Abstract
Cochlear-implant (CI) users have previously demonstrated perceptual restoration, or successful repair of noise-interrupted speech, using the interrupted sentences paradigm [Bhargava, Gaudrain, and Başkent (2014). “Top-down restoration of speech in cochlear-implant users,” Hear. Res. 309, 113–123]. The perceptual restoration effect was defined experimentally as higher speech understanding scores with noise-burst interrupted sentences compared to silent-gap interrupted sentences. For the perceptual restoration illusion to occur, it is often necessary for the masking or interrupting noise bursts to have a higher intensity than the adjacent speech signal to be perceived as a plausible masker. Thus, signal processing factors like noise reduction algorithms and automatic gain control could have a negative impact on speech repair in this population. Surprisingly, evidence that participants with cochlear implants experienced the perceptual restoration illusion was not observed across the two planned experiments. A separate experiment, which aimed to provide a close replication of previous work on perceptual restoration in CI users, also found no consistent evidence of perceptual restoration, contrasting the original study's previously reported findings. Typical speech repair of interrupted sentences was not observed in the present work's sample of CI users, and signal-processing factors did not appear to affect speech repair.
I. INTRODUCTION
Difficulty understanding speech in noisy listening environments is often a primary concern among cochlear-implant (CI) users (Zhao et al., 1997; Loizou et al., 2009; Wilson, 2017). One source of this difficulty is aspects of the device itself: CI processing provides impoverished speech signals that contain reduced spectral resolution and no temporal fine structure (Shannon et al., 1995), which can make it more difficult to separate the speech from background noise. Another source of this difficulty is biological, driven by the advanced age of many CI users and the duration of deafness prior to cochlear implantation (Dorman et al., 1998; Friesen et al., 2001; Fetterman and Domico, 2002; Nelson et al., 2003; Nelson and Jin, 2004; Loizou, 2006; Jin et al., 2013; Oxenham and Kreft, 2014; Sladen and Zappler, 2015). One approach to mitigate the “speech-in-noise problem” for CI users includes implementing various sound pre-processing algorithms as part of the device's signal processing scheme. Sound pre-processing algorithms in the CI may attempt to remove noise from incoming speech signals prior to transmitting these signals to the electrodes and/or may adjust aspects of the CI microphones and other settings to make target speech information more salient. These algorithms, which are proprietary and vary across CI manufacturers, will be referred to as “front-end preprocessing.”
CI users' speech understanding in noise generally improves with access to front-end preprocessing (Davidson et al., 2010; Gifford and Revit, 2010; Mauger et al., 2014; Wolfe et al., 2015; Rakszawski et al., 2016). What is unknown is whether front-end preprocessing reduces the use of perceptual restoration, a mechanism used by normal-hearing (NH) listeners to repair noisy, interrupted speech signals (Miller and Licklider, 1950; Warren, 1970; Verschuure and Brocaar, 1983; Bashford et al., 1992). Perceptual restoration occurs when parts of a speech signal are missing, due to being masked by or, as is often the case in laboratory investigations of this phenomenon, replaced with noise. In NH listeners, perceptual restoration creates the illusion of an intact, uninterrupted speech signal being presented in the context of the interrupting noise, or the sensation of two distinct and complete auditory objects. That is, despite the original speech information no longer being salient or present in that portion of the signal, NH listeners will “hear” the speech as still being salient or present, as their brain “fills in” the missing speech information. When an interrupting noise is removed entirely, creating a silent gap in the signal, or when noise is reduced in intensity, the restoration illusion does not occur and speech understanding is negatively affected (Başkent, 2012). Similarly, CI users' speech understanding is often disproportionately negatively affected by periodic reductions in salience or complete loss of speech signal information, whether by the addition of periodic gated masking noise or by the creation of silent-gap (SG) interruptions (Nelson and Jin, 2004; Bhargava et al., 2016). If front-end preprocessing, which attempts to reduce the salience of noise, also reduces opportunities to utilize perceptual restoration in CI users, this unintended side effect of front-end preprocessing could be one contributor to speech-in-noise difficulties among CI users. It is possible that speech-in-noise understanding could be improved further if front-end preprocessing strategies were adjusted to promote the use of perceptual restoration.
While NH listeners can achieve excellent speech understanding in the presence of intermittent noise by using perceptual restoration (Miller and Licklider, 1950; Warren, 1970; Verschuure and Brocaar, 1983; Bashford et al., 1992), they are less successful at perceptually restoring speech that has been degraded to simulate aspects of CI processing, like reduced spectral resolution (Başkent, 2012; Bhargava et al., 2014; Clarke et al., 2016; Jaekel et al., 2018). Evidence for perceptual restoration in CI users is mixed. In one study, CI users appeared to demonstrate perceptual restoration benefits [defined as significantly better speech understanding with noise-burst (NB) interrupted speech compared to SG interrupted speech], particularly when larger portions of the speech signal remain uninterrupted (Bhargava et al., 2014). In contrast, Jaekel et al. (2021), using a similar interrupted speech experimental paradigm, found no consistent evidence of perceptual restoration among CI users. One explanation for the differences between the two studies could be the impact of front-end preprocessing, which was not controlled for across the two studies.
The present study examined the extent to which front-end preprocessing diminishes perceptual restoration in CI users. Specifically, Study 1 compared SG vs NB interrupted speech understanding in CI users when adjustable front-end preprocessing schemes were enabled compared to when these schemes were disabled. The term “adjustable” here means that the CI user, in consultation with their audiologist, can decide whether these programs are enabled. Using the same interrupted sentence paradigm, Study 2 measured perceptual restoration in CI users when non-adjustable front-end preprocessing (i.e., the automatic gain control) was or was not engaged. The term “non-adjustable” here means that the CI user has no control over whether these programs are activated in response to incoming stimuli; engagement is instead controlled by the nature (specifically, the intensity) of the stimuli itself.
II. STUDY 1: FRONT-END PREPROCESSING EFFECTS ON INTERRUPTED SPEECH UNDERSTANDING
Front-end preprocessing algorithms generally appear to improve CI users' speech understanding in certain noisy conditions (Davidson et al., 2010; Gifford and Revit, 2010; Mauger et al., 2014; Gilden et al., 2015; Wolfe et al., 2015), but it is possible that these algorithms could inadvertently decrease the use or experience of perceptual restoration. The hypothesis for Study 1 was that perceptual restoration in CI users would be reduced when adjustable front-end preprocessing was enabled compared to when it was not. Since front-end preprocessing aims to remove or reduce noise, it may also remove or reduce the salience of the cue that prompts restoration, and thus could paradoxically decrease speech intelligibility in a setting with intermittent noise. By presenting SG interrupted and NB interrupted speech signals to CI users, this study isolated the effect of front-end preprocessing on perceptual restoration specifically, rather than generally analyzing how front-end preprocessing affects speech-in-noise perception. Note that for the purposes of this study, all adjustable front-end preprocessing programs were either enabled or disabled; in reality, a variety of configurations of enabled and disabled front-end preprocessing programs are possible in the CI.
A. Methods
1. Participants
Eleven bilaterally implanted adult CI users participated in this study. Demographic and case history information about participants is presented in Table I. Participants were native monolingual speakers of American English. All participants used a Cochlear-brand CI in both ears with N6 processors and had their most recent CI activated at least six months prior to testing. The age at onset of non-normal hearing was defined as the age at which the participant was no longer considered “normal hearing” in the target ear. The duration of non-normal hearing prior to implantation was defined as the difference in age between when the ear was implanted with a CI and when the participant was no longer considered “normal hearing” in the target ear. Participants reported this information (age at onset of non-normal hearing, duration of non-normal hearing) for each ear individually. This ensured adequate experience with the devices. Participants completed the Montreal Cognitive Assessment (MoCA), and earned scores of ≥22, indicating a lack of mild/moderate cognitive impairment (Nasreddine et al., 2005).
TABLE I.
Participant information for Study 1 (n = 11).
| Mean | Standard deviation | Range | |
|---|---|---|---|
| Age (years) | 62.8 | 13.4 | 32–79 |
| Age at onset of non-normal hearing (years; averaged across ears) | 26.9 | 18.0 | 2–55 |
| Duration of non-normal hearing prior to implantation (years; averaged across ears) | 25.3 | 18.9 | 0.5–59 |
| Baseline intact sentence understanding (percent words correct; averaged across “on” and “off” conditions) | 93.4 | 7.3 | 76–100 |
| Vocabulary (age-corrected standard score) | 105.6 | 12.5 | 93–134 |
| Working memory (age-corrected standard score) | 102.0 | 13.6 | 82–122 |
| Processing speed (age-corrected standard score) | 103.7 | 16.4 | 64–127 |
| Attention (age-corrected standard score) | 101.4 | 9.7 | 83–120 |
Participants also completed a battery of cognitive tests available from the NIH Toolbox, as it was of interest whether cognitive variables were associated with the use of perceptual restoration (Gershon et al., 2013). While Benard et al. (2014) found no relationship between perceptual restoration and cognitive skills (defined as scores on a composite assessment that included measures of working memory and processing speed) among NH listeners, it is possible that such a relationship could be observed among CI users. Such findings have occurred in other contexts; for example, Moberly et al. (2016) found a relationship between inhibition/concentration and sentence recognition in noise in CI users, but not in age-matched NH listeners. CI users with better working memory, processing speed, and inhibitory control may demonstrate higher performance on the perceptual restoration task, as they may be more successful at storing and processing incoming speech and inhibiting irrelevant input. Age-corrected standard scores from these tests are presented in Table I. The tests composing this battery were List Sorting Working Memory Test Age 7+, Pattern Comparison Processing Speed Test Age 7+, and Flanker Inhibitory Control and Attention Test Age 12+. Finally, participants completed the Peabody Picture Vocabulary Test, Version 4 (Dunn and Dunn, 2007), with mean age-corrected standard scores presented in Table I.
2. Stimuli
Stimuli were 320 declarative sentences (created by the authors), which contained 5–12 words with a range of speech sounds and semantic content. Examples of these sentences include “She cried because her seashell was broken” and “The volcano erupted after hundreds of years.” The sentences were recorded by a female speaker with a Standard American English dialect. Twenty sentences were “intact” (i.e., uninterrupted by silent gaps or noise) and were used to measure baseline speech understanding performance. Fifty percent of the remaining sentences (n = 150) were interrupted with silent gaps by applying a 5-Hz periodic nominally square wave with an 80% duty cycle to the signal, with 1-ms raised cosine on/off ramps. Duty cycle, or the amount of intact speech in a sentence, has been shown to affect restoration in CI users (Bhargava et al., 2014). For the present study, the 80% duty cycle was chosen based on pilot testing with four adult CI users. On average, this duty cycle produced a perceptual restoration effect—an improvement in performance with noise bursts compared to silent gaps—of approximately 6% with IEEE sentences (Rothauser et al., 1969) across the four pilot participants. The 80% duty cycle means that with 5-Hz interruptions, within each 200-ms speech segment, the first 160 ms remained intact and the subsequent 40 ms was replaced with a silent gap. The remaining sentences (n = 150) were interrupted with steady-state speech-spectrum-shaped noise bursts instead of silent gaps. The average level of the remaining speech was 55 dB SPL, and the average level of the interrupting noise bursts was 65 dB SPL, resulting in a –10-dB SNR.
3. Equipment
Stimuli were presented through two modes: one mode with several front-end preprocessing features enabled (“on” condition) and one mode with these same front-end preprocessing features disabled (“off” condition). Participants wore two research N6 CI sound processors through which these front-end preprocessing features could be enabled or disabled. Each participant's own clinical maps were uploaded to the research processors prior to the experiment. Since the participants were bilateral CI users, they were tested with two CIs simultaneously, as listening bilaterally was their typical mode for listening to speech.
In the on condition, the following algorithms were enabled and could potentially affect the incoming interrupted speech signal: SCAN (scene analysis), SNR-NR (signal-to-noise ratio noise reduction), WNR (wind noise reduction), ADRO (adaptive dynamic range optimization), and ASC (autosensitivity control). SCAN is a scene analyzer that can adapt microphone directionality; it automatically detects information about the listening environment and initiates specific programs for enhancing speech understanding based on the selected scene (e.g., Speech, Music, or Speech in Noise). These programs can change microphone settings and the scene classifier is updated about once per second. However, SCAN's classifications do not always properly identify the sound environment presented (Mauger et al., 2014), potentially introducing variability across and within listeners regarding how noise is processed during restoration. SNR-NR and WNR are both noise-reduction algorithms that aim to reduce the level of background noise and irrelevant sounds, which could affect how the interrupted noise bursts are perceived. The SNR-NR program detects which frequency channels in the CI contain continuous noise, for example, noise produced by a generator or car's engine. The channels below a certain SNR are then attenuated. This attenuation process, and potential delays with which the algorithm is applied, could lead to over-attenuation of several channels during and/or after the presentation of a noise burst, thereby reducing restoration. The WNR program detects decorrelation of sound signals recorded between a single sound processor's microphones, the presence of which is potentially indicative of turbulence from wind. This program likely has little effect during the restoration task given the stimuli that were used, though it is technically a noise-reduction algorithm. ADRO and ASC were expected to create compression effects (Gilden et al., 2015; Wolfe et al., 2015). Compression could change the relationship between speech and noise signals during noise burst conditions, perhaps by changing the effective SNR. Previous work has found that a negative SNR is typically needed for the perceptual restoration effect to occur (Başkent, 2012), so changes in the effective SNR could reduce the perception of the illusion. That said, Bhargava et al. (2014) did not find evidence that SNR was a factor in successful perceptual restoration in CI users – their participants succeeded in showing restoration across negative, equal, and positive SNRs. ADRO functions by slowly changing the channel gains based on input from the listening environment which could, over time, impact the effective SNR. ASC is one adjustable loop of the tri-loop automatic gain control (AGC) strategy available in Cochlear-brand CIs, and functions by attempting to reduce signal intensity. Other aspects of AGC and additional ways that AGC could influence restoration will be described in depth in Study 2. Three participants used all of these front-end preprocessing strategies with their everyday programs in both ears; four participants used all of these front-end preprocessing strategies with their everyday programs in one ear and only a subset of these strategies in the second ear; and four participants used only a subset of these strategies during everyday use in both ears.
In the off condition, SCAN, SNR-NR, WNR, ADRO, and ASC were manually turned off by the experimenter. These changes in listening mode, which involved changing front-end preprocessing strategy settings from the CI users' typical programming settings, were expected to have minimal negative impact on uninterrupted speech understanding in quiet, unlike, for example, a change to a frequency-to-electrode allocation, which would necessitate training and increased time for adaptation (Fu and Shannon, 1999).
4. Procedure
Participants completed the MoCA, cognitive battery, vocabulary test, and restoration experiment during a single visit. For the experiment, participants were seated in a soundproof booth (Industrial Acoustics, Inc., Bronx, NY), 1 m away from a pair of loudspeakers located at ±45° from the seated participant. Loudspeakers were calibrated prior to testing using speech-weighted noise at 55 dB SPL. The experimenter sat in the booth with the participant during testing and controlled the presentation of sentences.
Order of listening mode (on vs off) was counterbalanced across participants. The first ten sentences in each listening mode were intact, containing no interruptions, and served as a baseline measure of speech understanding. All other sentences (n = 300) were randomly assigned to listening mode and interruption type (silent gap vs noise burst) for each participant. Therefore, any sentence had an equal chance of being presented in either of the two listening modes and with either of the two interruption types.
On each trial, one sentence was presented, and participants reported aloud what they heard. Responses were recorded on a voice recorder and graded in real time in terms of number of words correct per sentence. Grading was lenient, similar to the approach used in Jaekel et al. (2018), meaning that incorrect verb conjugations were accepted. To ensure accuracy, a second individual separately graded a subset of responses (n = 8 participants) using the voice recordings. Inter-rater reliability was 88.1% based on the number of sentences with agreed-upon scores. Inconsistencies were resolved by averaging the scores of the two graders for that specific trial.
B. Results
The impact of the availability of front-end preprocessing on baseline intact speech understanding was minimal: on average, intact speech understanding scores differed 4.2% across conditions (on: 95.5% words correct; off: 91.3% words correct). Interrupted speech understanding results are presented in Fig. 1(A). Contrary to expectations, performance with SG interrupted sentences was always better than performance with NB interrupted sentences, both at the mean level and the individual level. This was further confirmed by investigating perceptual restoration effects for each participant, calculated by subtracting performance with SG interruptions from performance with NB interruptions. No individual participant showed positive restoration benefits either in the on or off conditions [Fig. 1(B)].
FIG. 1.
(A) Average percent words correct from Study 1. Individual data are presented with open circles and mean data are presented with filled circles. Error bars indicate ±1 standard error. Performance with SG interrupted sentences and performance with NB interrupted sentences are presented for each front-end preprocessing condition. “Off” indicates that front-end preprocessing algorithms were disabled, while “On” indicates that these algorithms were enabled. (B) Perceptual restoration effect results from Study 1. Individual data are presented with open circles and mean data are presented with filled circles. Error bars indicate ±1 standard error.
A multilevel model was used to analyze the data. The analysis was conducted in R/R-Studio (Version 4.0.0) using the buildmer package. This package creates a “maximal” model from the input that can still converge (Barr et al., 2013), then performs backward stepwise elimination based on significant changes in log-likelihood.
The dependent variable was percent words correct. The independent variables were interruption type (effect coded: −0.5 indicated SG interruptions, and +0.5 indicated NB interruptions), front-end preprocessing status (effect coded: −0.5 indicated the off condition, and +0.5 indicated the on condition), and their interaction. The covariates were standardized z-scores of age (years), age at onset of non-normal hearing (averaged across ears), duration of non-normal hearing prior to implantation (averaged across ears), non-interrupted baseline speech understanding score (averaged across on and off conditions), and age-corrected standardized z-scores for vocabulary, working memory, attention, and processing speed (Table I). All covariates were entered into the model as main effects. The random effects entered into the model were interruption type × front-end status by participant and interruption type × front-end status by sentence (i.e., item).
The results of the multilevel model are presented in Table II. Compared to performance with SG interruptions, NB interruptions significantly decreased performance (Table II: interruption type main effect, p < 0.001). Enabling front-end preprocessing strategies significantly improved performance compared to when these strategies were deactivated (Table II: front-end preprocessing status main effect, p < 0.001). The negative effect of NB interruptions on performance was lessened (i.e., performance relatively increased) when front-end preprocessing strategies were enabled (Table II: interaction term, p = 0.027). In terms of covariates, age significantly affected overall performance on the task, regardless of front-end preprocessing condition. With each one standard deviation increase in age above the sample mean (standard deviation, SD = 13.4 years; see Table I), overall performance decreased by approximately 20% (Table II: age main effect, p = 0.005); thus, older age was associated with poorer interrupted speech understanding in general. Another covariate, age at onset of non-normal hearing, appeared to have a weak but significant effect: with each standard deviation increase in age at onset of non-normal hearing (SD = 18.0 years; see Table I), overall performance increased by approximately 6% (Table II: age at onset of non-normal hearing main effect, p = 0.049). Later onsets of non-normal hearing, therefore, may be associated with better interrupted speech understanding in general.
TABLE II.
Multilevel model results for Study 1.
| Fixed effects | Estimate | Standard error | t | p |
|---|---|---|---|---|
| Intercept | 0.60 | 0.07 | 9.12 | <0.001 |
| Interruption type | –0.16 | 0.01 | –15.32 | <0.001 |
| Front-end preprocessing status | 0.06 | 0.01 | 5.55 | <0.001 |
| Age (years) | –0.20 | 0.07 | –2.84 | 0.005 |
| Age at onset of non-normal hearing | 0.14 | 0.07 | 1.96 | 0.049 |
| Interruption type × Front-end preprocessing status | 0.05 | 0.02 | 2.21 | 0.027 |
| Random effects | Variance | Standard deviation | ||
| Participant: Intercept | 0.047 | 0.217 | ||
| Item: Intercept | 0.016 | 0.126 | ||
| Residual | 0.060 | 0.244 |
Individual mean participant performance accounted for 38.3% of the variability, and individual mean performance per item accounted for 13.0% of the variability in the data. Approximately 48% of the variability in the data remained unexplained (Table II: random effects). Note that based on buildmer's model comparison algorithm, adding random slopes to the random effects did not improve model fit, and thus random slopes were not included in the final, reduced model presented in Table II.
C. Discussion
Study 1 evaluated whether front-end preprocessing strategies negatively affected perceptual restoration in CI users. Two surprising results emerged. First, no positive restoration effects (i.e., better NB interrupted sentence understanding compared to SG interrupted sentence understanding) were detected in any of the CI users tested [Fig. 1(B)]. Unlike Bhargava et al. (2014), who found that CI users could restore speech in certain conditions (i.e., experienced significant positive restoration effects), but similar to Jaekel et al. (2021), the present study found that NB interruptions acted as interferers rather than facilitators for interrupted speech understanding in the perceptual restoration paradigm. This may be unsurprising as many of the same participants (n = 9) chose to participate in both the current study and the study by Jaekel et al. (2021). On average, performance with noise bursts in the present study was 15.9% lower than with silent gaps [Fig. 1(A)]. Second, while front-end preprocessing strategies did not affect perceptual restoration of the interrupted sentences (as perceptual restoration was not observed in either the on or off conditions), NB interrupted speech understanding was most improved by access to front-end preprocessing strategies [Fig. 1(A)]. Improvement from the off to on conditions for NB interruptions was 8.0%, while improvement for SG interruptions was only 3.5%. If the hypothesis about restoration had been supported, the opposite pattern would have occurred: performance with NB interrupted speech would be negatively impacted by the presence of enabled front-end preprocessing strategies, not enhanced. In actuality, the front-end preprocessing served its intended purpose, even when speech was, for brief periods of time, completely interrupted by noise, rather than completely or partially masked by it (as would occur in conventional listening situations): it improved speech-in-noise perception. The remaining question, then, is why noise bursts, compared to silent gaps, were always harmful rather than helpful for speech understanding in these CI users. This contrasts with the commonly accepted mechanism of restoration, as well as with some previous research indicating that the mechanism can provide a benefit for speech understanding in CI users (Bhargava et al., 2014). NB interruptions should generally serve a useful rather than harmful purpose (compared to SG interruptions), particularly when there are longer durations of intact speech information between noise interruptions (Bhargava et al., 2014).
Participants who were older than the sample's average age (62.8 years) had significantly poorer interrupted speech understanding, whether interruptions were silent gaps or noise bursts. Poorer SG interrupted speech understanding among older CI users (average age = 53.8 years) and older NH listeners (average age = 66 years) has been reported previously (Saija et al., 2014; Bhargava et al., 2016). The fact that older CI users in the present study also struggled with NB interrupted speech contrasts with findings among older NH listeners. Older NH listeners tend to show greater perceptual restoration benefits—that is, significantly better understanding of NB interrupted speech compared to SG interrupted speech—with both unprocessed speech and speech that is vocoded to simulate aspects of CI processing (Saija et al., 2014; Jaekel et al., 2018). This may be part of a general pattern whereby older listeners rely to a greater extent on contextual top-down knowledge and less on the perceptual signal. Specifically, less reliance on the bottom-up auditory signal, and greater reliance on top-down information like vocabulary knowledge and sentence context, have been correlated with aging in NH listeners (Pichora-Fuller, 2008). However, Jaekel et al. (2021) reported no influence of age on NB interrupted speech understanding in CI users, even when semantic cues to the content of the interrupted sentence were provided. CI users may be more reliant on the bottom-up auditory signal in comparison, and thus are more impacted by any type of interruption to the signal.
In summary, whether front-end preprocessing strategies affect restoration in CI users remains inconclusive, as restoration (measured via the interrupted sentences paradigm) could not be detected within our sample. NB interrupted speech performance was improved by access to front-end preprocessing strategies, but not by enough to overtake performance with SG interrupted speech and reveal a perceptual restoration effect. There are several possibilities that could explain these findings: (1) front-end preprocessing strategies do not decrease NB interrupted speech understanding via noise-reduction effects and temporal envelope changes, (2) participants found the general quality of the speech signal to be improved with the front-end preprocessing strategies enabled, and thus were more able to effectively process the difficult NB interrupted speech, or (3) since many of the participants' daily programs included access to at least one of the front-end preprocessing strategies, previous experience with speech processed in this way allowed for better NB interrupted speech understanding in the on condition. A final possibility is that other aspects of CI processing outside of the controllable front-end preprocessing algorithms prohibited the ability of these CI users to restore noise-interrupted sentences. Thus, the extent to which built-in CI algorithms that cannot be modified, such as AGC, which introduces compression to the signal in response to high intensity sounds, could potentially inhibit restoration was investigated in Study 2.
III. STUDY 2: AUTOMATIC GAIN CONTROL EFFECTS ON INTERRUPTED SPEECH UNDERSTANDING
The intensity of speech stimuli, especially in relation to noise bursts, is important for the restoration effect to occur in NH listeners (Bashford et al., 1992; Başkent, 2012). In laboratory settings utilizing interrupting noise bursts (rather than masking noise) to elicit restoration effects, the interrupting noise bursts should be perceived as louder than speech, as the noise is then more able to serve as a plausible masker of the speech and promote the illusion that speech is continuing through the noise, intact and uninterrupted (Başkent, 2012). Bhargava et al. (2014) studied how noise burst intensity affects restoration in CI users at various SNRs. Speech was presented at 60 dB(A) and noise was presented at either 55, 60, 65, or 70 dB(A), which is equivalent to 5, 0, –5, or –10-dB SNR, respectively. Unlike NH listeners, who typically require negative SNRs in order to demonstrate restoration, CI users showed a similarly sized, significant restoration effect at every SNR tested in the 75% duty cycle (at the 50% duty cycle, no restoration was observed at any SNR).
One aspect of CI processing that could be affecting the results by Bhargava et al. (2014) is the role of AGC. Because the dynamic range for a CI user is smaller compared to that of a NH listener, compression is necessary to convey typical speech signals within the range from absolute audibility to the level of reportedly comfortable loudness (Khing et al., 2013), a range that is set by the CI user's audiologist. The SNRs presented by Bhargava et al. (2014) may not have been perceived as intended by the CI users as a result of dynamic compression from the AGC. Fast-acting AGC will respond quickly (<10 ms attack time) to sudden loud noises like a door slam, while slow-acting AGC will adjust the intensity of the incoming speech signals over time (Khing et al., 2013). The level at which AGC begins to compress the signal in Cochlear-brand CIs is approximately 70 dB SPL. Therefore, the most negative SNR measured in the Bhargava study (in which the majority of participants used Cochlear-brand CIs), for which the signal contained noise bursts at the 70 dB(A) level and had an SNR that would be expected to show a particularly large, prominent restoration effect, may have been impacted by compression. The compression may have reduced the perceived loudness differences between speech and noise bursts, decreasing the size of the restoration effect in this condition specifically. The ultimate effect of compression in the Bhargava study is unclear because a –10-dB SNR condition below the approximate AGC knee point of 70 dB(A) was not tested.
Besides the general presence of AGC in the processing algorithm of the CI, the speed with which the AGC turns on and off could also impact restoration (Başkent et al., 2009). On/off changes distort the amplitude envelope of speech, and resulted in reduced restoration in NH listeners presented vocoded speech (Başkent et al., 2009). Thus, the hypothesis for Study 2 was that CI users may experience a reduced restoration effect when noise bursts engage AGC due to envelope distortions and changes to the effective SNR, compared to when noise bursts are less intense and do not engage AGC.
A. Methods
1. Participants
Twelve bilaterally implanted adult CI users participated in Study 2 (11 of the 12 participants also participated in Study 1; order of study participation was randomized for each of these 11 participants, and testing for both studies occurred during the same visit). Information about these participants is presented in Table III. All participants were tested in their “functionally better ear,” which was hypothesized to have better peripheral auditory encoding, and therefore potentially better bottom-up acoustic quality—a key component for perceptual restoration (Başkent, 2012). Any potential mismatch in signal quality or encoding between ears could be avoided with this method of testing only the functionally better ear. Such “mismatches” can occur when bilateral CI users experience different etiologies or durations of hearing loss, histories of device use, device programming parameters, and/or device implantation depths across ears (Reiss et al., 2014). Controlled studies of NH listeners presented simulations of bilateral CI mismatch showed reduced ability to perceive target speech signals in masker contexts (Xu et al., 2020). Mismatch could have also affected the results of Study 1, as both ears were tested simultaneously. In Study 2, the functionally better ear was determined based on speech understanding performance with intact, uninterrupted sentences. Ten sentences were presented at a lower intensity level (55 dB SPL) and ten sentences were presented at a higher intensity level (65 dB SPL) to each ear, for a total of 40 sentences. The ear with the highest average performance across the two levels was designated the functionally better ear. For this sample, the average intact speech understanding performance in the functionally better ear was 92% words correct, with a range of 68%–100% (Table III). The functionally better ear experienced an onset of non-normal hearing at 33.9 years on average, and the average duration of non-normal hearing prior to implantation was 21.2 years (Table III).
TABLE III.
Participant information for Study 2 (n = 12).
| Mean | Standard deviation | Range | |
|---|---|---|---|
| Age (years) | 64.3 | 14.4 | 32–81 |
| Age at onset of non-normal hearing in functionally better ear (years) | 33.9 | 25.3 | 2–70 |
| Duration of non-normal hearing prior to implantation in functionally better ear (years) | 21.2 | 21.0 | 1–60 |
| Baseline intact sentence understanding in functionally better ear (percent words correct; averaged across two loudness levels) | 92.0 | 9.4 | 68–100 |
| Vocabulary (age-corrected standard score) | 105.8 | 12.5 | 93–134 |
| Working memory (age-corrected standard score) | 103.8 | 14.9 | 82–123 |
| Processing speed (age-corrected standard score) | 103.5 | 16.5 | 64–127 |
| Attention (age-corrected standard score) | 102.3 | 10.1 | 83–120 |
Participants were native monolingual speakers of American English. All participants used Cochlear-brand CIs in both ears with N6 processors and had their most recent CI activated for at least six months prior to testing. When corrected for age, most participants scored within one standard deviation of an average standard score for vocabulary and cognition (Table III).
2. Stimuli
In total, 310 sentences, created by the authors, were used as stimuli in Study 2. An additional 270 sentences were presented to nine of the twelve participants, specifically to the functionally poorer ear, but because testing was not completed by three of the 12 participants, these data will not be reported here.
Forty sentences were used for a baseline test of intact speech understanding. Each ear was presented with 20 of these 40 sentences, selected randomly without replacement: half (ten) of the sentences were presented at the lower intensity (55 dB SPL) and half (ten) of the sentences were presented at the higher intensity (65 dB SPL).
Thirty sentences were used to test control conditions. Control conditions were administered to measure whether any positive restoration effects in the experimental conditions were influenced by the absolute levels of the speech or noise, rather than (or in addition to) changes in AGC engagement. All control sentences were interrupted with noise bursts (method for interruption is described below). Fifteen control sentences had an SNR of −20 dB, with speech presented at 55 dB SPL and NB interruptions presented at 75 dB SPL. Fifteen control sentences had an SNR of 0 dB, with speech and NB interruptions presented at 65 dB SPL. Positive restoration effects were not observed in any control condition; therefore, absolute intensity values did not appear to impact restoration benefits.
The remaining sentences (n = 240) were used as experimental test sentences. Sentences were randomly assigned to one of four interruption/level conditions: (1) speech presented at 55 dB SPL with silent gaps, (2) speech presented at 55 dB SPL with 65 dB SPL noise bursts, (3) speech presented at 65 dB SPL with silent gaps, and (4) speech presented at 65 dB SPL with 75 dB SPL noise bursts. Sixty sentences were presented per condition.
The experimental conditions with speech presented at 55 dB SPL tested restoration below the AGC knee point, meaning AGC was not expected to be engaged. The experimental conditions with speech presented at 65 dB SPL tested restoration above the AGC knee point, meaning AGC was expected to more likely be engaged in the NB interruption condition and to thus change the effective SNR by compressing peaks of incoming sound signals to reduce overall levels. This compression was hypothesized to weaken the illusion of speech continuing through noise and therefore to decrease restoration.
SG and NB interruptions were applied to sentences using the same technique described in Study 1, using a 5-Hz interruption rate and an 80% duty cycle.
3. Equipment
During the task, participants wore one calibrated research N6 processor, containing the participants' clinical maps associated with that ear with all front-end preprocessing algorithms turned on, in the target ear. Participants removed the CI from the non-target ear. Participants who reported residual acoustic hearing had the ear(s) with residual hearing plugged.
4. Procedure
Participants were seated in a soundproof booth with the same setup as described in Study 1. The baseline intact speech understanding task was completed first, followed by the control conditions, followed by the experimental conditions. Assignment of sentence to condition (baseline, control, or experimental) was random for each participant. The two experimental blocks were (1) speech presented at 55 dB SPL and (2) speech presented at 65 dB SPL. The order of these blocks, each of which contained SG and NB interrupted speech signals, was randomized for each participant.
On each trial, the experimenter presented a sentence, and the participant reported the words they heard. Responses were both live-graded by the experimenter and recorded on a voice recorder for off-line grading by a different grader. A subset of responses (n = 7 participants) were graded by this different grader. Inter-rater reliability was 88.0% based on number of sentences with agreed-upon scores. Inconsistencies were resolved by averaging the scores of the two graders for that specific trial.
B. Results
At both intensity levels, participants typically performed better with SG than NB interrupted speech [Fig. 2(A)], and showed no positive restoration effect on average [Fig. 2(B)]. Across levels, 56.5% words were reported correctly for SG interrupted speech and 44.7% words were reported correctly for NB interrupted speech. Across interruption types, 50.3% words were reported correctly when speech was presented at 55 dB SPL, and 50.8% words were reported correctly when speech was presented at 65 dB SPL.
FIG. 2.
(A) Average percent words correct from Study 2. Individual data are presented with open circles and mean data are presented with filled circles. Error bars indicate ±1 standard error; because all standard errors were <2%, they are not visible on this plot. Performance with SG interrupted sentences and performance with NB interrupted sentences are presented for each level condition. (B) Perceptual restoration effect results from Study 2. Individual data are presented with open circles, and mean data are presented with filled circles. Error bars indicate ±1 standard error.
The multilevel model used to analyze the Study 2 data were constructed similarly to the model described in Study 1. The dependent variable was percent words correct per sentence, and the independent variables were interruption type (effect coded, with –0.5 indicating SG interruptions and +0.5 indicating NB interruptions), level (effect coded, with –0.5 indicating speech levels of 55 dB SPL and +0.5 indicating speech levels of 65 dB SPL), and their interaction. Covariates (all variables listed in Table III) were entered into the model as main effects. Random effects entered into the initial model were the main effects and interactions of the independent variables for both participants and items. Table IV presents the reduced, best-fitting model, following the removal of working memory as a covariate: initially, the model contained a significant negative association between working memory and performance, but upon closer inspection of the data, this relationship was found to be the result of a Simpson's paradox (Blyth, 1972) and was removed.1
TABLE IV.
Multilevel model results for Study 2.
| Fixed effects | Estimate | Standard error | t | p |
|---|---|---|---|---|
| Intercept | 0.51 | 0.06 | 9.03 | <0.001 |
| Interruption type | –0.12 | 0.02 | –5.17 | <0.001 |
| Level | 0.01 | 0.02 | 0.44 | 0.66 |
| Intact baseline speech understanding score | 0.23 | 0.05 | 4.88 | <0.001 |
| Interruption type × Level | 0.06 | 0.02 | 3.06 | 0.002 |
| Random effects | Variance | Standard deviation | ||
| Participant: Intercept | 0.038 | 0.195 | ||
| Participant: Interruption type slope | 0.005 | 0.069 | ||
| Participant: Level slope | 0.004 | 0.062 | ||
| Item: Intercept | 0.015 | 0.122 | ||
| Item: Interruption type slope | 0.021 | 0.146 | ||
| Residual | 0.049 | 0.221 |
Interrupted speech understanding was poorer with NB interruptions than for SG interruptions (Table IV; p < 0.001). Therefore, on average, no positive restoration benefit was observed for CI users. The main effect of level was not significant (p = 0.66), but the interaction of level with interruption type was significant (p = 0.002), indicating that with more intense stimuli, the difference in performance across interruption types was reduced. Specifically, with less intense stimuli, the difference across interruption types was 14.6%: performance was 43.0% with noise bursts and 57.6% with silent gaps. With more intense stimuli, the difference across interruption types was only 8.9%: performance was 46.4% with noise bursts (an improvement of 3.4%) and 55.3% with silent gaps (a decrease in 2.3%).
In terms of covariates, only intact baseline speech understanding scores were significantly associated with performance (Table IV; p < 0.001). With better baseline speech scores (i.e., above the sample's mean of 92.0%), overall performance with interrupted speech was predicted to increase: that is, better intact speech understanding was associated with better interrupted speech understanding. In terms of random effects, individual participant intercepts, individual sentence intercepts, and interruption type slopes for sentences explained the largest amount of variance. Approximately 37% of the variance was left unexplained by the model.
C. Discussion
The presence of a plausible masker, usually in the form of a noise that is louder than the surrounding speech, is required to promote the restoration illusion of speech (Bashford et al., 1992; Başkent, 2012). One form of compression in CIs, which occurs via an algorithm called AGC and is necessary due to dynamic range constraints in CI users, may generally change the intensity of noise in relation to speech and distort speech envelopes (Başkent et al., 2009; Khing et al., 2013). It was hypothesized that this form of compression could thus potentially reduce restoration when AGC turns on at the knee point level of 70 dB SPL.
Study 2 tested 12 participants to detect whether the restoration ability of interrupted sentences was reduced when speech and noise stimuli straddled the AGC knee point vs when speech and noise stimuli were presented at levels below the AGC knee point. In fact, no restoration ability (i.e., a positive restoration effect) was observed in either level condition, on average [Fig. 2(B)]. That is, speech understanding with NB interruptions was generally much poorer than with SG interruptions. Two participants technically demonstrated a positive perceptual restoration effect—one with speech at 55 dB SPL, and one with speech at 65 dB SPL—but these restoration benefits were small (≤0.5% words correct benefit).
Differences in performance with SG and NB interruptions shrank when stimuli levels straddled the AGC knee point and appeared to be driven both by a decrease in understanding SG interrupted speech and an increase in understanding NB interrupted speech. Thus, when compression was engaged, with all its potential concomitant envelope distortions, the processing of NB interrupted speech actually improved, contrary to the hypothesis. Perhaps reductions in the effective SNR due to AGC were helpful in the high intensity (65 dB SPL) condition. The influence of SNR on restoration was not observed by Bhargava et al. (2014), who presented CI users with a range of SNRs from –10 to +5 dB. While NH listeners have been shown to require negative SNRs to perceive a noise as a plausible masker and thus prompt the restoration illusion for interrupted speech, the findings from Study 2 suggest that CI users may require more favorable SNRs in order to successfully process NB interrupted speech. Furthermore, having access to more intense speech information in the 65 dB SPL condition appeared to be helpful, at least in the context of NB interruptions with more favorable SNRs (i.e., the control condition at 0-dB SNR, where both speech and noise bursts were presented at 65 dB SPL). NB interrupted speech understanding was 50.5% in this 0-dB SNR control condition, the highest percent words correct for any NB interrupted speech in the study (however, note that this control result is based on average performance with 15 sentences, while each interruption/level experimental condition result is based on average performance with 60 sentences—the results from the control conditions are thus less stable in comparison). More intense speech, compared to less intense speech, therefore may be more resilient against the distortions introduced by AGC in response to NB interruptions.
To summarize, restoration benefits in the interrupted sentences paradigm were again not observed among participants with CIs, in line with findings from Study 1 and some previous work in the field (Jaekel et al., 2021). Engaging AGC appeared to slightly but significantly improve NB interrupted speech understanding, contrary to the hypothesis that AGC engagement would introduce envelope distortions that would specifically harm noise-interrupted speech understanding.
IV. STUDY 3: REPLICATING PREVIOUS WORK ON INTERRUPTED SPEECH UNDERSTANDING IN CI USERS
Following the failure to measure perceptual restoration benefits in CI users in Studies 1 and 2 using the interrupted sentences paradigm, a “close” replication (LeBel et al., 2017) of the original study conducted by Bhargava et al. (2014) was attempted. In addition to testing the parameters used in that original study, the present study also tested other duty cycles and interruption rates to measure how these factors might influence restoration in CI users.
One interest of the authors of the Bhargava et al. (2014) study was in measuring whether CI users' difficulties with speech understanding in noisy environments were due to reduced restoration ability, with this reduction being caused by how CI processing affects bottom-up acoustic cues. The authors tested restoration in CI users as well as in NH listeners who were presented unprocessed and vocoded speech, using SG and NB interrupted sentences. The main findings of the study were that CI users could experience restoration benefits only under certain conditions and that the absence of restoration benefits in other conditions was likely due to CI processing transmitting degraded bottom-up cues. The latter was concluded because when NH listeners were presented vocoded speech, which simulates aspects of CI processing, they also failed to show restoration benefits. Another finding from the original study was that there was considerable variation in the magnitude and/or direction of the restoration effect across the CI users tested. In a more difficult condition (i.e., the 50% duty cycle condition, in which less speech information was provided between interruptions), restoration effects ranged from –10 to +20 Rationalized Arcsine Units (RAUs) in CI users. RAUs are similar to percentages except that they include a range of values extending beyond 0 and 100 to provide a more normally distributed dependent measure (Studebaker, 1985). Achieving a negative RAU score on this task indicates that NB interrupted speech was more difficult to understand than SG interrupted speech — the opposite of the expected outcome of a perceptual restoration experiment. Put another way, any CI user with a negative RAU score is not experiencing the speech understanding benefits associated with perceptual restoration. In this more difficult condition (i.e., the 50% duty cycle condition), six of 13 CI users (46%) had negative RAUs. In the easier condition (i.e., the 75% duty cycle condition, in which more speech information was provided between interruptions), restoration effects ranged from –5 to +20 RAUs in CI users, with only two of the 13 CI users (15%) having negative RAUs. The authors of the original study found that experiencing restoration was associated with higher intact, non-interrupted sentence understanding scores in quiet, at least in the more difficult listening condition (i.e., the 50% duty cycle condition). Potential reasons for this, posited by the authors of the original study, included that higher-performing CI users may be able to make better use of speech information and/or may be more sensitive to acoustic speech cues in general. In the easier condition (i.e., the 75% duty cycle condition), where nearly all CI users demonstrated restoration benefits, no significant relationship between intact sentence scores and restoration benefits was detected. Thus, it is possible that restoration benefits are only tied to speech understanding ability when the task is sufficiently difficult.
The present replication study had several goals. First, it aimed to replicate the methods used in the Bhargava study as closely as possible in order to identify restoration of interrupted sentences in CI users. Second, it aimed to vary specific parameters used in the Bhargava study to see whether they influenced restoration and could explain the differences in results between the original study and Studies 1 and 2. Third, it aimed to measure whether intact speech understanding scores were positively correlated with restoration benefits.
A. METHODS
1. Participants
Thirteen participants were recruited for the study, which is equivalent to the number tested in the original Bhargava study. Four of the 13 participants also chose to participate in Studies 1 and 2. Participant information for the original study and the present study is presented in Table V. According to the original study, the designation of “N/A” for “Age at onset of hearing loss (years)” indicated that “readings were not available in the patient record” (Bhargava et al., 2014, p. 114). As seen in Table V, the participants in the present study were older, had lower baseline intact sentence understanding scores, and had a greater range of phoneme scores than participants in the original study. These differences were not intentional and were not entirely avoidable; the participant pool to which the authors of the current study had access likely inherently had different characteristics than the original study's participant pool. An additional key difference between the present study's participants and those of the original study is that the present study's participants were native English speakers and completed the study with an English corpus. In contrast, the original study's participants were Dutch speakers and completed the study with a Dutch corpus. Finally, while participants in the original study may have used different generations of devices than participants in the present study, and thus may have had access to different front-end preprocessing strategies, results from Study 1 indicate that access to newer strategies should not result in a benefit in terms of perceptual restoration.
TABLE V.
Participant information for the original study by Bhargava et al. (2014) compared to participant information for Study 3.
| Original study (Bhargava et al., 2014) | Replication study (Study 3) | Notable differences | |
|---|---|---|---|
| Number of participants | n = 13 | n = 13 | |
| Gender | 6 Female, 7 Male | 9 Female, 4 Male | |
| Age at testing (years) | Mean = 48.5 (SD = 16.9) Range = 22–65 | Mean = 60.4 (SD = 15.3) Range = 31–79 | Replication study participants were older. |
| Age at onset of hearing loss (years) | Mean = 24.7 (SD = 24.0) Range = 0–61 6 N/As | Mean = 12.0 (SD = 17.2) Range = 0–55 0 N/As | The extent of missing data in the original study (6/13 participants) makes it difficult to conclude whether samples differed. |
| CI brand | 12 Cochlear, 1 Advanced Bionics | 12 Cochlear, 1 Advanced Bionics | |
| Intact speech scores (in RAUs) | Mean = 97.4 (SD = 11.8) Range = 78.4–117.8 | Mean = 83.2 (SD = 18.8) Range = 41.0–111.4 | Replication study participants had poorer intact speech scores. |
| Consonant-Vowel-Consonant phoneme score % (Bhargava), or Consonant-Nucleus-Consonant phoneme score % (Replication) | Mean = 84.8 (SD = 9.2) Range = 67–95 | Mean = 80.9 (SD = 11.5) Range = 51–95 | Replication study participants had a greater range of phoneme scores. |
2. Stimuli
The original study used sentences from the Vrij University corpus (Versfeld et al., 2000), which is composed of meaningful Dutch sentences. An example of one of these sentences (translated into English) is “Outside it is dark and cold.” The original study drew from 38 lists produced by a male talker, with each list containing 13 sentences. One list was used for familiarization purposes, two lists were used for measuring baseline intact sentence understanding, and 20 lists were used for measuring each of the 10 conditions (described below), with two lists randomized to each condition.
The present study used sentences from the IEEE corpus (Rothauser et al., 1969). An example of one of the sentences is “Two blue fish swam in the tank.” The sentences were recorded by a male talker, and each list contained ten sentences. One list was used for familiarization, two lists for measuring baseline intact sentence understanding, and 24 lists for measuring each of the 12 conditions (described below), with two lists randomized to each condition.
The original study's sentences were processed in the following way. SG and NB interruptions were applied to sentences using a nominally periodic square wave with an interruption rate of 1.5 Hz and raised 5-ms cosine ramps, and duty cycles of either 50% or 75%. The SNR of NB interruptions was either –10, –5, 0, or 5 dB. Thus, overall, there were ten conditions: two SG interruption conditions at the 50% and 75% duty cycle, and eight NB interruption conditions (four SNRs × two duty cycles). Speech was always presented at 60 dB SPL.
The present replication study included sentences processed identically to the original study's: SG interrupted sentences and NB interrupted sentences with a 1.5-Hz interruption rate, a duty cycle of 75%, and, for the NB interrupted sentences specifically, an SNR of –5 dB. This condition revealed a significant restoration effect for CI users in the original study (Bhargava et al., 2014). To expand upon that original study, the present study also tested several other conditions that varied interruption rates and duty cycles. In total, the following 12 conditions were created: two interruption types (SG and NB interrupted) × two interruption rates (1.5 and 5 Hz) × three duty cycles (65%, 75%, and 85%). Since every condition was tested twice, each time with a new list of ten sentences, the present experiment was composed of 240 sentence trials. Order of conditions and sentence list allocation to conditions was randomized for each participant. As no effect of SNR on CI users' ability to restore speech was detected by the original study, the present study did not vary this factor, and all SNRs were fixed at –5 dB. Speech was always presented at 60 dB SPL, identical to the original study.
3. Equipment
In the original study, participants sat in a soundbooth in front of a computer monitor and loudspeaker located at 0°. For the present study, participants were seated in front of two loudspeakers located at ±45° in a soundproof booth, and a computer monitor faced the experimenter, who controlled the experiment. Participants wore their personal devices set to everyday settings, and any ear with residual acoustic hearing was plugged.
4. Procedure
In both the original study and the present study, participants first completed the baseline speech understanding conditions (i.e., two sentence lists), then familiarization (one sentence list), then the main experiment. Familiarization consisted of participants being presented five SG and five NB interrupted sentences processed with a combination of parameters that differed from those used in the main experiment. For the present study, the familiarization sentences were interrupted at 1.5-Hz with an 80% duty cycle and –10-dB SNR. Participants listened to each sentence during familiarization and repeated aloud what was heard. Feedback during familiarization was an auditory presentation of the intact sentence coupled with the visual display of the text of the sentence, followed by the same interrupted sentence presented auditorily again. The first five sentences during familiarization were interrupted with noise bursts, and the second five sentences were interrupted with silent gaps. Following completion of familiarization, the main experiment began. Prior to each sentence list presented during the main experiment, an example sentence (not tested during the experiment) was presented, processed in the same way that the upcoming sentence list was processed. For example, if the upcoming sentence list was interrupted with silent gaps at a 5-Hz interruption rate and 65% duty cycle, the example sentence was also interrupted this way. The example sentence was always, “The horse trotted around the field at a brisk pace.” In the present study, 24 sentence lists were presented for the main experiment. Participants took breaks as needed. All testing was completed during the same visit and took approximately 1.5 h.
In the original study, for the main experiment, participants were scored on the number of words they correctly reported per list. The percent-correct scores were converted into RAUs in the original study. No feedback was provided to participants, and incorrect/absent scores were not penalized. In the replication study, sentences were graded for number of words correctly reported per list, but used lax grading that accepted as correct some changes in tense (“shop” for “shopped” but not “went” for “go”) and changes in plurality (“cats” for “cat” or “cup” for “cups”), in line with previous research practices (Stilp et al., 2010; Jaekel et al., 2018). Similar to the original study, percent-correct scores were transformed into RAUs, no feedback was provided, and incorrect answers were not penalized.
B. Results
The original study by Bhargava et al. (2014) found the following at the “replicated” condition (1.5-Hz interruption rate, 75% duty cycle, and –5 dB SNR): an approximate +4.6 RAU benefit of restoration, in that performance with NB interrupted speech was 4.6 RAUs higher than performance with SG interrupted speech. Furthermore, in the original study, CI users showed significant restoration effects at every SNR presented with a 75% duty cycle (reported by the original study's authors as an average benefit of +5.6 RAUs across all SNRs). The original study observed no association between participant variables and restoration at 75% duty cycles, but did find that higher baseline intact speech understanding scores were associated with positive restoration benefits at the 50% duty cycle. To summarize, it was expected that the present study would find restoration benefits in the replicated condition, and no significant relationship between intact speech understanding and restoration benefits; if such a relationship were observed, it could possibly indicate that the replicated condition was more challenging for our sample than it was for participants in the original study. The present study's results are described below. First, performance on the replicated condition is compared with the original study. Then, results from the other experimental conditions are discussed.
The +4.6 RAU benefit in the replicated condition, which was found in the original study, was not observed in the present study. Instead, a –0.9 RAU restoration effect was found [Fig. 3(C)]. Thus, similar average restoration effects to the original study were not found despite using the same interruption parameters and similar experimental procedures. Per paired-samples t-test, performance with the two interruption types in the present study in the replicated condition was not significantly different [t(12)=0.26, p = 0.802].
FIG. 3.
(A) Words correct (RAUs) are presented for four duty cycles: 50, 65, 75, and 85%. Data from the original study by Bhargava et al. (2014) are presented with triangles. The original study tested interrupted speech understanding at two duty cycles: 50% and 75%. Data from the present study are presented with squares (note that only the data from the 1.5-Hz interruption rate conditions are presented in this panel, this was the interruption rate used in the original study). The present study tested interrupted speech understanding at three duty cycles: 65%, 75%, and 85%. The open symbols indicate performance with silent-gap interrupted speech and filled symbols indicate performance with noise-burst interrupted speech. The filled box highlights the replicated condition between the two studies. Error bars represent standard error. Note that standard errors were <2% for the present study and are thus not visible on the plot. (B) Words correct (RAUs) are presented for the six conditions tested in the present study. The open symbols indicate performance with silent-gap interrupted speech and filled symbols indicate performance with noise-burst interrupted speech. Squares represent 1.5-Hz and circles represent 5-Hz interruption rate conditions. Error bars represent standard error; again, because standard errors were <2% for the present study, they are not visible on the plot. (C) Perceptual restoration effects (in RAUs) are presented for the six conditions tested in the present study. The three duty cycles tested are noted on the x axis. Squares represent 1.5-Hz and circles represent 5-Hz interruption rate conditions. The open symbols indicate individual data and the filled symbols indicate mean data. Error bars represent ±1 standard error.
Figure 3(A) places the accuracy results from the replicated condition in context with other conditions in the present study (i.e., the other 1.5-Hz rate interruption conditions at the 65% and 85% duty cycles) and the Bhargava study (i.e., the other 1.5-Hz rate interruption condition at 50% duty cycle). This figure shows an overall increase in speech understanding as duty cycle increases, but shows no consistent pattern in terms of the effect of interruption type. Figure 3(B) presents speech understanding results for every condition tested in the present study; again, there is a general trend of improved speech understanding as duty cycle increases, but no clear pattern in terms of the effects of interruption type or interruption rate.
In the present study, restoration effects were positive, on average, in two of the six conditions tested: the 1.5-Hz interruption rate with a 65% duty cycle (+2.4 RAUs) and the 5-Hz interruption rate with a 75% duty cycle [+2.6 RAUs; Fig. 3(C)]. The highest average accuracy was in the 5-Hz interruption rate with an 85% duty cycle and SG interruptions (64.4 RAUs), and the lowest average accuracy was in the 5-Hz interruption rate with a 65% duty cycle and NB interruptions [24.9 RAUs; Fig. 3(B)].
Accuracy was analyzed using a repeated measures analysis of variance (RM ANOVA). Note that the original study's statistical analysis could not be applied to the present study's data. The original study used Dunnett's test for multiple comparisons, as several NB conditions that varied in SNR were compared to a single SG condition. In the present study, in contrast, each NB condition had a partner SG condition against which it could be compared.
The variables of interruption type (two levels: NB and SG interruptions), interruption rate (two levels: 1.5 and 5 Hz), and duty cycle (three levels: 65, 75, and 85%) were entered as the within-subjects independent variables in the RM ANOVA. The dependent variable was accuracy in RAUs. No variables violated Mauchly's Test of Sphericity, so no corrections were needed.
The only significant main effect was duty cycle [F(2,24) = 156.8, p < 0.001, = 0.93]. Neither the main effect of interruption type was significant (p = 0.78) nor was the main effect of interruption rate (p = 0.99). No interactions were significant (p > 0.05 for all). For the main effect of duty cycle, post hoc paired-samples t-tests Bonferroni-corrected for multiple comparisons revealed that performance for every duty cycle was significantly different from every other (p < 0.001 for all). The best performance was for the 85% duty cycle, followed by the 75% duty cycle, followed by the 65% duty cycle.
To summarize, interruption type did not affect accuracy in the present study: similar scores were obtained from both SG interrupted sentences and NB interrupted sentences. Thus, no positive or negative restoration effects were observed. Because no interactions with interruption type were significant, no combination of parameters (i.e., duty cycles and interruption rates) explained interrupted speech understanding performance. Accuracy was also unaffected by interruption rate: similar scores were obtained regardless of how frequently interruption appeared (i.e., every 666.6 ms vs every 200 ms). In contrast, the use of longer duty cycles, which provided greater access to speech information across segments, significantly improved overall performance. The original study found a similar effect, specifically that performance was greater for the 75% duty cycle condition compared to the 50% duty cycle condition.
The original study found no relationship between baseline intact speech understanding scores and perceptual restoration effects in the 75% duty cycle. Figure 4 plots the original and present study's individual participant data in the replicated condition. When all data were considered from the original and present study, there appeared to be a possible relationship between baseline speech understanding scores and restoration effects in the replicated condition, in that restoration effects tend to increase (i.e., become positive) as intact speech understanding scores increase. Low-scoring CI listeners may be receiving such a degraded signal through their implants that it is difficult to tease apart speech information and noise bursts in noise-interrupted conditions. CI listeners in the mid-to-high scoring range may be most likely to experience and benefit from perceptual restoration—perhaps due to receiving comparatively higher-quality signals. Finally, it is possible that the highest scoring CI listeners are less likely to benefit from the perceptual restoration effect — achieving excellent performance with both SG and NB interrupted speech, resulting in restoration effects near zero. However, the two datasets contain little data in the high and low speech understanding score regions, making it difficult to understand the exact nature of this potential relationship between restoration effects and intact speech understanding scores. To adequately analyze the above hypothesis, additional data should be collected, particularly in the high and low regions of speech understanding performance.
FIG. 4.
Baseline intact speech understanding scores (in RAUs) are plotted against perceptual restoration effects (in RAUs) in the replicated condition for participants in the present study (filled circles) and the original Bhargava et al. (2014) study (open circles). Perceptual restoration effects were calculated for each participant by subtracting performance with silent-gap interrupted speech from performance with noise-burst interrupted speech.
C. Discussion
Study 3 had three goals: (1) provide a “close” replication of the original CI restoration study by Bhargava et al. (2014); (2) vary parameters that could impact restoration and assess their effects; and (3) measure whether intact speech understanding scores correlated with restoration benefits, which was a finding in the original study.
The original study found a +4.6 RAU restoration effect in the replicated condition. In contrast, the present study found a –0.9 RAU restoration effect in this condition. While the participants in the original study found the NB interruptions helpful for understanding speech in this condition, the present study's participants, on average, did not show this benefit. However, a range of performance was observed in the present study: six participants showed a negative restoration effect in this condition, and seven participants showed a positive restoration effect. A wide range of perceptual restoration effects was similarly observed in Jaekel et al. (2018), which used the same IEEE corpus on a restoration task with older and younger adult NH listeners presented vocoded and unprocessed speech. In that study, IEEE sentences were processed with a 2.5-Hz interruption rate and 50% duty cycle, presented at 65 dB SPL with a –5 dB SNR (i.e., it had somewhat different parameters from the present study). In the 16-channel vocoded condition, which is the fewest number of channels at which NH listeners have been shown to be able to restore vocoded speech (Clarke et al., 2016), younger adult NH listeners showed restoration effects that ranged from –25 to +31 RAUs, and older adult NH listeners ranged from –12 to +30 RAUs.
In the present study, when analyzed statistically, no significant differences in performance between SG and NB interrupted sentences were observed for the replicated condition. Thus, we could not reproduce the findings reported by Bhargava et al. (2014). This lack of restoration benefit in the interrupted sentence paradigm aligns with findings from Studies 1 and 2, where participants similarly showed no advantage with NB interruptions and instead tended to show higher performance for SG interrupted speech, the opposite of the restoration effect.
The second goal of the present study was to measure how varying the parameters of duty cycle and interruption rate affected restoration in CI users. The analysis revealed no effects of either interruption type or interruption rate; the only significant effect was a main effect of duty cycle, which indicated that increasing the amounts of speech available to the listener within each speech segment resulted in better performance overall. As duty cycle did not interact with interruption type, this usefulness of access to greater amounts of speech information did not increase restoration.
The third goal was to measure whether baseline sentence understanding scores were associated with restoration benefits. The addition of the present study's data appeared to put the original study's data into greater context, in that baseline speech understanding score may correlate with the extent to which a participant can perceptually restore speech interrupted at a 75% duty cycle (Fig. 4). Note that the original study found a significant relationship between intact speech scores and perceptual restoration benefits in the 50% duty cycle, but not in the 75% duty cycle (Bhargava et al., 2014). In the original study, the 50% duty cycle condition was considered to be challenging for listeners; perhaps our study's listeners, who had lower intact baseline speech scores (on average) than the CI users in the original study, found the 75% duty cycle condition challenging, and therefore a relationship between intact speech scores and restoration effects began to emerge. However, due to the lack of a substantial amount of data in the high-scoring regions, as well as an outlier in the low-scoring region, it was difficult to ascertain the exact nature and strength of this relationship in the current study. Additional data should be collected to characterize the relationship of baseline sentence understanding and perceptual restoration in CI listeners, particularly in the low and high sentence understanding score regions, as well as whether that relationship changes in response to task difficulty. Very low performers may be experiencing extremely poor spectral resolution, which has been shown in NH listeners presented vocoded speech to cause less restoration (Başkent, 2012; Bhargava et al., 2014; Clarke et al., 2016). Very high performers may not be sufficiently challenged by the interruption parameters used in the study to show a restoration effect, in that performance with SG interrupted speech is near-perfect, resulting in little need to utilize noise bursts to accurately perceive speech—a similar effect appeared to occur among the NH participants in the original study when presented unprocessed speech (Bhargava et al., 2014).
To summarize, restoration of interrupted sentences may exist among certain CI users—many of whom the original study had access to for their experiment—and under specific interruption conditions, but does not appear to function in the present study's sample across different interruption contexts, on average.
V. GENERAL DISCUSSION
Previous work in the field demonstrated that perceptual restoration (as measured via the interrupted speech paradigm) provided a benefit to CI users listening to noise-interrupted speech, as long as longer durations of intact speech information were made available to the listener between the interrupting noise bursts (Bhargava et al., 2014). A study using a similar paradigm, in contrast, did not replicate this effect (Jaekel et al., 2021); however, one hypothesized explanation for the different results between studies was the extent to which front-end preprocessing was activated for the various participants.
While speech understanding generally improves with the activation of front-end preprocessing like noise reduction algorithms (Mauger et al., 2014), it was unknown how speech repair mechanisms like perceptual restoration are affected by such processing. Perceptual restoration relies on the sensation of speech continuing through a masking or interrupting noise burst. If front-end preprocessing reduces the level of the masking or interrupting noises such that they fail to elicit the illusion of plausibly masked speech, the ability of perceptual restoration to help “fill in” the missing speech information may not be as successful (Başkent, 2012).
Study 1, which utilized the interrupted speech paradigm and compared CI users' performance with activated vs inactivated adjustable front-end preprocessing algorithms, detected no perceptual restoration benefits in either condition [Fig. 1(B)]. Instead, while NB interrupted sentence understanding improved with the activation of signal processing, performance with noise was significantly poorer compared to performance with SG interrupted speech [Fig. 1(A)]. It was unclear whether this result indicated that CI users are unable to consistently utilize restoration mechanisms in the interrupted speech paradigm, or, as posited by Study 2, if non-modifiable compression algorithms like AGC were contributing to reduced restoration ability.
Compression algorithms could reduce restoration by distorting the amplitude envelope of speech immediately subsequent to noise bursts and/or by changing the effective SNR of the stimuli. Study 2 measured restoration benefits with interrupted sentences that were expected to either engage or not engage AGC (Başkent et al., 2009; Khing et al., 2013). As occurred in Study 1, no perceptual restoration benefits were detected in CI users, on average, in either condition [Fig. 2(B)]. Similarly, while NB interrupted sentence understanding improved with the engagement of AGC, performance with SG interrupted speech was significantly more successful [Fig. 2(A)]. Thus, across both planned studies, the interrupted sentences paradigm failed to elicit perceptual restoration benefits in any condition in CI users. Despite attempts to alter the incoming noise signals via compression activation and front-end preprocessing, the result was essentially the same—the CI users recruited for these studies were less successful at NB interrupted speech understanding compared to SG interrupted speech understanding.
Based on these findings, we developed a third study: a “close” replication (LeBel et al., 2017) of the original study by Bhargava et al. (2014). The aims of this third study were not only to assess whether we could replicate results reported in the literature but also to measure whether the frequency and duration of interruptions influenced restoration in CI users. Studies 1 and 2 both utilized the same duty cycle and interruption rate; thus, perhaps this parameter combination was not conducive to restoration in CI users.
We found that we were unable to replicate the +4.6 RAU restoration benefit reported in the original study by Bhargava et al. (2014), despite utilizing the same interruption parameters and similar experimental procedures. Instead, we found a −0.9 RAU restoration effect [Fig. 3(C)]. However, it is important to note that several differences existed between our close replication study and the original Bhargava study, any of which could contribute to the discrepancy in restoration effects found. First, as it was necessary to test our participants in the English language, we were unable to use the Dutch speech corpus used in the original study (Versfeld et al., 2000). Instead, we presented the IEEE corpus (Rothauser et al., 1969), which may differ from the Dutch speech corpus in terms of sentence complexity, length, and/or amount of context provided. Choice of speech corpus may impact restoration effects: both younger and older NH listeners showed different magnitudes of perceptual restoration benefits when presented two different speech corpora, whether speech was unprocessed or vocoded to simulate aspects of CI processing (Jaekel et al., 2018). Thus, the choice of speech corpus could have contributed to the differences between the current study and the original study by Bhargava et al. (2014).
A second important difference between the studies is that our sample was older than the sample tested by Bhargava et al. (2014): the mean age in our study was 60.4 years, while the mean age in the original study was 48.5 years (Table V). Among NH listeners, older age is typically associated with greater perceptual restoration benefits, whether speech is unprocessed (Saija et al., 2014; Bologna et al., 2018; Jaekel et al., 2018) or vocoded (Jaekel et al., 2018). Older adults may be more likely to rely on top-down processing of incoming speech, and more likely to utilize context and linguistic knowledge to repair speech, than younger adults (Pichora-Fuller, 2008; Saija et al., 2014; Jaekel et al., 2018). However, the signal distortions introduced by SG interruptions may also disproportionately reduce older adults' scores for SG interrupted speech compared to younger adults (Bologna et al., 2018). Compared to older NH adults, it is unclear whether older CI users show a similar “aging benefit” for perceptual restoration. Similar to older NH adults, older age likely contributed to decreased SG interrupted speech understanding in CI users (Bhargava et al., 2016). However, unlike older NH adults, Jaekel et al. (2021) found no influence of aging on CI users' NB interrupted speech understanding – even when semantic cues were made available, which could potentially increase the opportunity for older CI users to use top-down processing. This could indicate that older CI users do not experience the “aging benefit” for perceptual restoration, and thus our older group would be expected to show less restoration than the group tested in the original study. However, in the current study, older age did not appear to be systematically related to reduced restoration; both negative and positive restoration effects were observed across the age range. Thus, despite our sample being older, on average, than participants in the original study, age did not appear to be directly affecting restoration results.
Our study and the original study by Bhargava et al. (2014) also differed in terms of participants' intact speech understanding scores (Table V). Participants in our study had lower intact speech scores, on average, possibly indicating that they were poorer performers than the participants enrolled in the Bhargava study. Indeed, Fig. 3(A) shows that the average performance with interrupted speech in the replicated condition was lower in our study than in the original study. While the original study found no significant relationship between intact speech understanding scores and restoration benefit with a 75% duty cycle, our data indicate a possible positive relationship—participants with lower intact scores also tended to have more negative restoration effects (Fig. 4). Had we enrolled participants who better fit into the range of intact speech understanding scores reported by Bhargava et al. (2014), we may have found restoration benefits similar to the original study's.
Beyond the reasons why our study's results did not match those reported by Bhargava et al. (2014), it is important to explore why the results from Study 3 were also different from the results reported in Studies 1 and 2. One of the most substantial changes among the studies was that the experimental procedures in Study 3 differed from those used in Studies 1 and 2. Study 3 involved a familiarization phase, where participants practiced listening to both SG and NB interrupted sentences, receiving both visual and auditory feedback about accuracy. Both written (visual) and auditory feedback have been shown to increase the perceptual learning and understanding of noise-vocoded speech (Davis et al., 2005). Perceptual learning of interrupted speech material may have been strongly supported by simultaneously including both types of feedback in Study 3. While training with NB or SG interrupted vocoded speech has not been shown to result in greater perceptual restoration benefits in NH listeners (Benard and Başkent, 2014), it is possible that CI users are amenable to such training, and that even with the short, ten-sentence training session used in Study 3, some CI users experienced improved NB interrupted sentence processing. Another difference across Studies 1–3 was that a familiar example sentence, processed in the same manner as the subsequent test sentences, was always presented at the beginning of each sentence list in Study 3. Access to this example sentence may have helped participants adjust their listening strategy and expectations for the upcoming sentences. Overall, task familiarization, multimodal feedback, and access to an exemplar sentence may have all contributed to the generally higher accuracy with NB interrupted speech observed in this third study [Fig. 3(B)] compared to Studies 1 and 2 [Figs. 1(A) and 2(A)]. A possible consequence of this improvement is that, while individual participants' restoration effects in Studies 1 and 2 were nearly always negative [see Figs. 1(B) and 2(B)], a wider range of restoration effects was observed in Study 3 [Fig. 3(C)]. That is, while many participants still demonstrated negative restoration effects, many other participants demonstrated positive restoration effects or experienced no restoration effects at all (i.e., because performance with SG and NB interrupted sentences were approximately equivalent, the restoration effect was calculated as zero).
It is possible that the differences in interruption rate and duty cycle parameters across studies also exerted effects on performance. However, the 5-Hz, 85% duty cycle condition in Study 3 was quite similar to the parameters used in Studies 1 and 2 (5-Hz, 80% duty cycle), and yet the restoration results were strikingly different across studies. While none of the studies demonstrated overall average restoration benefits in CI users (except for in two conditions; see below), the results from Study 3 showed a much less negative effect of NB interruptions on speech understanding in comparison to SG interruptions for many CI users.
Finally, despite not detecting restoration benefits in the replicated condition, restoration benefits were observed, on average, in two listening conditions in Study 3: sentences with less frequent interruptions but longer durations of missing speech information (the 1.5-Hz, 65% duty cycle condition), and sentences with more frequent interruptions but shorter durations of missing speech information (the 5-Hz, 75% duty cycle condition). However, these average restoration benefits were small (+2.3 to +2.6 RAUs). When viewed in the context of the small but negative restoration effects in the remaining four conditions (−4.3 to −0.9 RAUs), it appears that across all listening conditions, average restoration effects were essentially near zero. While individual variability existed in terms of the ability to utilize perceptual restoration in our interrupted speech paradigm, CI users on average did not appear to have access to, or receive benefit from, this speech repair mechanism. While the interrupted speech paradigm provides a rather unnatural listening task for the CI user, our data are at least suggestive that individuals with CIs may, as a group, likewise show less use of perceptual restoration in real-world listening environments.
VI. CONCLUSION
This study aimed to measure the extent to which adjustable and non-adjustable signal processing algorithms in CIs affected users' ability to perceptually restore interrupted speech. While we were unable to detect consistent perceptual restoration benefits in our sample, both the adjustable front-end preprocessing and non-adjustable compression algorithms, when activated, increased NB interrupted sentence understanding. While perceptual restoration benefits in CI users were not observed, on average, in our interrupted speech paradigm, it is possible that other experimental methods—especially when utilized with more diverse samples of CI users—could help characterize how this speech repair mechanism functions in this population.
ACKNOWLEDGMENTS
Research reported in this publication was supported by the National Institute On Deafness and Other Communication Disorders of the National Institutes of Health under Award Nos. F31DC017362 (B.N.J.), T32DC000046 (Trainee: B.N.J.), and R01DC014948 (M.J.G.). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. Portions of this work were presented at the Conference on Implantable Auditory Prostheses (2021). Thank you to Deniz Başkent and Etienne Gaudrain for helpful discussions on portions of this work. Thank you also to Alyssa Giammetta, Stefanie Kuchinsky, Jan Edwards, Samira Anderson, Catherine Carr, Maureen Shader, Kristina Milvae, Nicole Nguyen, Olga Stahkovskaya, Julie Cohen, Elizabeth Kolberg, Ginny Alexander, Will Bologna, Zilong Xie, Emily Shroads, Debbie Moon, Emma Peterson, Bobby Gibbs, and Kelly Miller for assistance with stimuli creation, participant recruitment, and data collection, and/or for providing feedback on this work.
Footnotes
The “Simpson's Paradox” took the following form. Inspection of our data revealed that two “clusters” of participants were inadvertently recruited for this study—one cluster with high overall speech understanding scores, and one cluster with low overall speech understanding scores. Within each of these clusters, there was a relationship between working memory scores and speech understanding scores; specifically, better working memory scores were associated with better speech understanding scores. However, when all the data were analyzed together (i.e., these two clusters were ignored), this relationship between variables appeared to reverse: for the entire sample of participants, it appeared that poorer working memory scores were associated with better speech understanding scores. Recruitment of more mid-range performers could have possibly averted this paradoxical result.
References
- 1. Barr, D. J. , Levy, R. , Scheepers, C. , and Tily, H. J. (2013). “ Random effects structure for confirmatory hypothesis testing: Keep it maximal,” J. Mem. Lang. 68, 255–278. 10.1016/j.jml.2012.11.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 2. Bashford, J. A. , Riener, K. R. , and Warren, R. M. (1992). “ Increasing the intelligibility of speech through multiple phonemic restorations,” Percept. Psychophys. 51, 211–217. 10.3758/BF03212247 [DOI] [PubMed] [Google Scholar]
- 3. Başkent, D. (2012). “ Effect of speech degradation on top-down repair: Phonemic restoration with simulations of cochlear implants and combined electric-acoustic stimulation,” JARO 13, 683–692. 10.1007/s10162-012-0334-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4. Başkent, D. , Eiler, C. , and Edwards, B. (2009). “ Effects of envelope discontinuities on perceptual restoration of amplitude-compressed speech,” J. Acoust. Soc. Am. 125, 3995–4005. 10.1121/1.3125329 [DOI] [PubMed] [Google Scholar]
- 5. Benard, M. R. , and Başkent, D. (2014). “ Perceptual learning of temporally interrupted spectrally degraded speech,” J. Acoust. Soc. Am. 136, 1344–1351. 10.1121/1.4892756 [DOI] [PubMed] [Google Scholar]
- 6. Benard, M. R. , Mensink, J. S. , and Başkent, D. (2014). “ Individual differences in top-down restoration of interrupted speech: Links to linguistic and cognitive abilities,” J. Acoust. Soc. Am. 135, EL88–EL94. 10.1121/1.4862879 [DOI] [PubMed] [Google Scholar]
- 7. Bhargava, P. , Gaudrain, E. , and Başkent, D. (2014). “ Top-down restoration of speech in cochlear-implant users,” Hear Res 309, 113–123. 10.1016/j.heares.2013.12.003 [DOI] [PubMed] [Google Scholar]
- 8. Bhargava, P. , Gaudrain, E. , and Başkent, D. (2016). “ The intelligibility of interrupted speech: Cochlear implant users and normal hearing listeners,” JARO 17, 475–491. 10.1007/s10162-016-0565-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9. Blyth, C. R. (1972). “ On Simpson's Paradox and the Sure-Thing Principle,” J. Am. Stat. Assoc. 67, 364–366. 10.1080/01621459.1972.10482387 [DOI] [Google Scholar]
- 10. Bologna, W. J. , Vaden, K. I., Jr. , Ahlstrom, J. B. , and Dubno, J. R. (2018). “ Age effects on perceptual organization of speech: Contributions of glimpsing, phonemic restoration, and speech segregation,” J. Acoust. Soc. Am. 144, 267–281. 10.1121/1.5044397 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11. Clarke, J. , Başkent, D. , and Gaudrain, E. (2016). “ Pitch and spectral resolution: A systematic comparison of bottom-up cues for top-down repair of degraded speech,” J. Acoust. Soc. Am. 139, 395–405. 10.1121/1.4939962 [DOI] [PubMed] [Google Scholar]
- 12. Davidson, L. S. , Geers, A. E. , and Brenner, C. (2010). “ Cochlear implant characteristics and speech perception skills of adolescents with long-term device use,” Otol. Neurotol. 31, 1310–1314. 10.1097/MAO.0b013e3181eb320c [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13. Davis, M. H. , Johnsrude, I. S. , Hervais-Adelman, A. , Taylor, K. , and McGettigan, C. (2005). “ Lexical information drives perceptual learning of distorted speech: Evidence from the comprehension of noise-vocoded sentences,” J. Exp. Psychol. Gen. 134, 222–241. 10.1037/0096-3445.134.2.222 [DOI] [PubMed] [Google Scholar]
- 14. Dorman, M. F. , Loizou, P. C. , Fitzke, J. , and Tu, Z. (1998). “ The recognition of sentences in noise by normal-hearing listeners using simulations of cochlear-implant signal processors with 6–20 channels,” J. Acoust. Soc. Am. 104, 3583–3585. 10.1121/1.423940 [DOI] [PubMed] [Google Scholar]
- 15. Dunn, L. M. , and Dunn, D. M. (2007). The Peabody Picture Vocabulary Test, 4th ed. ( NCS Pearson, Inc., Bloomington, MN: ). [Google Scholar]
- 16. Fetterman, B. L. , and Domico, E. H. (2002). “ Speech recognition in background noise of cochlear implant patients,” Otolaryngol. Head Neck Surg. 126, 257–263. 10.1067/mhn.2002.123044 [DOI] [PubMed] [Google Scholar]
- 17. Friesen, L. M. , Shannon, R. V. , Başkent, D. , and Wang, X. (2001). “ Speech recognition in noise as a function of the number of spectral channels: Comparison of acoustic hearing and cochlear implants,” J. Acoust. Soc. Am. 110, 1150–1163. 10.1121/1.1381538 [DOI] [PubMed] [Google Scholar]
- 18. Fu, Q. J. , and Shannon, R. V. (1999). “ Recognition of spectrally degraded and frequency-shifted vowels in acoustic and electric hearing,” J. Acoust. Soc. Am. 105, 1889–1900. 10.1121/1.426725 [DOI] [PubMed] [Google Scholar]
- 19. Gershon, R. C. , Wagster, M. V. , Hendrie, H. C. , Fox, N. A. , Cook, K. F. , and Nowinski, C. J. (2013). “ NIH toolbox for assessment of neurological and behavioral function,” Neurology 80, S2–S6. 10.1212/WNL.0b013e3182872e5f [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20. Gifford, R. H. , and Revit, L. J. (2010). “ Speech perception for adult cochlear implant recipients in a realistic background noise: Effectiveness of preprocessing strategies and external options for improving speech recognition in noise,” J. Am. Acad. Audiol. 21, 441–451. 10.3766/jaaa.21.7.3 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21. Gilden, J. , Lewis, K. , Grant, G. , and Crosson, J. (2015). “ Improved hearing in noise using new signal processing algorithms with the Cochlear Nucleus 6 sound processor,” J. Otol. 10, 51–56. 10.1016/j.joto.2015.09.001 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22. Jaekel, B. N. , Newman, R. S. , and Goupell, M. J. (2018). “ Age effects on perceptual restoration of degraded interrupted sentences,” J. Acoust. Soc. Am. 143, 84–97. 10.1121/1.5016968 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23. Jaekel, B. N. , Weinstein, S. , Newman, R. S. , and Goupell, M. J. (2021). “ Access to semantic cues does not lead to perceptual restoration of interrupted speech in cochlear-implant users,” J. Acoust. Soc. Am. 149, 1488–1497. 10.1121/10.0003573 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24. Jin, S. H. , Nie, Y. , and Nelson, P. (2013). “ Masking release and modulation interference in cochlear implant and simulation listeners,” Am. J. Audiol. 22, 135–146. 10.1044/1059-0889(2013/12-0049) [DOI] [PubMed] [Google Scholar]
- 25. Khing, P. P. , Swanson, B. A. , and Ambikairajah, E. (2013). “ The effect of automatic gain control structure and release time on cochlear implant speech intelligibility,” PLoS One 8, e82263. 10.1371/journal.pone.0082263 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26. LeBel, E. P. , Berger, D. , Campbell, L. , and Loving, T. J. (2017). “ Falsifiability is not optional,” J. Pers. Soc. Psychol. 113, 254–261. 10.1037/pspi0000106 [DOI] [PubMed] [Google Scholar]
- 27. Loizou, P. C. (2006). “ Speech processing in vocoder-centric cochlear implants,” Adv. Otorhinolaryngol. 64, 109–143. 10.1159/000094648 [DOI] [PubMed] [Google Scholar]
- 28. Loizou, P. C. , Hu, Y. , Litovsky, R. , Yu, G. , Peters, R. , Lake, J. , and Roland, P. (2009). “ Speech recognition by bilateral cochlear implant users in a cocktail-party setting,” J. Acoust. Soc. Am. 125, 372–383. 10.1121/1.3036175 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 29. Mauger, S. J. , Warren, C. D. , Knight, M. R. , Goorevich, M. , and Nel, E. (2014). “ Clinical evaluation of the Nucleus 6 cochlear implant system: Performance improvements with SmartSound iQ,” Int. J. Audiol. 53, 564–576. 10.3109/14992027.2014.895431 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30. Miller, G. A. , and Licklider, J. C. R. (1950). “ The intelligibility of interrupted speech,” J. Acoust. Soc. Am. 22, 167–173. 10.1121/1.1906584 [DOI] [Google Scholar]
- 31. Moberly, A. C. , Houston, D. M. , and Castellanos, I. (2016). “ Non-auditory neurocognitive skills contribute to speech recognition in adults with cochlear implants,” Laryngoscope Investig. Otolaryngol. 1, 154–162. 10.1002/lio2.38 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32. Nasreddine, Z. S. , Phillips, N. A. , Bedirian, V. , Charbonneau, S. , Whitehead, V. , Collin, I. , Cummings, J. L. , and Chertkow, H. (2005). “ The Montreal Cognitive Assessment, MoCA: A brief screening tool for mild cognitive impairment,” J. Am. Geriatrics Soc. 53, 695–699. 10.1111/j.1532-5415.2005.53221.x [DOI] [PubMed] [Google Scholar]
- 33. Nelson, P. B. , and Jin, S. H. (2004). “ Factors affecting speech understanding in gated interference: Cochlear implant users and normal-hearing listeners,” J. Acoust. Soc. Am. 115, 2286–2294. 10.1121/1.1703538 [DOI] [PubMed] [Google Scholar]
- 34. Nelson, P. B. , Jin, S. H. , Carney, A. E. , and Nelson, D. A. (2003). “ Understanding speech in modulated interference: Cochlear implant users and normal-hearing listeners,” J. Acoust. Soc. Am. 113, 961–968. 10.1121/1.1531983 [DOI] [PubMed] [Google Scholar]
- 35. Oxenham, A. J. , and Kreft, H. A. (2014). “ Speech perception in tones and noise via cochlear implants reveals influence of spectral resolution on temporal processing,” Trends Hear. 18, 1–14. 10.1177/2331216514553783 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 36. Pichora-Fuller, M. K. (2008). “ Use of supportive context by younger and older adult listeners: Balancing bottom-up and top-down information processing,” Int. J. Audiol. 47, S72–S82. 10.1080/14992020802307404 [DOI] [PubMed] [Google Scholar]
- 37. Rakszawski, B. , Wright, R. , Cadieux, J. H. , Davidson, L. S. , and Brenner, C. (2016). “ The effects of pre-processing strategies for pediatric cochlear implant recipients,” J. Am. Acad. Audiol. 27, 85–102. 10.3766/jaaa.14058 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38. Reiss, L. A. J. , Ito, R. A. , Eggleston, J. L. , and Wozny, D. R. (2014). “ Abnormal binaural spectral integration in cochlear implant users,” JARO 15, 235–248. 10.1007/s10162-013-0434-8 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 39. Rothauser, E. , Chapman, W. , Guttman, N. , Nordby, K. , Silbiger, H. , Urbanek, G. , and Weinstock, M. (1969). “ IEEE recommended practice for speech quality measurements,” IEEE Trans. Audio Electroacoust. 17, 225–246. 10.1109/TAU.1969.1162058 [DOI] [Google Scholar]
- 40. Saija, J. D. , Akyürek, E. G. , Andringa, T. C. , and Başkent, D. (2014). “ Perceptual restoration of degraded speech is preserved with advancing age,” JARO 15, 139–148. 10.1007/s10162-013-0422-z [DOI] [PMC free article] [PubMed] [Google Scholar]
- 41. Shannon, R. V. , Zeng, F. G. , Kamath, V. , Wygonski, J. , and Ekelid, M. (1995). “ Speech recognition with primarily temporal cues,” Science 270, 303–304. 10.1126/science.270.5234.303 [DOI] [PubMed] [Google Scholar]
- 42. Sladen, D. P. , and Zappler, A. (2015). “ Older and younger adult cochlear implant users: Speech recognition in quiet and noise, quality of life, and music perception,” Am. J. Audiol. 24, 31–39. 10.1044/2014_AJA-13-0066 [DOI] [PubMed] [Google Scholar]
- 43. Stilp, C. E. , Kiefte, M. , Alexander, J. M. , and Kluender, K. R. (2010). “ Cochlea-scaled spectral entropy predicts rate-invariant intelligibility of temporally distorted sentences,” J. Acoust. Soc. Am. 128, 2112–2126. 10.1121/1.3483719 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44. Studebaker, G. A. (1985). “ A rationalized arcsine transform,” J. Speech. Lang. Hear. Res. 28, 455–462. 10.1044/jshr.2803.455 [DOI] [PubMed] [Google Scholar]
- 45. Verschuure, J. , and Brocaar, M. P. (1983). “ Intelligibility of interrupted meaningful and nonsense speech with and without intervening noise,” Percept. Psychophys. 33, 232–240. 10.3758/BF03202859 [DOI] [PubMed] [Google Scholar]
- 46. Versfeld, N. J. , Daalder, L. , Festen, J. M. , and Houtgast, T. (2000). “ Method for the selection of sentence materials for efficient measurement of the speech reception threshold,” J. Acoust. Soc. Am. 107, 1671–1684. 10.1121/1.428451 [DOI] [PubMed] [Google Scholar]
- 47. Warren, R. M. (1970). “ Perceptual restoration of missing speech sounds,” Science 167, 392–393. 10.1126/science.167.3917.392 [DOI] [PubMed] [Google Scholar]
- 48. Wilson, B. S. (2017). “ The cochlear implant and possibilities for narrowing the remaining gaps between prosthetic and normal hearing,” World J. Otorhinolaryngol. 3, 200–210. 10.1016/j.wjorl.2017.12.005 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 49. Wolfe, J. , Neumann, S. , Marsh, M. , Schafer, E. , Lianos, L. , Gilden, J. , O'Neill, L. , Arkis, P. , Menapace, C. , Nel, E. , and Jones, M. (2015). “ Benefits of adaptive signal processing in a commercially available cochlear implant sound processor,” Otol. Neurotol. 36, 1181–1190. 10.1097/MAO.0000000000000781 [DOI] [PubMed] [Google Scholar]
- 50. Xu, K. , Willis, S. , Gopen, Q. , and Fu, Q. J. (2020). “ Effects of spectral resolution and frequency mismatch on speech understanding and spatial release from masking in simulated bilateral cochlear implants,” Ear Hear. 41, 1362–1371. 10.1097/AUD.0000000000000865 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51. Zhao, F. , Stephens, S. D. G. , Sim, S. W. , and Meredith, R. (1997). “ The use of qualitative questionnaires in patients having and being considered for cochlear implants,” Clin. Otolaryngol. 22, 254–259. 10.1046/j.1365-2273.1997.00036.x [DOI] [PubMed] [Google Scholar]