MECHANISMS OF LOCALIZATION AND SPEECH PERCEPTION WITH COLOCATED AND SPATIALLY SEPARATED NOISE AND SPEECH MASKERS UNDER SINGLE-SIDED DEAFNESS WITH A COCHLEAR IMPLANT

Abstract

Objectives:

This study tested participants with a cochlear implant (CI) in one ear and acoustic hearing in the other ear, to assess their ability to localize sound and to understand speech in collocated or spatially separated noise or speech maskers.

Design:

Eight CI users with contralateral acoustic hearing ranging from normal hearing to moderate sensorineural hearing loss were tested. Localization accuracy was measured in five of the participants using stimuli that emphasized the separate contributions of interaural level differences (ILD) and interaural time differences (ITD) in the temporal envelope and/or fine structure. Sentence recognition was tested in all eight CI users, using collocated and spatially separated speech-shaped Gaussian noise and two-talker babble. Performance was compared with that of age-matched normal-hearing (NH) listeners via loudspeakers or via headphones with vocoder simulations of CI processing.

Results:

Localization improved with the CI, but only when high-frequency ILDs were available. Participants experienced no additional benefit via ITDs in the stimulus envelope or fine structure using real or vocoder-simulated CIs. Speech recognition in two-talker babble improved with a CI in 7 of 8 participants when the target was located at the front and the babble was presented on the side of the acoustic-hearing ear, but otherwise showed little or no benefit of a CI.

Conclusion:

Sound localization can be improved with a CI in cases of significant residual hearing in the contralateral ear, but only for sounds with high-frequency content, and only based on ILDs. In speech understanding, the CI contributed most when it was in the ear with the better signal-to-noise ratio with a speech masker.

INTRODUCTION

Single-sided deafness (SSD) refers to a condition in which people have normal or near-normal hearing in one ear and little or no residual hearing in the other ear. It is estimated that approximately 200 new cases of SSD per million people are diagnosed each year (Baguley et al. 2006), implying about 50,000 new cases per year in the US adult population alone. Despite having normal hearing in one ear, patients with SSD find it challenging to localize sounds and to understand speech in noisy environments, especially when sound originates on the side of the deaf or “poorer” ear (Bess et al. 1986; Sargent et al. 2001; Lieu 2004).

Most sound localization is mediated through binaural auditory input. Localization is typically achieved in normal-hearing listeners by combining information from interaural time differences (ITDs), interaural level differences (ILDs), and monaural high-frequency spectral cues produced by the head, torso, and pinnae (Blauert 1997). For broadband sounds in anechoic conditions, the low-frequency (< 1500 Hz) ITDs conveyed via the temporal fine structure (TFS) – rapid oscillations in the stimulus waveform – dominate localization in the horizontal plane (Kistler & Wightman 1992). However, when low-frequency cues are not available, localization can be achieved via ILD cues and by ITD cues in the temporal-envelope fluctuations of high-frequency complex sounds (Macpherson & Middlebrooks 2002). Most SSD patients experience substantial deficits in localizing sounds in the horizontal plane (Newton 1983; Slattery & Middlebrooks 1994). The magnitude of their localization errors are generally greatest when sounds originate on the side of the poorer ear, and sounds presented on the side of the poorer ear tend to be localized to the better-ear side (Angell & Fite 1901; Jongkees & Van der Veer 1957; Gatehouse & Cox 1972; Gatehouse 1976). Van Wanrooij and van Opstal (2004) found that SSD patients can use monaural level and spectral cues to localize sound somewhat when the sound originates from the better-ear side, but that accuracy degrades in unfamiliar listening environments.

The loss of function of one ear also affects the ability of SSD patients to understand speech in noisy environments in at least three ways. First, the head-shadow effect at high frequencies means that the signal-to-masker ratio (SMR) will generally be better on one side than the other when the speech and masker come from different locations. If the speech is located on the side of the impaired ear, the SMR at the normal ear will be reduced (Zurek 1993). Second, the loss of one ear results in an inability to use binaural interactions, including summation of information from the two ears and interaural timing and phase differences, to produce binaural masking release (Durlach 1963). Third, poorer localization may lead to a reduced ability to use perceived spatial differences between sources to assist in the segregation of the target speech from an interfering masker, especially when the target speech and masker are similar, leading to “informational masking” (Freyman et al. 1999; Brungart et al. 2001).

In principle, patients with SSD could take advantage of better-ear effects by orienting their head to ensure that the target speech is always on the side of the normal ear and/or that the masker is on the side of the impaired ear. In practice, however, such orientation may not always be possible. To address this problem, audiologists may fit SSD patients with devices that transmit signals from the impaired ear to the normal ear using contralateral routing of signals (CROS) hearing aids or bone-anchored hearing aids (BAHA). However, these devices suffer from some limitations. First, the devices may decrease the SMR when speech originates from the normal-ear side and the masker originates from the impaired-ear side, thereby reducing the better-ear advantage. Second, they do not restore hearing in the poorer ear and so do not provide any usable binaural cues for sound localization or spatial masking release (Agterberg et al. 2018). These limitations may explain why many SSD patients reject such devices (Pennings et al. 2011; Lin et al. 2006).

Recently, some patients with SSD have received a cochlear implant (CI) in their poorer ear. Many have sought a CI for tinnitus relief (Mertens et al. 2016; Arndt et al. 2017; Firszt et al. 2017; Ramos Macías et al. 2018), while others have sought this treatment option for hearing restoration (Sladen et al. 2017a,b; Sladen et al. 2018). Research suggests that localization accuracy for fixed single-source sounds improves after implantation and is superior to performance with rerouting devices (Hassepass et al. 2013a; Blasco & Redleaf 2014; van Zon et al. 2015; Kitterick et al. 2016; Arndt et al. 2011; Hansen et al. 2013; Stelzig et al. 2011), but that motion perception (both perceived direction and degree angle of movement) with the CI remains impaired (Litovsky et al. 2018). The effects on speech perception in noise are less clear, with a few studies suggesting a small benefit (Blasco & Redleaf 2014; Cabral Junior et al. 2016; Kamal et al. 2012; Vlastarakos et al. 2014; Sladen et al. 2017b) but other reviews and studies failing to identify significant improvements (Kitterick et al. 2016; van Zon et al. 2015; Litovsky et al. 2018). A recent FDA clinical trial followed SSD+CI subjects for one year and found that binaural benefits consistently appear in localization and spatialized speech in noise tests (with the masker originating from the 0° azimuth or 90° on the NH side) at 1 month but asymptote after 3 months (Buss et al. 2018).

Little is known about the specific cues used by SSD patients with a CI (SSD+CI) to localize sound and potentially improve speech understanding in spatial environments. Preliminary evidence suggests that SSD+CI listeners primarily rely on ILD information to localize sound (Dorman et al. 2015), similar to what has been observed with bilateral CI users (Seeber & Fastl 2008), whose ITD sensitivity is generally poor (Long et al. 2006; Van Hoesel et al. 2009; Aronoff et al. 2010; Noel & Eddington 2013). However, Dorman’s study used lowpass-filtered, highpass-filtered, and unfiltered noise, which did not allow them to isolate each binaural mechanism and estimate its unique contribution to sound source localization ability in SSD+CI patients. For instance, lowpass-filtered noise has both TFS and temporal-envelope cues that could be used for localization, and highpass-filtered noise has both level and temporal-envelope cues, at least in terms of onset and offset (Dorman et al. 2015). In a study involving bimodal hearing (patients with a CI in one ear and a hearing aid in the nonimplanted ear), 4 of 8 patients were able to detect ITDs of 91-341 µs between acoustically and electrically presented pulse trains (Francart et al. 2009). A more recent study from that group shows that, when frequency-specific delays are added to channels, ITD sensitivity can be not only measured in SSD+CI patients but also used to roughly determine the cochlear region that a given electrode stimulates (Francart et al. 2018). These findings suggest that some ITD sensitivity may also be found in SSD+CI patients, although it is not known whether this sensitivity is useful for sound localization or speech perception in spatially separated noise.

The aim of this study was to explore the mechanisms underlying localization and speech perception in spatially separated maskers in SSD+CI users. The first experiment expanded on Dorman et al.’s (2015) study; it involved the localization of different sounds that were designed to emphasize different spatial cues, including ILDs, ITDs based on TFS cues, and ITDs based on temporal-envelope cues. The second experiment measured speech intelligibility under different conditions of masker-target spatial separation. Different types of maskers were used to emphasize either energetic masking or informational masking, where listeners are more likely to confuse the target voice with voices in the masker. It was hypothesized that if the CI improves localization then it may also partially restore the benefits of spatial hearing, particularly in situations where spatial masking release relies on a release from informational masking due to a perceived spatial separation between the target and masker.

Age-matched normal-hearing listeners completed each of these experiments in the sound field and under headphones. Performance in the sound field was measured as a reference for the most benefit that SSD patients could potentially receive with a CI. In addition, vocoder simulations were performed under headphones with non-individualized head-related transfer functions (HRTFs), recorded in anechoic conditions, to determine whether acoustic hearing in one ear and a simulation of a CI in the other produce the same pattern of results observed in actual SSD+CI patients.

EXPERIMENT 1: LOCALIZATION

Methods

Participants

Five subjects with acquired SSD and a CI (SSD+CI) participated. Two of the SSD+CI listeners used MED-EL implants with “Fine-Structure Processing” (FS4-p in SSD-12 and FS4 in SSD-16) enabled in the four most apical electrodes of their clinical maps. This type of processing is designed to maintain some information about the TFS in the timing of the electrical pulses presented to the most apical electrodes. The other three listeners used Cochlear devices. See Table 1 for further details concerning the SSD+CI listeners. Pure-tone air conduction thresholds at all audiometric (octave) frequencies between 250 Hz and 8 kHz were greater than 70 dB HL in the CI ear and less than 30 dB HL, on average, at 0.5, 1, 2, and 4 kHz in the unimplanted ear (Figure 1). Five age-matched control subjects were also enrolled (see Table 1). The control subjects had normal hearing bilaterally, defined as pure-tone thresholds of 20 dB HL or less at all audiometric frequencies between 250 Hz and 8 kHz.

Table 1.

SSD+CI listener demographics.

SSD Subject ID	SSD+CI age	NH control age	Implant side	Etiology	Duration of SSD (mos)	Duration of implant use (mos)	CI manufacturer and model
15	33	30	Left	Meningitis	7	6	Cochlear Nucleus CI 422
12	39	36	Left	Unknown	18	18	MED-EL Flex 24; FS4-p
16	40	36	Right	Unknown	11	16	MED-EL Flex 28; FS4
3	50	55	Left	Otosclerosis	44	11	Cochlear Nucleus CI 422
1	62	62	Right	Onset following hysterectomy	24	7	Cochlear Nucleus Hybrid L24
4	52	49	Left	Unknown	38	12	Cochlear Nucleus CI 522
10	53	55	Right	Unknown	16	9	MED-EL Flex 28
11	50	49	Right	Acoustic Neuroma	70	36	MED-EL Flex 28

Fig. 1.

Audiograms of acoustic ear in SSD+CI patients. Numbers represent individual SSD subjects (see Table 1). Red and blue numbers represent hearing thresholds in patients’ non-implanted right and left ears, respectively. The black solid line represents average hearing thresholds in the acoustic ear across listeners. Symbol numbers that fall in the gray shaded region represent hearing thresholds that fall outside the normal range.

Stimuli

Listeners’ ability to localize sounds was tested with several stimuli, designed to convey different binaural cues (Tables 2 and 3). The low-frequency stimuli were designed to provide interaural time difference (ITD) cues (either in the TFS only, or in both the TFS and envelope) but little or no interaural level difference (ILD) cues; the high-frequency stimuli were designed to provide ILD cues, and in some cases also ITD cues in the temporal envelope. Low- and high-pass filtering of the speech was achieved with 4^th-order Butterworth filters. The non-speech stimuli were 300 ms in duration, including 100-ms onset and offset ramps to reduce the potential envelope cues associated with onset and offset. The average duration of each word from the NU-6 words was 463 ms (SD 72 ms).

Table 2.

Localization stimulus details. ITD represents interaural timing differences, ILD represents interaural level differences, and TFS represents temporal fine structure.

Freq Content	Stimulus Name	Acronym	Basic Details	Cues Available
Low	Unmodulated Tone	ULT	200-Hz pure tone	ITDs in TFS
Low	Modulated Tone	MLT	200-Hz pure tone; amplitude modulated by 40-Hz sinusoid, 100% modulation depth	ITDs in TFS, ITD envelope information via amplitude modulations
Low	Speech	LP	NU-6 words lowpass-filtered at 640 Hz	ITDs from TFS and temporal envelope
High	Unmodulated Complex	UHC	Inharmonic complex tone; pure tone frequencies depended on subject group SSD+CI participants:, used channel CFs above 1500 Hz in individual current maps (Table 3) NH listeners: used CFs of the standard MED-EL clinical map above 1500 Hz (Table 3)	ILDs, little to no ITD information in TFS, little to no ITD information from temporal envelope due to lack of temporal modulation or temporal interactions between the components
High	Modulated Complex	MHC	Same as high-frequency inharmonic complex tone; amplitude-modulated by 40-Hz sinusoid, 100% modulation depth	ILDs, ITDs in temporal envelope via 40-Hz amplitude modulation
High	Speech	HP	NU-6 words highpass-filtered at 1500 Hz	ILDs, ITDs from high-frequency temporal envelope
Broadband	Speech	U	NU-6 words	All natural binaural cues (ILDs, ITDs in TFS, ITDs in stimulus envelope)

Table 3.

Pure tone components used for modulated and unmodulated inharmonic complex tone. For the SSD+CI participants (listed in order of ascending age), the tone frequencies within the complex were selected to coincide with the center frequencies above 1500 Hz of their individual current maps.

Subject	PT1	PT2	PT3	PT4	PT5
SSD+CI 15	2184	2869	3805	4991	6486
SSD+CI 12	1977	2713	3858	5238	7335
SSD+CI 16	2227	3064	4085	5656	7352
SSD+CI 3	2184	2869	3805	4991	6486
SSD+CI 1	2184	2869	3805	4991	6486
NH	2207	2994	4045	5450	7331

All sounds were presented at a nominal root-mean-square (rms) level of 60 dB SPL, as measured at the location corresponding to the position of the listener’s head. This level was sufficiently high to be in the middle of the dynamic range of most SSD+CI participants’ programmed maps, and sufficiently low to fall well below the acoustic detection thresholds of the SSD+CI participants’ implanted ear. The level of all stimuli was roved by ±10 dB around the nominal level of 60 dB SPL on every presentation to reduce the reliability of monaural loudness cues (Van Wanrooij & Van Opstal 2005).

To avoid the detection of low-frequency distortion products potentially generated by the high-frequency stimuli, a threshold equalizing noise (TEN; Moore et al. 2000) was added to all the highpass-filtered non-speech conditions, and presented from all speakers in the horizontal plane (19 total) simultaneously, with the same (correlated) noise presented from each speaker. At the listener’s position in the sound booth, the level of the TEN in the equivalent rectangular bandwidth (ERB) of the auditory filter around 1 kHz was 10 dB below the level per component of the modulated and unmodulated high complex stimuli and was roved in the same way as the stimuli. This level was chosen because it was sufficiently intense to mask any low-frequency distortion products but not so intense as to mask the target stimuli themselves. The low- and high-pass cutoff frequencies of the TEN were 20 and 1500 Hz, respectively. The TEN was gated on and off using 100-ms raised-cosine ramps 1 s before and after stimulus onset and offset, respectively, for a total duration of 2.3 s. No masking noise was played in the speech conditions.

Procedure

Sound-field presentation.

The SSD+CI and NH listeners were tested individually in a large (3.05 m by 3.96 m by 2.59 m) sound-attenuating chamber with 10 cm foam on all walls to reduce reverberation. Speakers (Anthony Gallo Acoustics: A’Diva ti) were located along the horizontal plane approximately level with the participant’s head at a spacing of 10° from −90° to +90° azimuth. The speakers were placed along the walls of the chamber, with distances from the listener’s head ranging from 1.4 to 2.4 m, and were equalized (in terms of time delay, level, and spectrum) to match their outputs at the point corresponding to the center of the listener’s head to that of the frontal speaker (0°), which was 1.8 m from the listener’s head. On each trial a stimulus was presented from one of the speakers with equal a priori probability, and the participant was asked to indicate from which one of the 19 speakers they heard the sound source via a virtual button on a computer screen with a display of the speaker arrangement.

Each type of stimulus was played 5 times from each of the 19 speakers in the front horizontal plane for a total of 95 trials per stimulus and listening condition (monaural or binaural). One stimulus type was presented per block. The location of the stimulus varied from trial to trial and the presentation order was determined randomly at the beginning of each block. One block of each stimulus type was tested before any other was repeated. The SSD listeners performed this task with the acoustic-hearing ear only (i.e., with the CI turned off) and with the combination of the acoustic-hearing ear and a CI. The order of listening condition and general stimulus category (tonal or speech stimuli) was counterbalanced across subjects. Within each category, the stimulus presentation order was counterbalanced using a Latin squares design.

The NH listeners completed this task with and without masking noise in one ear to simulate SSD. Three types of noise maskers were used: an octave-band noise centered at 300 Hz for low-frequency pure-tone stimuli, a bandpass-filtered TEN with cutoff frequencies of 1700 and 8400 Hz for the high-frequency complex-tone stimuli, and noise with the same long-term average spectrum as the unfiltered NU-6 words for the speech stimuli. The masking noise was presented via an Etymotic Research ER1 insert earphone to the ear corresponding to the poorer ear of each NH listener’s age-matched SSD listener and was presented at a fixed level of 75 dB SPL, which exceeded the maximum possible stimulus level (70 dB SPL) by 5 dB but still avoided loudness discomfort.

Headphone and vocoder presentation.

The five NH listeners also completed a simulation of the localization task under headphones. After non-individualized (KEMAR) head-related transfer functions (Gardner & Martin 1995) were applied to the stimuli, they were presented either unprocessed or via a tone-excited envelope vocoder to simulate aspects of CI processing (Dorman et al. 1998; Whitmal et al. 2007). In one condition, the unprocessed sounds were presented to one ear (monaural condition, simulating SSD); in the other condition, one ear was presented with the unprocessed sounds, while the other ear was presented with the vocoded sounds (simulating SSD+CI).

A tone-excited envelope vocoder was used with 16 frequency subbands, equally spaced on a logarithmic scale with center frequencies between 333 and 6665 Hz, as shown in Table 4. This spacing corresponds to the standard clinical map for Advanced Bionics devices and has been used in several previous vocoding studies (e.g., Oxenham & Kreft 2014; Oxenham & Kreft 2016). The temporal envelope from each subband was extracted using a Hilbert transform, and then the resulting envelope was low-pass filtered with a fourth-order Butterworth filter and a cutoff frequency of 50 Hz. Although this cutoff frequency is much lower than those applied in most current CI speech processors, 50 Hz was used to reduce the possibility that vocoding produced spectrally resolved components via amplitude modulation.

Table 4.

Center frequencies for 16 sub-bands of Advanced Bionics cochlear implant. These sub-bands were used for the tone vocoder simulation of the speech understanding in noise task.

Channel	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
CF	333	455	540	642	762	906	1076	1278	1518	1803	2142	2544	3022	3590	4264	6665

The effects of current spread were incorporated as described by Oxenham and Kreft (2014) by including the summation of the intensity envelopes from neighboring channels with an attenuation corresponding to 12 dB/octave. Finally, the resultant temporal envelopes were used to modulate pure tones at the center frequencies of each subband before the modulated carriers were summed and presented to the participants.

Participants in the vocoder simulations were seated in a small sound-attenuating chamber and stimuli were presented at a nominal level of 60 dB SPL via Sennheiser HD650 headphones. All other aspects of the experiment, including the stimuli, the response screen, and the task, were the same as in the sound-field conditions. The participants always completed this task after completing the sound-field task.

Data Analysis

Three measures were used to quantify the accuracy of sound localization. The first was the root-mean-squared (rms) error, defined as the square root of the mean squared difference in degrees between the actual location of the sound source and the reported sound location. This measure provides a combined estimate of systematic bias and response variability, as has been used in many previous studies (Macpherson & Middlebrooks 2002; Middlebrooks 1992). The second measure was the systematic localization bias, derived by calculating the mean (signed) difference between the actual and perceived angle for each source. This measure separates the variability from the systematic response bias to determine whether the responses were biased towards any given direction. Finally, to evaluate whether errors were location dependent, the results from each location were evaluated separately using confusion matrices.

Statistical analysis was performed using repeated-measures analyses of variance (ANOVA). Three separate analyses were performed for rms error and bias. First, the non-speech conditions were organized according to whether they included high frequencies (high-frequency complex tones) or not (low-frequency pure tones) and whether they contained amplitude modulations or not. The difference in performance between monaural and binaural conditions served as the dependent variable. This was done for SSD+CI listeners only. By analyzing the non-speech conditions separately, it was possible to isolate each binaural hearing mechanism using similar stimuli and thus estimate the contribution of each mechanism to localization performance.

The next two analyses compared performance in the SSD+CI and vocoder groups, as these two groups were tested under the most comparable conditions. A repeated-measures ANOVA was performed for monaural performance alone with listener group as the between-subjects variable and stimulus (including both speech and non-speech stimuli) as a within-subjects factor. Finally, a repeated-measures ANOVA was performed on the difference in performance between monaural and binaural conditions with stimulus type (speech and non-speech) as a within-subjects factor and group as the between-subjects variable. In those cases where within-subjects effects did not meet Mauchly’s test of sphericity, Greenhouse-Geisser corrections were used.

Results

Figure 2 shows the individual and average performance of the five SSD+CI participants in terms of localization error. Dark blue bars represent performance without the CI and light blue bars represent performance with the CI. The first analysis was performed to determine whether and how the addition of a CI affected localization errors, based on the spectral and temporal cues available in the stimulus. To simplify the analysis, the non-speech conditions were organized according to whether they included only high frequencies (providing access to ILDs; UHC and MHC) or only low frequencies (providing access to ITDs in the temporal fine structure; ULT and MLT) and whether they provided ITD cues in the temporal envelope by containing amplitude modulations (MLT and MHC) or not (ULT and UHC). The dependent variable was the difference in performance (rms error) between monaural and binaural conditions. The ANOVA revealed a significant main effect of frequency content [F(1,4)=17.7, p=0.014, η_p²=.815], but no effect of amplitude modulation [F(1,4)=0.76, p=0.43, η_p²=0.16], and no interaction [F(1,4)=0.84, p=0.41, η_p²=0.17]. Paired t-tests on the difference between monaural and binaural rms errors separated by frequency content (low vs. high) and averaged across modulation type (with or without modulation in the temporal envelope) showed that the rms error decreased significantly with the addition of the CI in the high-frequency tone conditions [decrease in error 26°; t(4)=4.41, p=0.012, d=1.47], whereas the change in error was not significant in the low-frequency tone conditions [increase in error 12°; t(4)=−2.19, p=0.093], d=−0.96]. In summary, binaural localization errors were smaller than unilateral (acoustic-hearing ear only) localization errors when high-frequency information was present. The addition of the CI did not improve performance in conditions relying on low-frequency TFS, and no further improvement was found for either the low- or high-frequency stimuli when temporal-envelope ITD information was added via amplitude modulations. Therefore, the CI improved performance only when SSD+CI listeners had access to high-frequency ILDs.

Fig. 2.

Localization error (root-mean-square error in degrees) for SSD+CI listeners in the sound field, organized by stimulus group. Stimulus type appears on the x-axis: unmodulated low tone (ULT), modulated low tone (MLT), low-pass filtered speech (LP), unmodulated high complex tone (UHC), modulated high complex tone (MHC), high-pass filtered speech (HP), unfiltered speech (U). Root mean square (rms) error in degrees appears on the y-axis. Dark blue bars represent performance with the CI turned off and removed. Light blue bars represent performance with the CI turned on. Error bars in the final panel represent ±1 standard error.

The data from the speech conditions (LP, HP, and U in Fig. 2) appear consistent with this interpretation: the average rms error was 8° higher, 30° lower, and 17° lower in the binaural condition compared to the monaural condition in the lowpass, highpass, and unfiltered stimulus conditions, respectively. However, these trends failed to reach statistical significance, as there was no significant effect of speech condition (LP, HP, or U) on the difference in rms error between the monaural and binaural conditions [F(1.04, 4.17)=5.76, p=0.071].

Figure 3 replots the mean data from the SSD+CI group from Figure 2 (Figure 3B), along with the mean data from the NH group listening in the sound field (Figure 3A) and the NH group listening over headphones with the vocoded stimuli (Figure 3C). The intent of including NH listeners in the sound field was to provide a baseline measure for the maximum (though unrealistic) amount of benefit SSD listeners could obtain with the CI activated. The vocoder simulation, on the other hand, was designed to approximate localization with and without a CI using non-individualized HRTFs. Therefore, only the SSD+CI and vocoder simulation groups were compared in the statistical analysis, which included all speech and non-speech stimuli.

Fig. 3.

Localization error (root-mean-square in degrees) organized by stimulus group for NH listeners in the sound field (first panel), SSD listeners in the sound field (second panel), and NH listeners in the vocoder simulations (third panel). Stimulus type appears on the x-axis: unmodulated low tone (ULT), modulated low tone (MLT), low-pass filtered speech (LP), unmodulated high complex tone (UHC), modulated high complex tone (MHC), high-pass filtered speech (HP), unfiltered speech (U). Dark blue bars represent the monaural condition. Light blue bars represent the binaural condition. For the monaural condition, one ear was masked for NH listeners in the sound field, and the CI was removed for SSD listeners, and no sound was delivered to one ear in the NH vocoder simulation. For the binaural condition, masking noise was removed for NH listeners in the sound field, CI was replaced and turned on for SSD listeners, and vocoded stimuli were delivered to one ear with unprocessed sound presented to the other ear in the NH listeners under headphones. Improvement in localization error (binaural condition subtracted from monaural condition) for each stimulus group appears below the x-axis. Error bars represent ±1 standard error.

A repeated-measures ANOVA on the difference between the monaural and binaural rms localization error, with group (SSD+CI or vocoder simulation) as a between-subjects factor, showed a significant main effect of group [F(1,8)=12.30, p=0.008, η_p²=0.61], with the average improvement in the binaural conditions being 17° larger for the NH vocoder group than for the CI+SSD group (p=0.023). There was also a significant main effect of stimulus type [F(6,48)=18.70, p<0.001, η_p²=0.70]. The boxes at the bottom on Figure 3 demonstrate this effect by showing the difference in monaural and binaural localization error across stimuli in the low-frequency, high-frequency, and broadband stimulus groups. Although the interaction between group and stimulus type was statistically significant [F(6,48)=2.91, p=0.017, η_p²=0.27], none of the individual contrasts reached significance.

Figure 4 shows the localization bias for the individual SSD+CI participants, along with the average data. In the low-frequency conditions, where localization performance was very poor, adding the CI led to a bias towards the side of the CI. In the high-frequency conditions, where performance improved with the addition of the CI, the bias was reduced and tended towards zero. Figure 5 shows the mean localization bias for the CI+SSD and NH groups (both with and without the vocoder). The trend of hearing sounds towards the (simulated) CI was not observed with the vocoded NH group.

Fig 4.

Localization bias in degrees for SSD+CI listeners in the sound field, organized by stimulus group. Stimulus type appears on the x-axis: unmodulated low tone (ULT), modulated low tone (MLT), low-pass filtered speech (LP), unmodulated high complex tone (UHC), modulated high complex tone (MHC), high-pass filtered speech (HP), unfiltered speech (U). Dark blue bars represent the monaural condition in which the CI was removed. Light blue bars represent the binaural condition in which the CI was attached and turned on. Error bars in the final panel represent ±1 standard error.

Fig. 5.

Localization bias in degrees organized by stimulus group for NH listeners in the sound field (first panel), SSD listeners in the sound field (second panel) and NH listeners under vocoder simulation (third panel). Stimulus type appears on the x-axis: unmodulated low tone (ULT), modulated low tone (MLT), unmodulated high complex tone (UHC), modulated high complex tone (MHC), low-pass filtered speech (LP), high-pass filtered speech (HP), unfiltered speech (U). Dark blue bars represent the monaural condition. Light blue bars represent the binaural condition. For the monaural condition, one ear was masked for NH listeners in the sound field, the CI was removed for SSD listeners, and no sound was delivered to one ear in the NH vocoder simulation. For the binaural condition, masking noise was removed for NH listeners in the sound field, the CI was turned on for SSD listeners, and vocoded stimuli were delivered to one ear for NH listeners under headphones. Change in localization bias (binaural condition subtracted from monaural condition) for each stimulus group appears below the x-axis. Error bars represent ±1 standard error.

Figure 6 confirms the impressions provided by the localization accuracy and bias in Figures 2–5, by plotting actual location against perceived location for SSD+CI listeners when the CI is turned on. A confusion matrix for NH listeners’ binaural localization performance in the sound field for the unfiltered (U) word stimulus is included in the bottom right panel for comparison. Clearly, SSD listeners’ responses were more variable than those of the NH listeners and tended to fall along the diagonal only for the stimuli that included some high-frequency content. For low-frequency stimuli, SSD listeners tended to perceive sounds as originating on the side of their implant. The NH listeners, on the other hand, showed highly accurate and precise responses when listening with two ears, as expected.

Fig. 6.

Confusion matrices for each stimulus type in SSD+CI listeners in the binaural listening condition. The confusion matrix for NH listeners in the unfiltered binaural listening conditions was included (lower right hand plot) for comparison. For each matrix, ordinate values represent the actual stimulus location. Abscissa values correspond to position that listeners’ perceived the stimulus to be. Negative values represent degrees on the poorer ear (CI) side and positive values represent degrees on the acoustic ear side. Warmer colors correspond to higher response rates for a given actual/perceived location combination across listeners; cooler colors correspond to lower response rates. Stimulus type appears in white text in the upper right hand corner of each plot. Stimulus types are labeled as follows: unmodulated low tone (ULT), modulated low tone (MLT), unmodulated high complex tone (UHC), modulated high complex tone (MHC), low-pass filtered speech (LP), high-pass filtered speech (HP), unfiltered speech (U).

Discussion

In general, it appears that SSD listeners were able to integrate information from the CI and their NH ears to evaluate ILDs (Figure 2). Without the CI, SSD listeners tended to perceive high-frequency stimuli on the acoustic ear side and low-frequency stimuli as occurring from 0° azimuth. When the CI was switched on, localization errors were ~30° for stimuli with high-frequency energy and slightly biased toward the acoustic ear. Stimuli limited to low-frequency energy were biased entirely toward one ear or the other, depending on the listener (Figure 4), and localization did not improve with the addition of the CI. The results are consistent with those of Dorman et al. (2015), which also suggested that SSD+CI listeners rely primarily on ILD cues to localize sound.

Our SSD listeners’ post-implant localization ability is similar to that found in previous studies using broadband stimuli such as words (Firszt et al. 2012), sentences (Arndt et al. 2011; Hassepass et al. 2013b; Arndt et al. 2017), noise (Dillon et al. 2017; Jacob et al. 2011; Mertens et al. 2017), and environmental sounds (Hansen et al. 2013). In general, these studies have shown that CI users with significant residual monaural hearing localize sound better with the addition of a CI than when they are unaided or aided with devices that route signals to the acoustic ear.

In summary, the results confirm previous findings that SSD+CI listeners can use ILDs to some extent. The current study also extends previous work by showing that ITDs in either the temporal fine structure or temporal envelope do not contribute to SSD+CI listeners’ localization abilities. None of the SSD+CI listeners were able to use to low-frequency TFS information, including the two subjects (SSD-12 and −16) who had “Fine-Structure Processing” enabled in the four most apical channels of their clinical maps. It is possible that some of the differences observed between NH listeners in the vocoder simulation and SSD+CI listeners are due to the somewhat unnatural physical cues produced by non-individualized HRTFs in the NH group; HRTFs were measured in a room set up that was different from the one used in this experiment. Furthermore, NH listeners are not accustomed to listening to vocoded stimuli in conjunction with normal acoustic stimuli and so may experience less perceptual fusion than SSD+CI users, who have had time to adjust to the stimulation method. Overall, therefore, it appears that acute presentation of vocoder simulations of a localization task in NH listeners will not necessarily provide accurate predictions of performance by actual SSD+CI patients.

EXPERIMENT 2: SPEECH RECOGNITION WITH SPATIALLY COLOCATED AND SEPARATED NOISE AND SPEECH MASKERS

Methods

Participants

Three additional SSD+CI listeners and three additional normal-hearing age-matched controls were tested in experiment 2, along with the participants from experiment 1, for a total of 8 listeners in each group. Participant 4 (who only participated in this experiment) used a hearing aid in the contralateral ear during testing, as he uses the hearing aid in his daily life. Given the amount of hearing loss in this subject’s acoustic-hearing ear, he technically should be classified as bimodal (as opposed to SSD). This patient reports that he relies mostly on his hearing aid for speech understanding and uses his CI for sound localization. As in experiment 1, the NH participants completed the task both under sound-field and headphone (vocoded) conditions.

Stimuli and Procedure

Harvard IEEE sentence lists spoken by a single female talker were used as the target speech (Rothauser et al. 1969). These materials provided sufficient lists to test all the conditions under consideration without repetition in SSD+CI participants and without repeating each sentence more than once in NH participants (because NH participants completed both sound-field and headphone versions of the experiment). The target speech was always presented at an rms level of 55 dB SPL, and the masker level was varied to achieve the desired SMR. The masker was either time-reversed two-talker female babble, generated by concatenating random segments from the same IEEE materials as the target speech, or Gaussian noise, spectrally shaped to match the long-term average spectrum of the target sentences. These two maskers were chosen to produce high and low contributions of informational masking, respectively (Freyman et al. 2004; Balakrishnan & Freyman 2008). The SSD+CI listeners were tested in two listening conditions: with the acoustic-hearing ear only and with the combination of the acoustic-hearing ear and the CI. In the sound-field conditions, the NH control subjects were tested with unilateral masking noise to simulate unilateral listening conditions, and without unilateral masking to allow full use of binaural cues. In the headphone conditions, the NH control subjects were tested either unilaterally or bilaterally with the HRTF-filtered sound to one ear and the HRTF-filtered and then vocoded sound to the other ear.

The speech and masker were presented at five different SMRs. The SMRs were selected individually based on the results from pilot testing in each combination of masker type, listening status (CI on or off), and spatial configuration that was tested in the main experiment (12 total for SSD+CI listeners, 24 total for NH listeners). The pilot testing involved an adaptive 1-up 1-down procedure, where the SMR was increased if the participant correctly reported less than 4 of 5 keywords and was decreased otherwise. The SMRs tested in the main experiment corresponded to +3, 0, −3, −6, −9 dB, relative to the SMR obtained from the pilot adaptive procedure for each participant. In the main experiment, one SMR was tested for per list, with each list containing 10 sentences. Listeners entered their responses using a computer keyboard and were encouraged to guess as many words as possible. The proportion of keywords (out of five per sentence) correctly identified in each sentence was used to assess performance.

Performance was measured in three spatial configurations: speech and masker co-occurring from 0° azimuth (S₀N₀), speech at 0° and masker at 60° on the (actual or simulated) CI side (S₀N_CI), and speech at 0° and masker at 60° on the acoustic ear side (S₀N_NH). An angle of 60° was chosen because this angle has been found to produce the largest head-shadow effects (Culling et al. 2012). For each SSD listener, 60 conditions were tested (2 masker types, 3 spatial configurations, and 5 SMRs with and without the CI turned on). For each NH listener, 60 conditions were tested in the sound field (2 masker types, 3 spatial configurations, 5 SMRs, 2 listening modes) and then the 60 conditions were repeated under headphones for a total of 120 conditions. One sentence list was used per condition and SMR (i.e., 5 lists per condition when combining across SMRs).

Data Analysis

A speech reception threshold (SRT) was determined for each listener in each condition by logistic regression using maximum-likelihood estimation. The following equation was used to predict the SRT, defined as the SMR at which listeners correctly report 50% of the words in the sentences:

$$PC=\frac{1}{1+e^{-s(SMR-SRT)}}$$

where PC is the proportion of correct responses at a given SMR, SRT represents the SMR at PC=50%, and s represents the slope of the function where s is 1/4 * slope at PC=50% (Smits et al. 2004). SRT and s were treated as free parameters. A starting set of parameters (the mean SMR used in the experiment and 1 for the value of s) were then used within a minimization routine to find the parameters that produced the largest log-likelihood ratio.

Statistical analysis was performed using repeated-measures ANOVAs with one between-subjects factor (group; SSD+CI sound-field and NH vocoder simulation only) and three within-subjects factors (monaural/binaural listening, masker type, and masker location). The dependent variable was SRT in dB. In cases where within-subjects effects did not meet Mauchly’s test of sphericity, Greenhouse-Geisser corrections were used. Follow up analyses were completed using paired t-tests.

Speech recognition results from the individual SSD listeners (represented by the subject identification number from Table 1) are plotted along with the mean data in Figure 7; listeners with some hearing loss in the acoustic ear are represented by gray numbers and listeners with clinically normal hearing (no audiometric thresholds greater than 20 dB HL) are represented by black numbers. Lower SRTs correspond to better speech understanding. The data show some of the expected effects of spatial separation: for both the noise and two-talker maskers, SRTs were generally lowest (best) when the masker was presented to the side of the CI (S₀N_CI), and highest (worst) when the masker was presented to the side of the acoustic ear (S₀N_NH).

Fig. 7.

Speech recognition thresholds for SSD+CI listeners. Speech and noise location appear on the x-axis and SRT in dB appears on the y-axis. Blue bars represent conditions where the masker was speech-shaped noise (SSN). Red bars represent conditions where the masker was time-reversed two-talker babble (TTB). Darker shaded bars represent the monaural condition. Lighter shaded bars represent the binaural condition. Individual data appear as numbers where the numbers correspond to subject identification number from Table 1. Purple numbers represent subjects with hearing loss and black numbers represent subjects with normal hearing. Error bars in the final panel represent ±1 standard error.

A repeated-measures ANOVA using the SSD listeners alone, with SRT as the dependent variable and factors of CI status (on or off), masker type (noise or two-talker interferer), and masker position (CI side, middle, acoustic side) revealed no main effect of CI status (on or off) [F(1,7)=2.24, p=0.18, η_p²=0.24], but a significant interaction between CI status and masker type [F(1,7)=7.81, p=0.03, η_p²=0.53], reflecting a trend for more of an effect of CI status in the two-talker babble than in the noise. A significant main effect of masker type was observed [F(1,7)=451, p<0.001, η_p²=0.99], showing that two-talker, same-gender babble (average SRT: −0.47 dB SMR) was a more effective masker than speech-shaped noise (average SRT: −10.8 dB SMR). The main effect of masker position was statistically significant [F(2,14)=53.35, p<0.001, η_p²=0.88], showing that SRT decreased on average as the masker moved from the acoustic ear to CI ear side. In addition, the interaction between masker type and masker position was statistically significant [F(1.17,8.22)=10.68, p=0.01, η_p²=0.60], showing that SRT increased at a faster rate in two-talker babble than stationary noise when the noise source moved from the CI side to the acoustic ear side. The interaction between CI status and masker position just failed to reach significance [F(2,14)=3.40, p=0.06, η_p²=0.33]. Finally, the three-way interaction between CI status, masker type, and masker position failed to reach significance [F(2,8)=1.33, p=0.32, η_p²=0.25].

On average, trends towards benefit provided by the CI (compare dark and light shaded bars; without CI and with CI listening conditions, respectively) were observed in all spatially separated conditions, except when noise (blue bars) originated on the CI side. Notably, the CI made a difference in the most challenging listening situation: when babble originated on the acoustic-hearing side (S₀N_NH), 7 of 8 listeners showed an improvement in SRT, with an average improvement of 3.5 dB. Although the main effect of CI status and the interaction between CI status and masker position failed to reach significance, it was planned a priori to compare performance with and without the CI turned on when the maskers originated on the acoustic-hearing side, as this configuration was expected to most clearly demonstrate any observable benefit, based on the improved SMR at the CI. A paired t-test (monaural vs. binaural performance) revealed a statistically significant binaural benefit when two-talker babble [t(7)=2.98, p=0.02] but not speech-shaped noise [t(7)=1.21, p=0.26] originated on the acoustic-hearing side. This was the only case where the difference between the monaural and binaural SRT reached statistical significance.

Figure 8 replots the mean data from the CI users and compares them with the mean data from the NH listeners either under sound-field conditions (first panel) or over headphones with vocoder simulations of CI processing (third panel). Performance of the NH listeners with one ear (dark blue and red bars) was generally quite similar to that observed with the SSD+CI patients (second panel). The NH listeners in the sound-field gained a substantial benefit from having both ears available, as expected. In speech-shaped noise, the average SRT improved by 4.0 dB with the addition of a second ear. The benefit was much greater with the two-talker babble, with an average improvement of 9.7 dB in SRT with the addition of a second ear. This finding is consistent with those of Freyman et al. (2001), showing that release from informational masking tends to be larger than predicted by classic binaural interactions. As with the SSD+CI listeners, the benefit of the vocoded ear compared to the no-vocoder condition was minimal in some conditions (e.g., S₀N_CI in SSN; addition of the vocoder increased average SRT threshold by 2.3 dB) and more substantial in others (e.g., S₀N_NH in SSN; addition of vocoder decreased average SRT threshold by 5.8 dB).

Fig. 8.

Speech recognition thresholds for NH listeners in the sound field (first panel), NH listeners in the vocoder simulation (second panel), and SSD+CI in the sound field (third panel), groups, respectively. Speech and noise location appear on the x-axis and SRT in dB appears on the y-axis. Blue bars represent conditions where the masker was speech-shaped noise (SSN). Red bars represent conditions where the masker was time-reversed two-talker babble (TTB). Darker shaded bars represent the monaural condition. Lighter shaded bars represent the binaural condition. Error bars represent ±1 standard error.

A repeated-measures ANOVA with SRT as the dependent variable, listener group (SSD group and NH vocoder group only) as a between-subjects factor, and listening conditions (monaural/binaural), masker type (noise or two-talker interferer), and masker position (CI side, middle, acoustic ear side) as within-subjects factors revealed that all main effects were statistically significant. The main effect of group [F(1,14)=24.79, p<0.001, η_p²=0.64] showed that SRTs averaged across masker type and position were better (lower) for NH listeners under vocoder simulation conditions (SRT = −10 dB SMR) than SSD listeners in the sound field (SRT = −5.6 dB SMR). The main effect of listening condition (monaural/binaural) [F(1,14)=9.94, p=0.007, η_p²=0.42] showed that listening with two ears (SRT = −8.6 dB SMR) was better than listening with one ear (SRT = −7.0 dB SMR) overall. The main effect of masker type [F(1,14)=209.33, p<0.001, η_p²=0.94] showed that babble (informational masking) was a more effective masker than stationary speech-shaped noise (energetic masking) across groups, listening condition, and noise direction (SRT = −3.0 dB SMR and SRT = −13 dB SMR, respectively). Finally, as expected based on the head shadow effect, the main effect of masker location [F(2,28)=151.80, p<0.001, η_p²=0.92] was highly significant, with SRT improving on average as the masker moved from the acoustic to the (real or simulated) CI ear side.

The two-way interaction between listening condition and masker direction reached statistical significance [F(2,28)=15.61, p<0.001, η_p²=0.53]. Averaging across groups and masker type, performance was essentially the same for monaural and binaural listening when the masker originated at 0° azimuth (−5.64 dB SMR monaural, −6.68 dB SMR binaural) or on the (simulated) CI side (−12.04 dB SMR monaural, −12.35 dB SMR binaural), suggesting no binaural benefit in these configurations. However, there was a trend toward improvement in SRT across groups from monaural (−3.32 dB SMR) to binaural (−6.79 dB SMR) listening when the masker originated on the acoustic-hearing side, as would be expected based on the improved effective SMR at the CI ear in these conditions.

A similar pattern was observed in the significant two-way interaction between masker type and masker direction [F(2,28)=3.54, p=0.043, η_p²=0.20], perhaps because the difference in SRT between the two maskers appears somewhat greater for the S₀N_CI conditions than in the other two spatial conditions.

Finally, the three-way interaction between noise type, noise direction and group [F(2,28)=3.58, p=0.041, η_p²=0.20] was statistically significant. This arose because the differences between groups (SSD and NH vocoder simulation) were greater in some masker locations than others, depending on the masker type. All other two- and three-way interactions were not statistically significant.

Discussion

The purpose of the second experiment was to investigate whether and how SSD+CI listeners use their CI when listening to speech in noise under various spatial conditions. Marginal trends toward a benefit of the CI were observed in all spatial configurations, particularly when listening to speech in two-talker babble. The only statistically significant benefit of the CI, however, was observed when two-talker babble was presented to the acoustic ear side. Overall, the speech understanding in noise results seem to suggest that the addition of a CI does not restore binaural masking release in spatially separated conditions, but rather allows the listener to use the ear with the better signal-to-noise ratio, based on the head-shadow effect.

In terms of binaural summation, the addition of the CI did not significantly improve the SRT when the target and masker were collocated, consistent with earlier studies (Arndt et al. 2011; Buechner et al. 2010; Firszt et al. 2012; Stelzig et al. 2011; Vermeire & Van de Heyning 2009; Mertens et al. 2017; Sladen et al. 2017b), and consistent with the fact that speech perception via the CI is considerably worse than via a normal-hearing ear. To our knowledge, only two studies in postlingually deafened adults have demonstrated a small but significant improvement with a CI in the S₀N₀ condition where two- or four-talker babble was the interfering masker (Arndt et al. 2017; Távora-Vieira et al. 2013). In addition, Sladen et al. (2017a) found a small, yet significant improvement for speech in noise when speech was at 0° and speech maskers were coming from 8 speakers located 360° around the listener.

Our results did not mirror those reported by Bernstein et al. (2016), who found that the CI significantly improved sentence recognition in SSD listeners when target speech was presented to the NH ear and one or two same-gender interfering talkers were presented diotically. The difference may be related to the specific configuration used by Bernstein et al. (2016), with the monaural target and dichotic maskers, which does not reflect fully realistic listening conditions, and the coordinate response measure (CRM), which has a closed set of words and can produce very low SRTs. In our most challenging condition (S₀N_NH with the two-talker babble), the difference in SRT was in the right direction, but the size of the difference was small (3.5 dB) and did not reach statistical significance. Bernstein et al. (2015) observed a similar trend in vocoder simulations of SSD; unprocessed speech and one or two same-gender (informational) speech maskers presented one ear and vocoded masker(s) presented alone to the contralateral ear facilitated the separation between speech and masker(s). Likewise, the addition of the vocoded contralateral ear over silence had no effect when the competing maskers were stationary noise.

Although a larger cohort will be needed to address this question more conclusively, other studies with larger sample sizes have shown that a CI supplies a benefit in speech understanding when listening to speech in noise. Vermeire and Van de Heyning (2009), Mertens et al. (2017), Arndt et al. (2017), and Sladen et al. (2018) have all reported that patients with normal to moderate hearing loss in the acoustic-hearing ear can experience some improvements in performance with the addition of a CI. It is likely that the benefit of the CI increases with increasing degrees of hearing loss in the acoustic ear.

Results from the NH listeners were as expected, based on earlier studies: listening with two ears provided more speech understanding than listening with one ear alone when the target and masker were spatially separated, and the benefit was greater for the two-talker babble than for the speech-shaped noise masker. In general, the NH listeners with the vocoder showed a pattern of performance that was similar to that found with the SSD+CI participants, with a trend towards a small improvement in performance with the addition of the vocoded ear.

GENERAL DISCUSSION

Overall, the results from our experiments confirm that a CI partially restores spatial awareness in SSD patients. In addition, the results from the various stimulus conditions were consistent in showing that localization was possible via high-frequency ILD cues but was not improved by ITD cues in either the temporal envelope or fine structure.

Because most CIs do not transmit TFS, it was reasonable to assume that low-frequency fine-structure ITD cues were not available to SSD+CI patients. The lack of usable temporal fine-structure information was confirmed in all our patients. Thus, our study provides no evidence suggesting that SSD+CI listeners are sensitive to TFS-based ITDs in the low frequencies, even with processing schemes that explicitly code such information, such as the two participants (SSD 12 and 16) using fine-structure processing in the four most apical channels of their MED-EL CIs. In this respect, our results are consistent with those of Magnusson (2011), who also found no benefits associated with the FSP strategy. However, a larger sample size would be needed to confirm this finding.

In the absence of fine-structure ITD cues, it was possible that envelope ITDs could have been salient for SSD+CI users. In a sample size of five, Dorman et al. (2015) showed that SSD+CI listeners were able to localize low-pass noise above chance. Moreover, work by Stakhovskaya et al. (2016) suggested that SSD+CI users may be sensitive to interaural timing differences in the onset of electric and (bandpass-filtered) acoustic pulse trains. However, our results suggest that SSD+CI patients were not able to utilize ITD cues in the temporal envelope to improve sound localization.

In terms of speech perception, it was hypothesized that the benefit of binaural listening would be most apparent in cases involving informational masking, where a perceived spatial separation between the target and masker can lead to improved speech perception (Brungart et al. 2001; Freyman et al. 1999). In 7 of 8 SSD+CI participants a benefit of the CI when the speech and two-talker masker were spatially separated, but only when the noise masker originated on the acoustic ear side, suggesting a benefit based on improved SMR at the CI ear, rather than a use of binaural cues. Furthermore, the degree of benefit from the head shadow effect aligns with the SNR at the CI side. That is, the better the SNR, the larger the benefit.

Fortunately, in no conditions did the addition of the CI lead to poorer performance overall.

Overall, the results of this study suggest that, although the CI allows patients to compare the relative intensity level across ears to localize sound, SSD+CI users are not able to utilize binaural timing cues when listening to speech in noise under spatialized conditions. It is possible that other changes to the CI programming (such as better mapping to encourage binaural fusion between the ears) may allow SSD+CI patients to truly tap into binaural processes.

CONCLUSIONS

A CI can partially restore SSD listeners’ ability to localize sound, based on high-frequency ILD cues. The SSD+CI listeners were not able to make use of ITD cues in either TFS or the temporal envelope to improve performance. The addition of the CI did not produce a significant overall improvement in speech perception in noise or two-talker babble, when considering all conditions. However, some improvement was observed in specific cases, when the noise was presented on the side of the acoustic ear, thereby improving the speech-to-masker ratio at the CI. This effect was particularly evident in the cases of a two-talker babble. In no cases did the CI interfere with localization or speech understanding.