition (normal, slow, and fast) were recorded to compact disc. Three Speech-Language
Pathologists (SLP) served as raters for the recordings. The raters evaluated the discourse
recordings for rate, articulation, voice quality, and pitch. The Appendix contains
a copy of the rating scale used by the raters. Scores on the rating scale were totaled.
Next, the speech rate for each condition and each talker was calculated and displayed
in a tabular format. The relative percent increase or decrease in rate (e.g., 34%
compression and 32% expansion) was then calculated for each talker and added to the
table. Ratings, wpm, and compression/expansion percentages can be found in [Table 1]. The researcher then selected the primary stimulus recording based on the ratings
from the SLPs, the normal speech rate falling within normal limits, and ability of
the talker to maintain the targeted rate-alteration of 50% TC and 25% time expansion.
Talker 3 served as the primary stimulus recording. This recording was ultimately selected
because of the talker’s ability to consistently maintain the targeted rate, neutrality
of dialect, and overall quality of the vocal recordings. The talker with the lowest
rating from the SLP’s of the opposite gender of the primary talker was removed from
the stimulus pool so that the same number of female and male talkers remained for
creation of the background stimulus. On reviewing the ratings from the SLP’s, two
of the original male talker recordings were deemed unfit for use as background stimuli
because of the lack of dialect neutrality and consistency of rate-altered recordings.
Therefore, one additional male talker was recruited to serve as the final recording
for the background stimuli. The remaining four talkers’ recordings of the reading
from a novel were then used to create the background stimulus. The two remaining female
and two remaining male talker recordings from the novel reading were overlapped and
concatenated to create a five-minute background stimulus of twelve-talker babble.
This method was similar to [Kalikow et al (1977)]. Twelve-talker babble was selected for use as the background stimuli to remain consistent
with the commercially available ANL testing materials. For the slow and fast rate
recordings, the same procedure was completed by overlapping and concatenating the
recordings to create slow 12-talker babble and fast 12-talker babble recordings. The
primary stimulus recording was then concatenated to create a five-minute recording
of discourse. The relative RMS for the primary and background stimuli were then evaluated
and adjusted so that the two recordings had the same relative RMS. The stimuli were
then paired by speech rate (e.g., slow speech rate primary talker with slow speech
rate twelve-talker babble background) and recorded to compact disc.
Table 1
Primary Stimulus wpm, Time Expansion (TE), TC, and Gender for Each Talker at Slow,
Normal, and Fast Speech Rates
|
Talker
|
Slow (wpm)
|
Normal (wpm)
|
Fast (wpm)
|
TE (%)
|
TC (%)
|
Gender
|
Total Score
|
Stimulus Type
|
Notes
|
|
1
|
143
|
193
|
288
|
26
|
49
|
F
|
224
|
Background
|
Primary investigator
|
|
2
|
124
|
178
|
226
|
30
|
27
|
F
|
235
|
Background
|
Could not achieve desired TC%
|
|
3
|
130
|
187
|
288
|
30
|
54
|
F
|
222
|
Primary
|
Neutral dialect; achieved desired TC and TE%
|
|
4
|
118
|
161
|
253
|
27
|
57
|
M
|
227
|
Background
|
Difficulty reading aloud with consistency
|
|
5
|
144
|
201
|
264
|
28
|
31
|
M
|
203
|
|
Dialect was not neutral
|
|
Could not achieve desired TC%
|
|
6
|
119
|
170
|
244
|
30
|
44
|
M
|
196
|
|
Dialect was not neutral
|
|
Could not achieve desired TC%
|
|
7*
|
136
|
169
|
234
|
19
|
39
|
M
|
226
|
Background
|
Could not achieve desired TC%
|
Note: *Recorded background stimuli only.
To assess ANL at multiple speech rates, three stimulus pairings (slow primary stimulus
and slow twelve-talker babble paired, normal primary stimulus and normal twelve-talker
babble paired, and fast primary stimulus and fast twelve-talker babble paired) were
used for assessment of MCL and BNL. These stimulus pairings were at a slow speech
rate, normal speech rate, and a fast speech rate.
Procedures
Pre-experimental and experimental procedures were completed during one session lasting
approximately one hour. Participants completed a case history form that included age,
gender, and native language, as well as medical history pertaining to middle ear disease,
neurologic disorder, and speech language disorder. The cognitive screening (MoCA)
and a pure-tone audiometric screening were then completed. Experimental procedures
included measurement of MCL and BNL in three speech rate conditions (slow, normal,
and fast). Participants were provided written and verbal instructions for MCL and
BNL. For MCL, the participants were instructed to listen to the primary speech stimulus
and adjust the level in an up-and-down procedure until it was set at their desired
level. Participants used the “thumbs-up” hand signal to indicate an increase in loudness,
and the “thumbs-down” hand signal to indicate a decrease in loudness. The level of
the primary stimulus was first presented at 30 dB HL and the participants were instructed
to adjust the signal to a level that was above their MCL and then adjust the signal
to a level below their MCL. These adjustments were made using 5 dB step-sizes. For
every thumbs-up or thumbs-down signal, the investigator adjusted the level by one
step (i.e., 5 dB). Participants were then asked to adjust the level of the primary
stimulus to their MCL. For this instruction, 2 dB step-sizes were used. Three trials
were completed for each stimulus. Order of presentation for the three primary stimuli
was randomized. Because BNL measures require presentation of the primary stimulus
at the listener’s MCL, all MCL measures were completed before measurement of BNL.
For measurement of BNL, the primary stimulus was presented at the listener’s mean
MCL and the background stimulus was introduced at 30 dB HL. Participants were instructed
to adjust the level of the background stimulus to the level where they could no longer
hear the primary stimulus clearly and then adjust the level of the background stimulus
to a level where they could hear the primary stimulus clearly. These adjustments were
completed with a 5 dB step size. Participants were then instructed to adjust the level
of the background stimulus to the highest level of background noise they were able
to “put up with” or tolerate without becoming tense or tired. For the final adjustment,
the step size was reduced to 2 dB. Three trials were completed for each stimulus condition.
The order of presentation of the stimuli was randomized. For MCL and BNL, the three
trials were averaged for each condition. Mean MCL and BNL were used to calculate ANL
(MCL − BNL = ANL) for each condition.
RESULTS
Reliability of MCL and BNL trials were evaluated using Pearson product-moment correlations.
MCL correlation coefficients for each speech rate were statistically significant (p < 0.001), and r-values ranged from 0.863 to 0.933 indicating strong reliability for MCL measurements
across trials within each speech rate condition. BNL correlation coefficients for
each speech rate were significant (p < 0.001), and r-values ranged from 0.880 to 0.966 indicating strong reliability for BNL measurements
across trials within each speech rate condition. MCL and BNL correlation coefficients
for each speech rate are found in [Table 2]. Means were calculated for MCL and BNL. ANL was then calculated as MCL − BNL. Means
and standard deviations for ANLs across speech rate and group are displayed in [Figure 1].
Table 2
Correlation Coefficients for MCL and BNL to Slow, Normal, and Fast Speech Rates
|
Rate
|
MCL1-MCL2
|
MCL1- MCL3
|
MCL2-MCL3
|
BNL1-BNL2
|
BNL1-BNL3
|
BNL2-BNL3
|
|
Slow
|
0.863
|
0.872
|
0.911
|
0.880
|
0.909
|
0.966
|
|
Normal
|
0.921
|
0.879
|
0.933
|
0.955
|
0.918
|
0.961
|
|
Fast
|
0.883
|
0.895
|
0.882
|
0.963
|
0.945
|
0.962
|
Note: All correlation coefficients are significant (p < 0.05).
Fig. 1 Means and standard deviations for ANL across speech rates and group. Horizontal bars
with asterisks indicate significant differences between speech rates. Note: younger
adults (YA); older adults (OA).
A two-way mixed repeated measures ANOVA was conducted to evaluate the effect of age
and speech rate on the acceptance of background noise. The within-subjects factor
was speech rate (slow, normal, and fast), the between-subjects factor was group (younger
and older adults), and the dependent variable was ANL. Mauchly’s Test of Sphericity
was not statistically significant (p > 0.05) for speech rate and therefore sphericity was assumed for the remainder of
the statistical analyses. The main effect of speech rate [F
(2,56) = 27.625, p < 0.001] was statistically significant. The main effect of group [F
(1,28) = 0.021, p > 0.05] and the interaction effect between speech rate and group [F
(2,56) = 0.370, p > 0.05] were not statistically significant. To follow-up, the significant main effect
for speech rate and pairwise comparisons with Fisher’s least significance difference
(LSD) correction factor were conducted. Each pairwise comparison was significantly
different from one another. The ANL for the slow speech rate (M = 3.03, SD = 5.13) was significantly lower than for the normal speech rate (M = 4.50, SD = 4.74, p < 0.05) and for the fast speech rate (M = 7.03, SD = 4.55, p < 0.001). In addition, the ANL for the normal speech rate was also significantly
lower than the fast speech rate (p < 0.001). Pairwise comparisons can be found in [Table 3].
Table 3
Paired Comparisons of ANL with Slow, Normal, and Fast Speech Rates
|
Paired Comparison
|
p
|
|
Slow-normal
|
0.018
|
|
Slow-fast
|
0.000
|
|
Normal-fast
|
0.000
|
The between subjects factor of group was not significant; therefore, the data were
collapsed across group and a repeated measures ANOVA was completed with speech rate
as the factor. This follow-up repeated measures ANOVA revealed a significant main
effect of speech rate [F
(2,58) = 28.239, p < 0.001] on ANL. The pairwise comparisons for speech rate revealed that all pairings
were significantly different from one another with the same pattern as described in
the omnibus ANOVA previously. Additional statistical analyses were completed with
data collapsed across the group.
Analytical statistics were completed on MCL and BNL to determine their contribution
to the significant change in ANL due to speech rate. A one-way repeated measures ANOVA
was conducted to evaluate the effect of speech rate on MCL. The within-subjects factor
was speech rate and the dependent variable was MCL. Mauchly’s Test of Sphericity was
not violated (p > 0.05), and sphericity was assumed for the remainder of the statistical analyses.
The main effect of speech rate was statistically significant [F
(2,58) = 7.931, p < 0.001]. Pairwise comparisons using Fisher’s LSD correction factor revealed that
the MCL for the slow speech rate (M = 46.83, SD = 4.31) was significantly lower from the MCL of the normal speech rate
(M = 48.13, SD = 4.97, p < 0.05). MCL for the slow speech rate was also significantly lower from the MCL of
the fast speech rate (M = 48.97, SD = 4.49, p < 0.001). However, MCL for the normal speech rate was not significantly different
from the fast speech rate (p > 0.05). Means and standard deviations for MCL across speech rates can be found in
[Figure 2].
Fig. 2 Means and standard deviations for MCL across speech rates (group data collapsed).
Horizontal bars with asterisks indicate significant differences between speech rates.
Note: younger adults (YA); older adults (OA).
A one-way repeated measures ANOVA was conducted to evaluate the effect of speech rate
on BNL. The within-subjects factor was speech rate and the dependent variable was
BNL. Mauchly’s Test of Sphericity was not violated (p > 0.05), and sphericity was assumed for the remainder of the statistical analyses.
The main effect of speech rate was statistically significant [F
(2,58) = 4.906, p < 0.05]. Pairwise comparisons using Fisher’s LSD correction factor revealed BNL for
the slow speech rate (M = 43.77, SD = 6.32) was not significantly different from the BNL for the natural
normal speech rate (M = 43.63, SD = 6.26, p > 0.05); however, the BNL for the slow speech rate was significantly higher than
the BNL for the fast speech rate (M = 42.00, SD = 6.98). The BNL for the normal speech rate was also significantly higher
than the BNL for the fast speech rate (p < 0.05). Means and standard deviations for BNL across speech rates are displayed
in [Figure 3].
Fig. 3 Means and standard deviations for BNL across speech rates (group data collapsed).
Horizontal bars with asterisks indicate significant differences between speech rates.
Note: younger adults (YA); older adults (OA).
DISCUSSION
The purpose of the present study was to determine the effects of age and speech rate
on the acceptance of background noise. Research suggests older adults struggle with
temporally degraded speech. Older adults’ speech understanding decreases as speech
rate increases ([Gordon-Salant and Fitzgibbons, 1993]); conversely, older adults’ speech understanding increases as speech rate decreases
([Gordon-Salant et al, 2007]; [Adams et al, 2012]). Naturally fast speech has been documented to be more detrimental to speech understanding
than speech digitally time-compressed in a uniform manner ([Gordon-Salant et al, 2014]). The difficulty older adults experience with rapid speech may be due to memory
constraints ([Wingfield et al, 1994]), limited contextual cues ([Wingfield et al, 1985]; [Gordon-Salant and Fitzgibbons, 2001]), or a general decline in cognitive processes ([Wingfield, 1996]). Because of the negative impact on speech understanding for older adults when speech
is rate-altered, we were interested in how this population may adjust or change the
amount of background noise they willingly accept in a variety of speech rates (slow,
normal, and fast). The ANL was introduced in 1991 and has since been found to accurately
predict hearing aid success. While the ANL requires that the patient listen to speech
in noise, it is unrelated to the amount of speech in noise understood by the listener.
[Nabelek et al (2004)] evaluated ANL and speech understanding in noise using the R-SPIN. The R-SPIN and
ANL measures were found to be reliable over time, but not correlated with one another.
[Nabelek et al (2004)] concluded that ANL and speech understanding in noise were independent measures.
[Nabelek et al (2006)] further confirmed that R-SPIN scores and ANL were unrelated. [Mueller et al (2006)] evaluated ANL and speech understanding in noise using the Hearing in Noise Test.
The ANL and the Hearing in Noise Test were found to be uncorrelated as well. However,
more recent research indicates that ANL is affected by speech intelligibility of the
primary stimulus and intelligibility may serve as a criterion used to determine the
listener’s ANL ([Gordon-Hickey and Moore, 2008]; [Wu et al, 2014]; [Gordon-Hickey and Morlas, 2015]).
Previous work investigating the impact of rate-alteration of the speech signal on
speech understanding in noise tasks suggests that older adults are negatively impacted
by increases in speech rate and may be positively affected by decreases in speech
rate. Because changes in speech rate influence speech intelligibility, we hypothesized
that speech rate would influence a listener’s ANL, particularly for older adults.
Specifically, when slowing speech we thought that the subtle improvement in intelligibility
caused by slowing the speech may lead to a listener’s ability to cope with more background
noise (i.e., improved ANL). In addition, when speeding speech we thought that the
decreased intelligibility would decrease a listener’s ability to cope with background
noise (i.e., poorer ANL). Therefore, we anticipated older adults would accept significantly
more background noise in the slow speech rate condition and significantly less background
noise in the fast speech rate condition than their younger adult counterparts. However,
in the present study, there was no significant age effect. Younger adults and older
adults both accepted more background noise as the speech rate decreased and became
less accepting of background noise as the speech rate increased. This finding is consistent
with previous research indicating that ANL is not affected by age. Perhaps this is
due to the nature of ANL being a psychoacoustic decision that is intrinsic to each
individual. Findings of the initial ANL study indicated that the ANL did not correlate
with MCL, pure-tone average, or age ([Nabelek et al, 1991]). Later work by [Nabelek et al (2006)] confirmed that age does not influence ANL. The present study further supports these
findings.
Findings of the present study demonstrate that ANL is influenced by speech rate of
the primary stimulus. The ANL differed significantly across the three speech rates
(slow, normal, and fast). Regardless of age, when listening to a slower rate of speech,
listeners are more accepting of background noise. When in background noise, a slower
speech rate may reduce the effort and attention required for listeners to attend to
the talker. This appears to allow the listener to accept more background noise. A
fast speech rate has the opposite effect. When listening to a fast speech rate, more
effort is likely required to attentively listen to the talker. An even greater amount
of effort and attention may be required when listening to a fast speech rate in the
presence of background noise. This additional demand of effort and cognitive resources
(attention) required of the listener negatively impacts the amount of background noise
listeners are willing to accept. Increasing the speech rate may also lead to a decrease
in sound quality for the primary stimulus. In addition, this could lead to frustration
or annoyance for the listener and thus reduce the amount of background noise listeners
are willing to accept. Interestingly, [Adams et al (2010)] found ANL to be resistant to the effects of reverberation (often used as a measure
of temporal processing). In that study, acceptance of background noise was evaluated
under five stimulus conditions with varying degrees of reverberation (reverberation
time ranged from 0.4 to 2.0 sec). No significant effect of reverberation was found
for ANL, and there was no significant interaction between age and reverberation. Thus,
indicating reverberation does not impact the amount of background noise one is willing
to accept. It is unclear why the rate of the primary stimulus impacts ANL and the
reverberation of the primary stimulus does not affect ANL. Because rate-alteration
and reverberation are both methods for altering temporal aspects of the signal, findings
should be consistent. Future research should attempt to resolve this discrepancy.
To understand the changes in ANL due to speech rate, we evaluated differences in MCL
and BNL across speech rates. As with ANL, no group differences were observed; however,
consistent with ANL findings, speech rate significantly impacted MCL and BNL. Our
findings indicate that when listening to a slower rate of speech, individuals can
comfortably listen to connected discourse at lower loudness levels (i.e., lower MCL).
Although this finding reached statistical significance, it may not be clinically significant
because of the minor change (1 dB) in MCL, and thus should be interpreted with caution.
On further investigation of speech rate and BNL, we found that the slow speech rate
produced the highest BNL. With each subsequent increase in speech rate, BNL became
poorer. The fast speech rate condition produced a BNL significantly lower than the
other two conditions. This finding reached statistical significance; however, it may
not be clinically significant because of the minor change (<2 dB) in BNL. Results
of the present study indicate that individuals need a lower level of background noise
when the speech rate is fast to comfortably manage the listening situation. As the
speech rate decreases, individuals are better able to cope with the background noise
and can accept higher levels of background noise. [Gordon-Hickey and Moore (2008)] found a similar pattern of findings when they studied the influence of intelligibility
on ANL using intelligible and unintelligible speech stimuli. The results of that study
revealed that the intelligibility of the primary discourse did not affect MCL, but
affected BNL resulting in a change in ANL. The authors concluded that as speech intelligibility
changes, ANL may change, further resulting in a potential change for a patient’s hearing
aid success rate. In the present study, the subtle change in MCL combined with the
subtle changes in BNL, ultimately contributed to the overall significance across speech
rates for ANL.
Audiologists often recommend the use of communication strategies for those that are
hearing impaired and their communication partners. One such strategy is for a communication
partner to slow their speaking rate. As suggested by [Adams and Moore (2009)], even slowing from a fast speaking rate to a more normal rate may be of benefit
for hearing impaired patients. The present study provides support for this recommendation
as listeners in the present study were willing to cope with more background noise
in environments with slow and normal speech rates. Many listeners, especially older
adults, struggle with understanding speech in the presence of noise. This may be exacerbated
when another form of speech degradation, such as reverberation, is present in addition
to noise. Similarly, speech rate changes in addition to background noise produce a
difficult listening environment. To deal with listening situations where the speech
signal is degraded in two different manners (e.g., noise and fast speech rate), listeners
may move away from the noise source or have the noise source turned down. If there
are no effective methods to reduce background noise, findings from the present study
suggests that listeners may be more willing to listen if they ask the talker to slow
their speaking rate.
Future work should expand on this study and test speech understanding at varying speech
rates in addition to ANL. A speech understanding task may provide further clarification
on how the listeners determined their ANL for each speech rate and the relationship
of speech understanding in noise and ANL. In the present study, participants were
younger and older adults with normal hearing. Four participants in the older adult
group had some degree of hearing loss, but this was considered normal for their age
based on the [Cruickshanks et al (1998)] data. Future work should control more for hearing sensitivity and add a hearing
impaired group.
In summary, the findings of the current study suggest that ANL is influenced by the
primary talker’s speech rate. Specifically, when speech is altered to a rapid rate
(e.g., 288 wpm), listeners, regardless of age, accept significantly less background
noise than if the speech is of a normal rate (e.g., 186 wpm) or a slow rate (e.g.,
130 wpm). The current study also supports previous works indicating that the ANL is
not influenced by the age of the listener ([Nabelek et al, 1991]; [2006]).
Abbreviations
ANL:
acceptable noise level
ANOVA:
analysis of variance
BNL:
background noise level
CSL:
Computerized Speech Laboratory
MCL:
most comfortable listening level
MoCA:
Montreal Cognitive Assessment
SLP:
Speech Language Pathologist
TC:
time compression
APPENDIX
Instructions and Rating Scale for Speech Samples
Each track on this CD contains a 10 sec sample of a female or male talker reading
from a short story. There are 18 speech samples. Please place a check mark in the
box appropriate for each track. Thank you.
Track #
Which gender do you believe the talker is? Female or Male
|
Very True (5)
|
True (4)
|
Neutral (3)
|
False (2)
|
Very False (1)
|
|
The words spoken are intelligible and precisely articulated
|
|
|
|
|
|
|
The talker demonstrates normal voice quality (no hoarseness, breathiness, etc)
|
|
|
|
|
|
|
The talker is speaking at a rate that is not too fast or too slow
|
|
|
|
|
|
|
The talker’s speech does not differ significantly from Standard American English
|
|
|
|
|
|
|
The talker’s pitch is appropriate for his/her gender
|
|
|
|
|
|
|
All aspects of speech and vocal quality are acceptable for use as auditory stimuli
for a research task
|
|
|
|
|
|
|
Total
|
|
|
|
|
|
