Effects of the configuration of hearing loss on consonant perception between simulated bimodal and electric acoustic stimulation hearing

Abstract

Background:

Cochlear implant technology allows for acoustic and electric stimulation to be combined across ears (bimodal) and within the same ear (electric acoustic stimulation, or EAS). Mechanisms used to integrate speech acoustics may be different between the bimodal and EAS hearing, and the configurations of hearing loss might be an important factor for the integration. Thus, differentiating the effects of different configurations of hearing loss on bimodal or EAS benefit in speech perception (differences in performance with combined acoustic and electric stimulation from a better stimulation alone) is important.

Purpose:

Using acoustic simulation, we determined how consonant recognition was affected by different configurations of hearing loss in bimodal and EAS hearing.

Research Design:

A mixed design was used with one between-subject variable (simulated bimodal group vs simulated EAS group) and one within-subject variable (acoustic stimulation alone, electric stimulation alone, combined acoustic and electric stimulation).

Study Sample:

Twenty adult subjects (ten for each group) with normal hearing were recruited.

Data Collection and Analysis:

Consonant perception was unilaterally or bilaterally measured in quiet. For the acoustic stimulation, four different simulations of hearing loss were created by band-pass filtering consonants with a fixed lower cutoff frequency of 100 Hz and each of the four upper cutoff frequencies of 250, 500, 750, and 1000 Hz. For the electric stimulation, an 8-channel noise vocoder was used to generate a typical spectral mismatch by using fixed input (200-7000 Hz) and output (1000-7000 Hz) frequency ranges. The effects of simulated hearing loss on consonant recognition were compared between the two groups.

Results:

Significant bimodal and EAS benefit occurred regardless of the configurations of hearing loss and hearing technology (bimodal vs EAS). Place information was better transmitted in EAS hearing than in bimodal hearing.

Conclusions:

These results suggest that configurations of hearing loss are not a significant factor for integrating consonant information between acoustic and electric stimulation. The results also suggest that mechanisms used to integrate consonant information may be similar between bimodal and EAS hearing.

Introduction

Bimodal hearing, the use of a cochlear implant (CI) in conjunction with a hearing aid (HA) on opposite ears, produces a bimodal benefit in speech perception (Gifford et al., 2007). Electric Acoustic Stimulation (EAS) hearing, which is the combined use of a CI and a HA in the same ear, also provides a greater EAS benefit in speech perception (Gantz et al., 2009). Bimodal or EAS benefit is defined as a difference in performance with combined acoustic and electric stimulation from an electric stimulation alone (i.e., CI+HA performance – CI performance). However, bimodal or EAS benefit varies greatly within and between the groups. It has been thought that the variability in bimodal or EAS benefit is directly linked to the configurations of hearing loss, that is, better residual hearing in the HA ear, greater bimodal or EAS benefit in speech perception (Mok et al., 2006). However, there are mixed results. Ching et al. (2004) and Gifford et al. (2007) showed no correlation between bimodal benefit and hearing thresholds. Other researchers showed a significant correlation with unaided (Sheffield and Zeng, 2012, using a simulation) and aided (Mok et al., 2006; 2010) hearing thresholds. To our knowledge, limited correlation data for EAS hearing are available to date.

Bimodal and EAS hearing may have different mechanisms to integrate speech acoustics. As bimodal patients integrate speech cues across ears, the integration process might be limited due to less optimal binaural functioning in the central auditory system such as the cochlear nucleus complex or/and inferior colliculus. In contrast, the integration process in EAS patients might be less affected by binaural functioning than in bimodal patients because EAS patients integrate the speech acoustics within the same ear (Trevino et al., 2010). Fu and colleagues compared the effect of tonotopic mismatch between input and output frequency range of CI on vowel perception between simulated bimodal and EAS hearing (2017). Gifford et al. (2017) compared sentence perception in bimodal and EAS patients with a contralateral HA. Both studies found a greater EAS benefit over bimodal benefit. The results from the two studies support the idea that EAS hearing may provide listening environment for a better integration of speech information between acoustic and electric stimulations.

One limitation of Fu et al.’s study (2017) is the use of a fixed acoustic stimulation frequency range (200-600 Hz). Different configurations of hearing loss might have different effects on integration both within and between the bimodal and EAS patients. In addition, as Gifford et al.’s study (2007) tested EAS patients with a contralateral HA, we know little about the contribution from the contralateral HA to EAS benefit. Another challenging aspect of testing clinical population is a difficulty to control important patient-related variables such as durations of hearing loss, preoperative speech perception abilities, and duration of hearing device use. All are known to be important factors for speech perception (Gifford et al., 2007). In this study with normal hearing (NH) adults, we simulated bimodal and EAS hearing (i.e., simulated bimodal and EAS groups) with various configurations of hearing loss typically found from bimodal and EAS patients. The use of a simulation helped to explicitly control the extent of stimulation within the cochlea and directly compare the perception of combined acoustic and electric stimulations between bimodal and EAS hearing. The goal of this study was to determine how different configurations of hearing loss affect the integration of consonant information in simulated bimodal and EAS hearing.

Materials and Method

Participants

Two separate groups of ten adult NH subjects were recruited for the simulated bimodal group (8 females and 2 males; average and standard deviation: 23±4.7 years) and the simulated EAS group (6 females and 4 males; average age and standard deviation: 26±3.9 years). Subjects were native speakers of American English. All NH listeners had thresholds better than 20dB HL at 250, 500, 1000, 2000, 4000, and 8000 Hz. All procedures were approved by the Texas Tech University Health Sciences Center Institutional Review Board.

Stimuli

Each subject was presented with 16 consonants, followed by a common vowel /a/ (/ba, da, ga, pa, ta, ka, ma, na, fa, sa, ʃa, va, za, ʒa, ða, Ɵa/). The stimuli (mean duration and standard deviation: 425.71±18.35 ms) were produced by a single female talker (F0: 228 Hz) and delivered via Sennheiser HDA-200 headphones. After both acoustic and electric simulation, all processed stimuli were normalized to have the same long-term root-mean-square energy (65 dBA).

Bimodal and EAS Simulation

For both groups, there were nine listening conditions, as shown in Figure 1: four acoustic alone (A250, A500, A750, and A1000), one electric (E) alone, four combined acoustic and electric (E+A250, E+A500, E+A750, and E+A1000). For the bimodal condition, the acoustic and electric stimulation were delivered to opposite ears, whereas for the EAS condition, both acoustic and electric stimulations were presented to the same ear.

Figure 1:

Illustration of the four simulated acoustic (A) and one simulated electric (E) stimulation conditions. For the electric (E) stimulation, spectral mismatch between analysis and carrier bands was indicated by the arrows.

For the acoustic simulation, four different configurations of simulated high-frequency hearing loss typically found in bimodal and EAS patients were created using band-pass filters (Skarzynski and Lorens, 2010). Each consonant was lowpass filtered with a fixed lower cutoff frequency of 100 Hz and with one of the four upper cutoff frequencies of 250, 500, 750, and 1000 Hz (20^th order Butterworth filters; 240 dB/octave). A long-term root-mean-square of all lowpass filtered consonant syllables was adjusted to have the same 65 dBA.

For the electric stimulation alone, an 8-channel noise vocoder was used with fixed input (200-7000 Hz) and output (1000-7000 Hz) frequency ranges to create a typical tonotopic spectral mismatch, as indicated by the arrows in Figure 1. We used the fixed input frequency ranges to simulate typical frequency range for CI fitting. The lower output frequency (1000 Hz) is a mean of the upper edge of residual hearing for many EAS patients (Karsten et al., 2013). The upper output frequency (7000 Hz) is similar to the highest input frequency commonly used in commercial CI speech processors (Gifford et al., 2017). According to Greenwood’s function (Greenwood, 1990), the input frequency range was divided into eight bands (4^th order Butterworth filters; 48 dB/octave). The temporal envelope was extracted from each band by half-wave rectification and lowpass filtering (4^th order Butterworth filter with 160 Hz cutoff frequency; 24 dB/octave). The temporal envelope from each band was used to modulate with a white noise and then the modulated noise was frequency limited by filtering with the same bandpass filters used in the original analysis band. The modulated noise-bands were summed, and the output was adjusted to have the same long-term root-mean-square energy as the input speech signal (i.e., 65 dBA). The corner frequencies used for the analysis and carrier band simulations were given in Table 1.

Table 1:

Corner frequencies used for the electric simulation.

Channel	Analysis bands (Hz)	Carrier bands (Hz)
1	200-359.1	1000-1294.5
2	359.1-591.3	1294.5-166.5
3	591.3-930.5	1664.8-2130.3
4	930.5-1425.8	2130.3-2715.5
5	1425.8-2149.1	2715.5-3451.1
6	2149.1-3205.3	3451.1-4375.9
7	3205.3-4747.7	4375.9-5538.5
8	4747.7-7000	5538.5-7000

Procedures

Consonant recognition was measured in a single-walled sound booth (Industrial Acoustics) in quiet with the acoustic stimulation alone, electric stimulation alone, and combined acoustic and electric stimulation in the 16-alternative forced-choice paradigm. To reduce a right ear advantage in both groups, half of the participants were tested for the acoustic stimulation with their left ears and the electric stimulation with their right ears (Roup, 2011). The other half were tested in the opposite manner. The order of testing was counter-balanced within and across groups. To familiarize the subjects with the acoustic, electric, and combined stimulations, all subjects were provided 10-minutes familiarization (total 30 minutes) before the formal testing. All stimuli were presented at the listener’s most comfortable loudness level (range: 50 ~70 dB SPL) according to the Cox loudness rating scale in response to ten presentations of the consonants in quiet (Cox, 1995). Each consonant was presented 10 times in a random order for each of the nine listening conditions (16 consonants x 10 repetitions x 9 listening conditions = 1440 trials in total). The subject responded by clicking on one of the 16 response boxes labeled in /consonant-a/ context. The complete test protocol required two hours with enough break time. No trial-by-trial feedback was provided.

Data Analysis

A two-way mixed analysis of variance (ANOVA) was performed to determine the main effects of the group (i.e., bimodal vs EAS) and the listening condition (i.e., four acoustic alone, one electric alone, and four combined acoustic and electric). Pairwise multiple comparisons were performed with the Bonferroni correction. The effect size (η²) was also computed for the main effects and was interpreted as a small, medium, or large effect if η²≤0.01, η²≤0.059, or η²≥0.138, respectively (Cohen, 1988). A significance level of 0.05 with a two-tailed test was used.

Results

Figure 2 depicts the mean consonant perception scores with standard errors in bimodal and EAS groups as a function of the listening condition. The numbers in percentage above the bars indicate mean bimodal and EAS benefits. A significant main effect was observed for the listening condition, F(8, 162)=05.2, p<0.001; η²=.84 but not for the group, F(1, 162)=0.84, p=0.36; η²=.001. No significant interaction was found, F(8, 162)=0.21, p=0.99. Pairwise multiple comparisons showed significant differences for any pairs between combined stimulation and either acoustic or electric stimulation alone conditions in both groups (p<0.001). The multiple comparisons, however, showed no significant differences for any pairs among the combined stimulation in either group (p>0.05). The complete pairwise multiple comparison results are given in Table 2. Neither bimodal nor EAS interference (poorer performance with bimodal than with a better ear alone) was observed.

Figure 2:

Mean percent correct with standard errors between the groups for each listening condition. The numbers in percentage above the bars indicate mean bimodal and EAS benefits.

Table 2:

Pairwise multiple comparisons, along with average recognition scores in parentheses for the bimodal and EAS groups.

Bimodal Group	E+A1000 (69.9%)	E+A750 (67.1%)	E+A500 (66.2%)	E+A250 (66.5%)	E alone (52.3%)	A1000 (46.1%)	A750 (38.4%)	A500 (26.4%)	A250 (10.4%)
E+A1000		n.s	n.s	n.s	***	***	***	***	***
E+A750	n.s		n.s	n.s	***	***	***	***	***
E+A500	n.s	n.s		n.s	***	***	***	***	***
E+A250	n.s	n.s	n.s		***	***	***	***	***
E alone	***	***	***	***		***	***	***	***
A1000	***	***	***	***	***		n.s	***	***
A750	***	***	***	***	***	n.s		n.s	***
A500	***	***	***	***	***	***	n.s		***
A250	***	***	***	***	***	***	***	***
EAS Group	E+A1000 (73.2%)	E+A750 (70.3%)	E+A500 (70.1%)	E+A250 (64.9%)	E alone (53.4%)	A1000 (45.8%)	A750 (38.6%)	A500 (26.4%)	A250 (12.2%)
E+A1000		n.s	n.s	n.s	***	***	***	***	***
E+A750	n.s		n.s	n.s	***	***	***	***	***
E+A500	n.s	n.s		n.s	***	***	***	***	***
E+A250	n.s	n.s	n.s		n.s	***	***	***	***
E alone	***	***	***	n.s		***	***	***	***
A1000	***	***	***	***	***		n.s	***	***
A750	***	***	***	***	***	n.s		n.s	***
A500	***	***	***	***	***	***	n.s		***
A250	***	***	***	***	***	***	***	***

^***indicates p < 0.001 and n.s indicates not significant.

To quantify the contribution of voicing, manner, and place cues for consonant perception, we computed the percent information transmitted using information theory equations listed in the Wang and Bilger study (1973). In short, we first set aside consonant syllables in terms of voicing, manner, and place features and then computed the percent correct for each of the three feature-based group consonants. To obtain the percent information transmitted, the percent correct was divided by the total number of consonant syllables presented and then multiplied by 100. The results of these computations with standard errors are shown in Figure 3. For voicing, a significant main effect was found for the listening condition, F(3, 54)=9.85, p<0.001; η²=.33 but not for the group, F(1, 54)=0.85 p=0.37; η²=.04. A significant difference was observed for the bimodal group between the E+A1000 and E+A250 (p=0.001) and between the E+A500 and E+A250 (p=0.04), as indicated by asterisks. A significant difference was observed for the EAS group between the E+A1000 and E+A250 (p=0.01) and between the E+A750 and E+A250 (p=0.03). No significant interaction was found, F(3, 54)=0.35, p=0.78. Manner cue transmitted was not statistically different among the listening conditions, F(3, 54)=2.22, p=0.09; η²=.10 and between the groups, F(1, 54)=1.43, p=0.25; η²=.08. Interaction was also not significant, F(3, 54)=.58, p=0.64. Place cues transmitted was not significantly different among the listening conditions, F(3, 54)=0.49, p=0.69 ; η²=.02 but significantly different between the groups, F(1, 54)=4.40, p=0.04; η²=.02. Significant differences occurred in place cue for three A+E conditions (p<0.05), as indicated by asterisks. No significant interaction was found, F(3, 54)=0.48, p=0.41.

Figure 3:

Mean information transmitted for voicing, manner, and place features between the groups. *** indicates p < 0.001 and ** indicates p < 0.01.

Discussion

Our results show that significant bimodal and EAS benefits occurred regardless of differing configurations of hearing loss and hearing technology (bimodal or EAS). These results suggest that the configurations of hearing loss are not a significant factor for integrating consonant information between acoustic and electric stimulation. The results also suggest that mechanisms used to integrate consonant information may be similar between bimodal and EAS hearing for consonant recognition.

Our result, the significant bimodal and EAS benefits regardless of configuration of hearing loss, suggests that lower-spectral information provided by an acoustic stimulation is positively combined with the information obtained from an electric stimulation, irrespective of configuration of hearing loss. This result is consistent with other researchers who showed a significant correlation with unaided (Sheffield and Zeng, 2012; Yang and Zeng, 2013) and aided (Mok et al., 2006; 2010) hearing thresholds. However, the result is not consistent with Ching et al. (2004) and Gifford et al. (2007) who showed no correlation between bimodal benefit and hearing thresholds. The mixed results suggest that hearing thresholds may play an important role for integration when certain conditions are provided. For example, greater bimodal benefit occurred in bimodal patients who had better residual hearing at frequencies <1000 Hz (Sheffield and Zeng, 2012; Yang and Zeng, 2013), with poorer aided threshold at 1 and 2 kHz, or with no measurable aided thresholds at frequency >4 kHz (Mok et al., 2006; 2010). In contrast, bimodal interference occurred in those who had 4 kHz aided thresholds within audiometer limits (Mok et al., 2006; 2010). These results suggest that the finer spectral structures provided by better residual hearing at >1 kHz negatively interact with mismatched spectral information processed by the electric stimulation. The most apical electrodes are generally positioned in the region of the cochlea with characteristic frequencies of 1 to 2 kHz (Greenwood, 1990) but used to convey information about frequencies as low as 100 Hz. This situation could lead to conflict between information received from both acoustic and electric stimulations. The aided thresholds in the mid-to-high frequencies may therefore account for some of the individual variability in bimodal benefit.

Another main finding from the current study is that the bimodal (range: 13.8% to 15.3%) and EAS benefits (range: 11.5% to 19.8%) are not statistically different across the configurations of hearing loss. Overall, EAS hearing provided a 2.2-4.5% advantage (not statistically significant) over bimodal hearing across the listening condition except for the E+A250 condition in which bimodal hearing provided a 2.6% advantage (not statistically significant) over EAS hearing. These results suggest that integration mechanism is similar between bimodal and EAS hearing which is consistent with one of Fu et al.’s findings (2017). They showed that EAS benefit was approximately 5% greater than the bimodal benefit for the 700 Hz spectral mismatch between input and output frequency range in CI stimulation which is similar to our spectral mismatch (800 Hz). Gifford et al. (2017) compared sentence perception between bimodal and EAS patients with a contralateral HA. They found that EAS benefits were 5.2% to 12.2% greater than the bimodal benefit, depending on lower edge CI cutoff frequency. Due to different listening conditions (EAS with a contralateral HA), along with different testing conditions such as use of semi-diffused noise and varying signal-to-noise ratio, direct comparisons with our results should be cautious.

Our acoustic feature analysis showed that the place cues were significantly better transmitted in EAS hearing (9.2%~16.4% range) than in bimodal hearing for all bilateral listening conditions except for the E+A1000 (bottom panel of Figure 3). This explains a marginally greater EAS benefit (2.2-4.5%) over bimodal benefit. Numerous bimodal studies also showed that place cue is a primary contributor to bimodal benefit. Kong and Braida (2011) showed higher transmission of place information (9%) than voicing (5%), and manner (1.5%). Yoon et al. (2012) also showed higher transmission place cue (5%~20% range) than voicing (3%), and manner (1.2%). Perception of place cues requires more spectral information, particularly in higher frequency ranges (Kong & Mullangi, 2013). Our finding indicates that better spectral information is processed with EAS hearing than with bimodal hearing, even though group difference in bimodal and EAS benefit is not statistically significant.

This simulation study has a few limitations. Although acoustic simulations for the bimodal and EAS hearing provides better controls over subject- and device-related variables, they cannot simulate nonlinear phenomena seen in auditory-nerve responses and broadened auditory filters (Trevino et al., 2010). We used the fixed output (1000-7000 Hz) for CI simulation in both bimodal and EAS hearing, but spectral mismatch generated by shallower insertion depth of electrode array may be greatly varied within and between patients. Different results for bimodal and EAS benefits might be observed in real CI users with various types of residual hearing. The simulation also does not represent the effect of durations of deafness and experience of bimodal and EAS bimodal hearing which are known important factors (Gifford et al. 2017). Different results of the current study from others can in part be explained by the absence of these components in the simulations.

Abbreviations:

A

Acoustic

E

Electric

A+E

Combined acoustic and electric