Effects of rate and age in processing interaural time and level differences in normal-hearing and bilateral cochlear-implant listeners

Sean R Anderson; Kyle Easter; Matthew J Goupell

doi:10.1121/1.5130384

. 2019 Nov 12;146(5):3232–3254. doi: 10.1121/1.5130384

Effects of rate and age in processing interaural time and level differences in normal-hearing and bilateral cochlear-implant listeners^a)

Sean R Anderson ^1,^✉, Kyle Easter ², Matthew J Goupell ^2,^b)

PMCID: PMC6948219 PMID: 31795662

Abstract

Bilateral cochlear implants (BICIs) provide improved sound localization and speech understanding in noise compared to unilateral CIs. However, normal-hearing (NH) listeners demonstrate superior binaural processing abilities compared to BICI listeners. This investigation sought to understand differences between NH and BICI listeners' processing of interaural time differences (ITDs) and interaural level differences (ILDs) as a function of fine-structure and envelope rate using an intracranial lateralization task. The NH listeners were presented band-limited acoustical pulse trains and sinusoidally amplitude-modulated tones using headphones, and the BICI listeners were presented single-electrode electrical pulse trains using direct stimulation. Lateralization range increased as fine-structure rate increased for ILDs in BICI listeners. Lateralization range decreased for rates above 100 Hz for fine-structure ITDs, but decreased for rates lower or higher than 100 Hz for envelope ITDs in both groups. Lateralization ranges for ITDs were smaller for BICI listeners on average. After controlling for age, older listeners showed smaller lateralization ranges and BICI listeners had a more rapid decline for ITD sensitivity at 300 pulses per second. This work suggests that age confounds comparisons between NH and BICI listeners in temporal processing tasks and that some NH-BICI binaural processing differences persist even when age differences are adequately addressed.

I. INTRODUCTION

An increasing number of individuals with severe to profound sensorineural hearing loss are receiving a second cochlear implant (CI) (Peters et al., 2010) that is intended to improve the ability to localize sound sources in the horizontal plane by conveying interaural cues. Spatial information in the horizontal plane is important for personal safety (e.g., to identify the direction of an oncoming vehicle) and is used to help perceptually segregate a target from interfering sounds (Bregman, 1990; Cherry, 1953; Darwin and Hukin, 2000). Numerous studies have demonstrated that adult (Aronoff et al., 2010; Bernstein et al., 2016; Grantham et al., 2007; Loizou et al., 2009; van Hoesel, 2007) and child (Garadat and Litovsky, 2007; Gordon et al., 2014; Grieco-Calub and Litovsky, 2010) BICI listeners are sensitive to binaural cues, are able to localize sound sources, and show improved speech understanding in noise with spatial separation.

Listeners can localize sounds in the horizontal plane using interaural time differences (ITDs) and interaural level differences (ILDs). Of these two cues, NH listeners rely most heavily on low-frequency (<1500 Hz) ITDs conveyed in the temporal fine-structure of the acoustical signal when localizing broadband sounds, thus demonstrating a low-frequency ITD dominance for sound localization (Macpherson and Middlebrooks, 2002; Wightman and Kistler, 1992). This contrasts with bilateral CI (BICI) listeners who seem to rely almost entirely on ILDs, thus demonstrating an ILD dominance for sound localization (Aronoff et al., 2010; Grantham et al., 2008; Seeber and Fastl, 2008).

A. Limitations of hardware and programming approaches

Auditory signals can be broken down into two portions: slower changes in energy over time called the “temporal envelope,” and faster changes in amplitude and phase over time called the “temporal fine-structure.” Two problems with how the clinical sound processors convey ITDs likely contribute to ILD dominance in BICI listeners: (1) the replacement of temporal fine-structure information with an acoustically irrelevant, high-rate ( $≳$ 1000 pulses per second; pps) carrier and (2) the lack of bilateral synchronization.

Most CI sound processing strategies extract and convey only the temporal envelope ( $≲$ 300 Hz) (Loizou, 2006). The original temporal fine-structure is replaced with an electrical pulse train of a constant rate, where the amplitude of individual pulses is modulated by the temporal envelope extracted from the acoustic signal (Loizou, 2006). The resulting envelope-based signal is transmitted to the electrode array and ultimately encoded by the auditory nerve. Thus, the envelope-based interaural information (ILDs and envelope ITDs) is relatively preserved by signal processing in CIs, while fine-structure-based ITDs are removed entirely.

Bilaterally unsynchronized clinical sound processors also likely introduce uncontrolled, time-varying ILDs and envelope ITDs. For example, unsynchronized clinical sound processors can introduce asynchrony in time on a sub-millisecond scale within and across electrodes due to uncoordinated pulse delivery. Each sound processor has independent compressions and automatic gain controls, which can dynamically change the ILD in a way that is not reflected in the acoustic signal (Archer-Boyd and Carlyon, 2019; Potts et al., 2019).

To further complicate issues, clinicians program CI processors by determining stimulation rates and levels independently for each CI, and BICI programming is performed without consideration of the possible distortions to interaural cues. Loudness growth has been shown to change depending upon electrode site within each ear (Bierer and Nye, 2014). Thus, independent mapping can cause each ear to have substantially different dynamic ranges and result in different loudness growth across ears. Consequently, pulse trains that are modulated in amplitude [modulated-amplitude pulse trains (MAPTs)], may cause interaural decorrelation because of the potentially different loudness growth functions (Whitmer et al., 2014). Thus, MAPTs may be perceived away from the center of the head or as broader in width (Goupell et al., 2013a; Goupell, 2015; Goupell and Litovsky, 2015). In contrast, stimuli that do not elicit loudness changes over time, like constant-amplitude pulse trains (CAPTs), are more likely to be perceived as relatively punctate.

B. Effect of carrier and modulation rate on binaural processing

ITD just-noticeable differences (JNDs) in NH listeners when conveyed via CAPTs¹ are lowest (i.e., best) at 100–200 pps, with JNDs as low as about 50 μs. Note that lower JNDs indicate better performance. As pulse rate increases above 300 pps, JNDs increase until chance performance occurs around 500–1000 pps (Bernstein and Trahiotis, 2014; Laback et al., 2007; Majdak and Laback, 2009; Monaghan et al., 2015). Likewise, BICI listeners perform best on ITD discrimination tasks with CAPTs at pulse rates between 100 and 300 pps (Kan and Litovsky, 2015; Laback et al., 2015; Van Hoesel et al., 2009), though rates lower than 100 pps have seldom been addressed in the human CI literature and were included in the present study. As pulse rate increases beyond 300 pps, CAPT ITD sensitivity declines precipitously in most listeners (Noel and Eddington, 2013; van Hoesel et al., 2009). In single unit recordings of the inferior colliculus from anesthetized cats implanted with CIs, firing rate was highest with electrical stimulation for CAPTs up to about 100 pps (Smith and Delgutte, 2007), consistent with results from human perception. Single unit recordings in the inferior colliculus of unanesthetized rabbits implanted with CIs have further shown that the variance of responses decreased and neural JNDs increased for CAPT ITDs with pulse rates greater than 100 pps (Chung et al., 2016; Hancock et al., 2012).

Most clinical sound processors stimulate electrodes with MAPTs using high carrier rates. Thus, ITDs and ILDs in the envelope are much more likely to be useful in everyday listening and warrant further investigation. Decreasing ITD sensitivity with increasing rate also occurs for amplitude-modulated stimuli. NH listeners have lowest ITD JNDs around a 100-Hz envelope modulation rate (Bernstein and Trahiotis, 2002; Bernstein and Trahiotis, 2009; Majdak et al., 2006). Larger depths of amplitude modulation lead to lower ITD JNDs in NH listeners (Bernstein and Trahiotis, 2009). Similarly, BICI listeners have lowest ITD JNDs for MAPTs around a 100-Hz envelope modulation rate (Noel and Eddington, 2013; van Hoesel, 2007; van Hoesel et al., 2009). BICI listeners are also more sensitive to ITDs of MAPTs with larger modulation depths (Ihlefeld et al., 2014). Interestingly, when comparing ITD JNDs for CAPTs and MAPTs in BICI listeners, ITD JNDs are slightly but significantly higher for MAPTs at 300 Hz than CAPTs at 300 pps (van Hoesel, 2007; van Hoesel et al., 2009).

C. Effect of age on binaural processing

Age-related decreases in temporal processing abilities have been documented using numerous psychophysical tasks in older listeners with and without hearing loss (Gordon-Salant et al., 2010). Temporal gap detection thresholds are poorer in older NH (ONH) compared to younger NH (YNH) listeners (Gallun et al., 2014; He et al., 1999; Snell, 1997; Strouse et al., 1998). ONH listeners are also poorer than YNH listeners at discriminating the duration of sounds (Fitzgibbons et al., 2006; Gallun et al., 2014).

With respect to spatial hearing, ONH listeners have greater ITD JNDs compared to YNH listeners (Grose and Mamo, 2010; Strouse et al., 1998). ONH listeners also demonstrate higher root-mean-square errors in sound localization experiments (Abel et al., 2000; Freigang et al., 2014), larger minimum audible angles (Chandler and Grantham, 1992; Freigang et al., 2014), and smaller binaural masking level differences (Pichora-Fuller and Schneider, 1991; Pichora-Fuller and Schneider, 1998; Strouse et al., 1998) compared to YNH listeners.

Another task that has been used to assess sensitivity to binaural cues is intracranial lateralization. In some lateralization experiments, listeners hear a single stimulus interval and indicate its perceived intracranial location on a visual scale. Lateralization procedures have the advantage of being more comparable to free-field sound source localization and provide access to numerous aspects of perception. Using data from a lateralization task, it is possible to extract the range of responses, predict sensitivity to spatial cues, and account for bias (e.g., Baumgärtel et al., 2017; Kan et al., 2013; Macpherson and Middlebrooks, 2002; Stakhovskaya and Goupell, 2017). For example, studies examining lateralization in NH listeners have demonstrated an increased extent of perceived laterality, or the intracranial distance subjects hear ITDs from left to right, with greater modulation depth for amplitude-modulated stimuli (Bernstein and Trahiotis, 2011). In this case, a greater range of lateralization indicates a stronger impact of that spatial cue on perception, and therefore greater utility.

Using a lateralization task, Babkoff et al. (2002) showed that ONH listeners had smaller lateralization ranges compared to YNH listeners using short-duration, 10-Hz acoustic click trains. The lateralization range, however, has yet to be compared between ONH and BICI listeners. A recent study by Baumgärtel et al. (2017) explored the relationship between pulse rate and lateralization in YNH, middle-aged NH, and middle-aged BICI listeners. For NH listeners, they simulated electrical stimulation using band-pass filtered pulse trains. Their results showed that the lateralization range for BICI listeners was similar for roughly aged-matched middle-aged NH listeners.

There were several aspects of the study by Baumgärtel et al. (2017) that could be considered for further investigation. First, many adult BICI listeners in the literature are older than middle-aged (e.g., Goupell, 2015). Thus, it should be determined how lateralization changes in this older group with proper age matching and a larger sample size. Second, ITD JNDs tend to increase above 300 pps in both NH and BICI listeners, but the study by Baumgärtel et al. (2017) used a maximum of 200 pps. A higher rate would help clarify if there is an interaction between rate limitation and stimulation mode (electric vs acoustic). Third, while less temporally precise than the ITD circuit, temporal precision is also important in the neural ILD circuit (e.g., Brown and Tollin, 2016; Joris and Yin, 1995). ILD lateralization was not explored by Baumgärtel et al. (2017) and ILDs are the primary localization cue for BICI listeners.

D. Study aims and hypotheses

The purpose of the present study was to compare ITD and ILD lateralization in NH and BICI listeners to determine the effects of rate and temporal fine-structure vs envelope on lateralization. It was hypothesized that BICI listeners would exhibit greater variability in their lateralization responses for MAPTs than CAPTs, especially compared to NH listeners, since loudness growth functions may differ across electrodes and vary widely across individuals (e.g., Fu, 2005; Bierer and Nye, 2014). We further hypothesized that the range of ITD lateralization would decrease for older listeners because age-related temporal processing deficits should greatly affect the ITD circuits that rely on the finest temporal processing mechanisms in the central nervous system. In contrast, such an effect may not occur for ILD lateralization since ILDs may be more robust to temporal degradations. Finally, we hypothesized a rate by age interaction for ITD lateralization, where the age-related temporal processing deficits would diminish lateralization range more greatly for older listeners than younger listeners at higher rates.

II. EXPERIMENT 1: LATERALIZATION OF CAPTS AND MAPTS IN YNH AND BICI LISTENERS

A. Methods

1. Listeners and electrode selection

Eleven YNH (21–25 years; average = 22.4 years) and eight BICI (46–74 years; average = 61.9 years) listeners were tested in this study. The YNH listeners had hearing thresholds ≤20 dB hearing level (HL) for octave frequencies between 250 and 8000 Hz, and asymmetries that were no more than 10 dB at the tested frequencies. The BICI listeners were post-lingually deafened with at least six months of experience with their CIs (see Table I for listeners' demographic information). Most of the BICI listeners were familiar with the testing procedure from past participation in similar studies. All BICI listeners were implanted with 22-electrode CIs from Cochlear Ltd. For each listener, one middle electrode pair (about electrode 12 in each ear) was tested. The middle electrode region was chosen to minimize the possibility that an electrode would be deactivated. There are sometimes differences in ITD sensitivity according to place-of-stimulation in BICI listeners, with either apical or basal electrodes yielding better sensitivity, but these effects vary greatly across listeners (Kan and Litovsky, 2015; Laback et al., 2015). We also chose the middle electrode to minimize introducing a confound based on possible sensitivity differences due to place-of-stimulation. The study was approved by the University of Maryland Institutional Review Board for Human Subjects and all listeners gave informed consent.

TABLE I.

Demographics and etiology for BICI listeners. Table includes age at testing, estimated age of onset of deafness, duration of deafness, years of CI experience, and etiology. All listeners had implants and clinical processors from Cochlear Ltd.

Subject ID	Age	Onset (R/L)	Yrs. Deafness (L/R)	Yrs. CI (L/R)	Etiology (R/L)
CAB	68	Progressive/Progressive	36/43	17/10	Unknown/Unknown
CAD	73	Progressive/Progressive	3/9	10/4	Unknown/Unknown
CAE	61	Progressive/Progressive	4/3	6/7	Hereditary/Hereditary
CAG	63	Progressive/Progressive	0/5	8/11	Unknown/Unknown
CAI	69	Progressive/Progressive	not available	2/6	Unknown/Unknown
CAP	44	Progressive/Progressive	1/0	4/5	Unknown/Unknown
CAS	50	Progressive/Progressive	1/5	6/2	Unknown/Unknown
CAZ	67	Progressive/Progressive	not available	not available	Unknown/Unknown

Open in a new tab

2. Equipment

Stimuli were generated on a personal computer using matlab (MathWorks, Natick, MA). The electrical stimuli presented to BICI listeners were delivered by Nucleus Implant Communicator (NIC version 2; Cochlear Ltd., Sydney, Australia) software that controlled a pair of bilaterally synchronized L34 research processors (Cochlear Ltd.).

Acoustic stimuli presented to YNH listeners were delivered using a soundcard (UA-25 EX; Edirol/Roland Corp., Los Angeles, CA), amplifier (D-75A; Crown Audio, Elkhart, IN), programmable attenuator (PA5; Tucker-Davis Technologies, Alchua, FL), and circumaural headphones (HD650; Sennheiser, Hanover, Germany). Testing occurred in a double-walled sound attenuating booth (Industrial Acoustics Inc., Bronx, NY).

3. Stimuli

For BICI listeners, CAPTs were presented at pulse rates of 10, 30, 100, 300, and 1000 pps, and MAPTs were presented at modulation rates of 10, 30, 100, and 300 Hz. Electrical pulse trains were biphasic monopolar current pulses of 25 μs per phase with an interphase gap of 8 μs. The CAPTs were presented at a listener's most comfortable loudness level. The MAPTs had a 1000-pps carrier and a modulation depth of 100%. Modulation was sinusoidally imposed on a linear current unit (CU) scale, and peak modulation was fixed at the maximum comfortable level, which resulted in a comfortable loudness level. The control carrier condition was tested at comfortable level for the 1000-pps CAPT to gather a baseline for the MAPT condition when the stimulus was unmodulated. A carrier rate of 1000 pps was chosen because it is near the default rate of 900 pps provided to Cochlear Ltd. Patients and should have resulted in sufficient sampling of the amplitude modulation. One limitation of the present study is that loudness balancing was not completed across stimulation rates and amplitude modulation conditions. It is possible that some configurations resulted in greater loudness than others. Stimuli were presented for 500 ms.

For NH listeners, CAPTs were presented at pulse rates of 10, 30, 100, and 300 pps, and MAPTs were presented at modulation rates of 10, 30, 100, and 300 Hz. The acoustic CAPTs were Gaussian-shaped pulses with a center frequency of 4000 Hz and a bandwidth of 1.5 mm or 851 Hz [i.e., the distance along the cochlea to reach −3 dB from the peak; see Goupell et al. (2013b) for additional details]. Gaussian-shaped pulses were generated by multiplying a 4000-Hz pure tone by a Gaussian envelope and concatenating these pulses in time. The bandwidth of the pulses is inversely related to the temporal width of each pulse. All rates for acoustic CAPTs had the same peak amplitude. Pulse rates of 10–300 pps were used because at higher rates, acoustic pulses begin to temporally overlap. Similarly, listeners were tested with a 4000-Hz pure tone (i.e., carrier) to compare against the 1000-pps condition in BICI listeners. A sine tone was used since acoustic pulse trains with 1.5-mm bandwidth presented at 1000 pps result in temporal overlap, and therefore reduced modulation depth. The acoustic, nominally MAPT² stimuli had a 4000-Hz sine tone carrier and sinusoidal amplitude modulation with 100% modulation depth, with modulation rates of 10–300 Hz. MAPTs were presented at 65 dB-A and had a duration of 500 ms. The level of 65 dB-A for CAPTs was determined using 300-pps CAPTs, which would have resulted in a lower sound level at lower rates. Stimuli were cosine gated with 10-ms rise/fall times to avoid transients, and ITDs and ILDs were applied to the whole waveform after gating.

Low-frequency masking noise was added to the acoustic stimuli to prevent NH listeners from potentially exploiting low-frequency non-linear distortions. The noise had two cutoff frequencies: the first at 200 Hz where the rolloff was –3 dB/octave and the second at 1000-Hz where roll-off was –18 dB/octave (Klein-Hennig et al., 2011). The noise was also cosine gated with rise/fall times of 10 ms. Noise onset was 1000 ms before and offset was 1000 ms after the target signal, resulting in a total duration of 2500 ms. Noise presentation level was 61.1 dB-A, which equates to 30 dB spectrum level.

4. Procedure

a. Preliminary CI testing.

The direct stimulation procedures followed best practices as outlined in Litovsky et al. (2017). Threshold, comfortable, and maximum comfortable levels were measured for each listener using CAPTs at each rate for the middle electrode pair. Equal loudness of comfortable levels across the two electrodes was ensured by performing sequential presentation of the two electrodes with a 500-ms inter-stimulus interval.

Pitch-matching was performed in an attempt to have a matched interaural place-of-stimulation (Litovsky et al., 2012), which is likely within the range of good binaural sensitivity for monopolar CI stimulation (Goupell et al., 2019; Kan et al., 2015). Listeners were presented two CAPTs of equal loudness and duration in two separate intervals, and responded whether the second sound was lower, the same, or higher in pitch than the first sound in a five-alternative forced-choice task. For more details on this procedure, see Litovsky et al. (2012).

b. Lateralization task and stimuli.

Listeners performed a lateralization task by indicating the intracranial location of a stimulus on the picture of a face (Goupell et al., 2013b; Kan et al., 2013). Listeners could respond with up to three source locations, indicating the lack of a fused image. Each listener was instructed to place the strongest or most salient image on the top bar, which was used in all analyses if multiple images were reported. If a broad spatial image was perceived, listeners were instructed to match to the center of the image. Stimuli could be replayed as many times as needed.

For the BICI listeners, ITDs and ILDs were tested in separate blocks because a left or right bias can persist for bilateral equal-loudness comfortable levels (Fitzgerald et al., 2015; Goupell et al., 2013a; Stakhovskaya and Goupell, 2017). Left/right bias was measured by locating where the lateralization curve crossed the x axis. The amount of bias in current level units (CUs) was then subtracted from the stimulation rate of the biased side for CAPTs and from maximum comfortable level for MAPTs for the ITD experiment. Interaural centering (applying an ILD that resulted in a sound perceived in the center of the head with a 0-μs ITD) was performed for the ITDs using the ILD data, but no centering was completed for the ILDs. Table II summarizes the stimuli tested in experiments 1 and 2.

TABLE II.

Summary of stimuli tested in experiments 1 and 2 by group. A cell including an X indicates that group was tested with the corresponding stimulus. For BICI listeners, the carrier of MAPTs was a 1000-pps pulse train. For NH listeners, the carrier of MAPTs was a 4000-Hz sine tone.

Stimulus	Rate	Group
Stimulus	Rate	YNH	BICI	ONH
Carrier	N/A	X	X
CAPT	10	X	X
	30	X	X	X
	100	X	X	X
	300	X	X	X
MAPT	10	X	X
	30	X	X
	100	X	X
	300	X	X

Open in a new tab

For most conditions with the BICI listeners, ILDs of 0, ±2, ±5, ±10, ±20, and ±30 CUs were tested. Negative ILDs were created by reducing stimulation level in CUs in the right ear and positive ILDs were created by reducing the stimulation level in the left ear. However, not all listeners were tested with these values due to differences in sensitivity. Likewise, for most conditions with the BICI listeners, ITDs of 0, ±100, ±200, ±400, ±800, ±1600, and ±2500 μs were tested. Large ITDs were used to determine whether lateralization range would asymptote for greater ITDs in BICI listeners. ITDs were generated by delaying the pulse train to one ear. For example, an ITD of –800 μs was created by delaying the pulse train in the right ear by 800 μs.

ILDs or ITDs from one condition in Table II were presented in random order within each testing block, and the order of conditions was chosen randomly across BICI listeners. In all BICI listeners, the greatest magnitude ILD or ITD was presented first with a 100-pps CAPT. This should have resulted in the maximum observable lateralization range in the experiment. The remaining range of ILDs and ITDs to be tested in the experiment was determined based on these results.

For NH listeners, ILDs of 0, ±1.5, ±3, ±6, ±9, and ±15 dB and ITDs of 0, ±100, ±200, ±400, ±800, and ±1000 μs were imposed on the stimulus for a total of eleven conditions. Acoustic ILDs were imposed by reducing the level in one ear and increasing the level, by an equal amount, in the other ear. Negative ILDs correspond to a greater stimulation level in the left ear and positive ILDs correspond to a greater stimulation level in the right ear. Acoustic ITDs were created by delaying the waveform to the opposite ear (as in BICI listeners). Left/right bias was not corrected in the NH listeners. Unlike BICI listeners, the same ILDs and ITDs were tested in NH listeners. Each ILD or ITD was presented in random order for each condition in Table II within the same testing block, resulting in a total of 20 repetitions per ILD or ITD for each condition.

c. Analysis.

The selected locations for a particular ILD or ITD were converted into numerical values within the range of ±10, corresponding to the left (negative) and right (positive) sides. These data were fit using the following function:

y = A erf (\frac{x - μ_{x}}{\sqrt{2} σ}) - μ_{y},

(1)

where A corresponds to maximum extent of lateralization achieved by the fit, erf corresponds to the error function which takes as input any real number and whose output is bounded between ±1, x corresponds to the value of the ITD or ILD, μ_x corresponds to the horizontal shift, σ is related to the lateralization slope, and μ_y corresponds to the vertical shift and from now on is referred to as “lateralization offset.” All curve fitting was completed using non-linear least squares via the curve fitting toolbox in matlab.

Several indices of performance were used to characterize lateralization of the different stimuli presented to groups in this experiment. The impact of spatial cues on lateralization of a perceived image can be assessed from the efficacy of the spatial cue on altering the listener's perceived intracranial lateralization to the left or right. Accordingly, lateralization range was calculated as the absolute difference of the perceived left and right positions for ITD values of ±1000 μs or ILD values of ±15 dB or ±20 CUs from the fitted curve for each listener. The lateralization offset and ILD corresponding to a centered image (where the mean lateralization response was 0) were used as indices of bias. Finally, the standard deviation of lateralization responses for each magnitude of ITD or ILD was used to determine the reliability of the lateralized image for each stimulus and group. For each of these indices (lateralization range, bias, and standard deviation of lateralization response), two mixed-effects analysis of variance tests (ANOVAs) were conducted for ITDs and ILDs. The first tested for differences between envelope conditions and group; the second tested for a subset of envelope conditions, rates, and group. Two ANOVAs were completed because rate was not varied for the carrier condition in the same respect as CAPTs and MAPTs. Thus, rate was excluded as a factor in the first ANOVA. Fixed-effects of envelope (CAPT and MAPT), rate (10, 30, 100, and 300 Hz), group (YNH and BICI), and their interactions were estimated in the second set of ANOVAs. All post hoc paired comparisons were completed using estimated-marginal means with Tukey adjustments for multiple comparisons. All degrees of freedom and hypothesis tests were estimated using the Satterthwaite method (Satterthwaite, 1946).

B. Results

The goal of this experiment was to determine the effects of pulse rate and temporal fine-structure vs envelope on the lateralization of acoustic or electric pulse trains in YNH and BICI listeners. Individual data for BICI listeners are shown in Fig. 1 for ITDs and Fig. 2 for ILDs. Figure 1 shows that large ITDs (≥1500 μs) were needed before listeners perceived the sound as lateralized completely to the left or right. For CAPTs, listeners heard 10- and 30-pps CAPTs as most lateralized with ITDs, and as the pulse rate increased, the lateralization range decreased. For MAPTs, lateralization range peaked near 100 Hz and decreased as the modulation rate increased or decreased. Finally, ITDs applied to the 1000-pps CAPT pulse trains (i.e., carrier) did not produce a change in perceived intracranial lateralization except for listener CAZ. Figure 2 shows that, in contrast to ITDs, lateralization range for ILDs increased with increasing rate for CAPTs and remained relatively consistent across rate for MAPTs.

FIG. 1. — (Color online) Lateralization of ITDs by listener and pulse/modulation rate for BICI listeners. Each row corresponds to one BICI listener, where their three-character ID is written in the leftmost panel. From left to right, each panel corresponds to a pulse rate (for CAPTs) or modulation rate (for MAPTs) of 10, 30, 100, and 300 pps/Hz. Closed squares represent CAPTs, closed circles represent MAPTs, and open triangles represent the high-rate (1000 pps) carrier only for each pulse/modulation rate. The carrier rate does not change for each column but is included for comparison. The x axis corresponds to the ITD presented with negative values indicating left-leading ITDs. The y axis corresponds to the extent of perceived left or right intracranial lateralization (see Sec. II A 4). Error bars represent ±1 standard deviation from the mean.

FIG. 2. — (Color online) Lateralization of ILDs by listener and pulse/modulation rate for the BICI listeners. Plotted as in Fig. 1 with the exception that the x axis corresponds to the ILD presented, with negative values indicating greater CUs in the left ear.

Average lateralization results for YNH and BICI listeners are shown in Fig. 3. The BICI listeners showed smaller amounts of lateralization and required larger ITDs and ILDs to reach full lateralization range [Figs. 3(A) and 3(C)] compared to YNH listeners [Figs. 3(B) and 3(D); note the difference in scale along the x axes]. In general, YNH listeners had greater variability in their lateralization responses compared to BICI listeners. See Supplementary Figs. 1 and 2 for raw YNH listeners' data.³

1. Lateralization range

It was of greatest interest to determine the left- and right-most extent to which listeners lateralized sounds, or lateralization range, when a large ITD or ILD was applied. Average lateralization ranges for each condition are shown in Fig. 4 for each group. Additionally, Supplementary Fig. 3 shows lateralization range across rate and stimulus plotted by individual.³

FIG. 4. — (Color online) Lateralization range for (A) ITDs and (B) ILDs with CAPTs (left panels) and MAPTs (right panels). Closed shapes represent average and individual data for YNH listeners and open shapes represent BICI listeners. The x axis corresponds to the pulse rate for CAPTs or modulation rate for MAPTs. The y axis corresponds to the lateralization range as defined in Sec. II A 4 c. Open shapes show results for BICI listeners and closed shapes show results for YNH listeners. Error bars represent 95% within-subject confidence intervals as described by Baguley (2012) and can be interpreted as informal *post hoc* comparisons at α = 0.05. Individual data are shown offset to the right of the mean.

a. ITDs.

For all results in this section, the ITD applied to the signal was varied, and the ILD was fixed at 0 dB for YNH listeners while an ILD (in CUs) resulting in a centered image was applied for BICI listeners. First, we performed an analysis on signal type and averaged across rates for CAPTs and MAPTs. The results indicated that lateralization range of ITDs was significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(1,164) = 27.53, p < 0.0001]. YNH listeners showed larger lateralization ranges than BICI listeners [F(1,24) = 10.39, p < 0.01]. There was also a significant stimulus × group interaction [F(2,164) = 7.79, p < 0.001]. For BICI listeners, lateralization ranges were not significantly different between CAPTs and MAPTs [t(164) = 0.40, p > 0.05], but lateralization ranges were significantly larger for CAPTs compared to the carrier [t(164) = 5.04, p < 0.0001] and MAPTs compared to the carrier [t(164) = 4.82, p < 0.0001]. For YNH listeners, lateralization ranges for CAPTs were significantly larger compared to MAPTs [t(164) = 6.07, p < 0.0001] and CAPTs compared to the carrier [t(164) = 4.79, p < 0.001], but lateralization ranges were not significantly different for MAPTs and the carrier [t(164) = 1.36, p > 0.05]. Together, these results showed that lateralization range was mostly largest for CAPTs.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. The results indicated that lateralization range was significantly affected by stimulus [F(1,133) = 48.80, p < 0.0001] and that increasing rate decreased lateralization range [F(3,133) = 36.12, p < 0.0001]. YNH listeners showed larger lateralization ranges than BICI listeners [F(1,19) = 5.28, p < 0.05]. There were significant interactions of stimulus × rate [F(3,133) = 74.68, p < 0.0001] and stimulus × group [F(1,133) = 34.96, p < 0.0001], but not rate × group [F(3,133) = 1.40, p > 0.05]. Lateralization range was significantly affected by a three-way interaction of stimulus × rate × group [F(3,133) = 5.11, p < 0.01]. For the sake of brevity and reader convenience, an extensive report of all paired-comparisons is excluded. Instead, the error bars in Fig. 4 represent 95% within-subject confidence intervals; that is, in Fig. 4 where errors bars do not overlap, differences between error bars within the same panel can be interpreted as informal paired-comparisons (Baguley, 2012). The three-way interaction was driven by two primary patterns in the data. First, lateralization ranges were smaller for BICI listeners compared to YNH listeners with CAPTs, especially at higher rates (100–300 Hz), but not MAPTs. Second, lateralization ranges decreased as pulse rate increased for CAPTs, but for MAPTs lateralization ranges were maximal at 100 Hz. Together, these results showed that the effect of rate depended upon the stimulus type (CAPT or MAPT).

b. ILDs.

For all results in this section, the ILD applied to the signal was varied and the ITD was fixed at 0 μs. First, we performed an analysis to account for the carrier signal and averaged across rates for CAPTs and MAPTs. The results indicated that lateralization range of ILDs was significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(2,164) = 46.19, p < 0.0001] and a stimulus × group interaction [F(2,164) = 13.74, p < 0.0001], but was not significantly affected by group [F(1,20) = 3.76, p > 0.05]. This interaction was analyzed using post hoc analysis with Tukey adjustments for multiple comparisons. For BICI listeners, lateralization ranges were significantly smaller for CAPTs compared to MAPTs [t(164) = −8.85, p < 0.0001], CAPTs compared to carrier [t(164) = −5.91, p < 0.0001], but not carrier and MAPTs [t(164) = 0.32, p > 0.05]. For YNH listeners, lateralization ranges were significantly smaller for CAPTs compared to MAPTs [t(164) = −3.35, p < 0.05]. There was not a significant difference between CAPTs and the carrier [t(164) = −2.15, p > 0.05] or MAPTs and the carrier [t(164) = 0.04, p > 0.05]. Together, these results showed that lateralization ranges were mostly smallest for CAPTs, especially with BICI listeners.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. The results indicated that lateralization range was significantly affected by stimulus [F(1,133) = 114.53, p < 0.0001], rate [F(3,133) = 7.24, p < 0.001], and group [F(1,19) = 5.98, p < 0.05]. There were also significant interactions of stimulus × rate [F(3,133) = 11.78, p < 0.0001], stimulus × group [F(1,133) = 33.68, p < 0.0001], and rate × group [F(3,133) = 8.00, p < 0.0001]. Lateralization range was significantly affected by a three-way interaction of stimulus × rate × group [F(3,133) = 6.63, p < 0.001]. Data in Fig. 4(B) show lateralization range results for ILDs with 95% within-subject confidence intervals. The three-way interaction was driven by two primary patterns in the data. First, lateralization ranges were smaller for BICI compared to YNH listeners with CAPTs but not MAPTs. Second, lateralization ranges increased as pulse rate increased for CAPTs in BICI but not YNH listeners. Lateralization ranges did not change across rate for MAPTs in either group. Together, these results demonstrated that there were different effects of rate for CAPTs between the YNH and BICI groups.

2. Lateralization slope

One notable difference between YNH and BICI listeners in Figs. 1–4 is the change in lateralization with changing ITD or ILD. This is best characterized by fitting a linear regression through the linear portion of the lateralization function and directly relates change in the spatial cue with change in perception. For ITDs, the slope of the lateralization function significantly predicted the lateralization range when data from all groups in experiment 1 were included and the intercept was allowed to vary with listener [F(21,167) = 12.37, p < 0.0001; adjusted R² = 0.56]. For ILDs, the slope of the lateralization function predicted lateralization range to an even greater extent [F(21,167) = 17.11, p < 0.0001; adjusted R² = 0.64]. Both regressions were completed using the analysis of covariance procedure outlined by Bland and Altman (1995). Since slope was highly correlated with lateralization range, no additional analyses were completed. Lateralization slopes by group and stimulus in experiments 1 and 2 are shown in Supplementary Fig. 4 for reference.³

3. Bias

Left-right lateralization bias was examined because previous experiments in BICI listeners showed that they can demonstrate large amounts of bias in their lateralization responses for single-electrode stimulation (Goupell et al., 2013a; Stakhovskaya and Goupell, 2017) even when they are bilaterally loudness balanced (Fitzgerald et al., 2015). Further, centering was completed using the ILD lateralization data to help ensure that BICI listeners heard a centered image when a 0-μs ITD was presented during ITD lateralization for each stimulus. Thus, only the ILD data should contain bias. Bias was characterized in two ways that are plotted in Fig. 5.

FIG. 5. — (Color online) (A) Lateralization offset and (B) ILD corresponding to a centered image with CAPTs (left panels) and MAPTs (right panels). Closed shapes represent YNH listeners and open shapes represent BICI listeners. The x axis corresponds to the pulse rate for CAPTs or modulation rate for MAPTs. The y axis in (A) corresponds to the lateralization offset parameter in the fitted curve as defined in the Sec. II A 4 c. The y axis in (B) corresponds to the ILD that resulted in a centered image (i.e., a lateralization response of 0). Error bars represent 95% within-subject confidence intervals as described by Baguley (2012) and can be interpreted as informal *post hoc* comparisons at α = 0.05. Individual data are shown offset to the right of the mean.

a. Lateralization offset.

Figure 5(A) shows the results for lateralization offset for ITDs. The left panel shows results for CAPTs and right panel shows results for MAPTs. The parameter associated with lateralization offset in the lateralization function in Eq. (1) is μ_y. Individuals plotted in Fig. 5(A) show that one listener from the YNH and BICI groups had inordinately large lateralization offsets at 100 Hz. Variability across individuals was more uniform for other rates and MAPTs.

First, we performed an analysis to account for the carrier signal and averaged across rates for CAPTs and MAPTs. The results indicated that the lateralization offset of ILDs was not significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(2,156) = 1.30, p > 0.05], group [F(1,36) = 0.06, p > 0.05], or an interaction of stimulus × group [F(2,156)= 0.64, p > 0.05]. This result demonstrates that the mean lateralization offset was similar across conditions and groups.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. Data in Fig. 5(A) show lateralization offset results for ILDs with 95% within-subject confidence intervals. The results indicated that the lateralization offset for ILDs was neither significantly affected by stimulus [F(1,144) = 2.49, p > 0.05] nor group [F(1,144) = 1.17, p > 0.05], but was significantly affected by rate [F(3,144) = 3.07, p < 0.05]. There was also a significant interaction of stimulus × rate [F(3,144) = 2.86, p < 0.05]. There was neither a significant interaction of stimulus × group [F(1,144) = 0.61, p > 0.05] nor rate × group [F(3,144) = 1.04, p > 0.05]. There were significantly smaller values of lateralization offset for 10 compared to 100 Hz [t(144) = −2.68, p < 0.05], but there was no significant difference between 10 and 30 Hz [t(144) = −0.48, p > 0.05], 10 and 300 Hz [t(144) = −0.22, p > 0.05], 30 and 100 Hz [t(144) = −2.20, p > 0.05], 30 and 300 Hz [t(144) = 0.26, p > 0.05], or 100 and 300 Hz [t(144) = 2.46, p > 0.05]. This result may have been driven by a subset of listeners (addressed in Sec. II C 2 b). In fact, when this analysis was repeated excluding the listeners with these three extreme results (CAD, CAI, and SCT), there was no significant effect of stimulus [F(1,105) = 0.25, p > 0.05], rate [F(3,105) = 2.33, p > 0.05], or group [F(1,15) = 0.80, p > 0.05]. There was a significant rate × group interaction [F(3,105) = 3.16, p < 0.05] and three-way interaction [F(3,105) = 3.32, p < 0.05], but no significant interaction between stimulus × rate [F(3,105) = 2.29, p > 0.05] or stimulus × group [F(1,105) = 0.06, p > 0.05]. Thus, if the mean lateralization offset changed between stimuli or groups, it was not robust as the results of the ANOVA changed drastically when these three listeners were removed.

To explore whether the variance of the lateralization offset differed between groups, Bartlett's test for heterogeneity was completed for each stimulus (CAPT or MAPT) and rate with Bonferroni corrections to account for multiple comparisons. The results indicated that variance of lateralization offset was significantly different between YNH and BICI listeners for CAPTs of 10 Hz [K²(1) = 13.18, p < 0.05], but not 30 Hz [K²(1) = 2.27, p > 0.05], 100 Hz [K²(1) = 5.95, p > 0.05], or 300 Hz [K²(1) = 0.09, p > 0.05]. There was no evidence of significant differences in variance between YNH and BICI listeners for MAPTs of 10 Hz [K²(1) = 2.10, p > 0.05], 30 Hz [K²(1) = 0.18, p > 0.05], 100 Hz [K²(1) = 0.28, p > 0.05], or 300 Hz [K²(1) = 0.46, p > 0.05]. Together, these results suggest that the variance of lateralization offset was similar across both groups for the majority of the stimuli in this experiment.

b. ILD resulting in centered image.

It was also of interest to determine the ILD resulting in a centered image for the participant. During lateralization of ITDs in BICI listeners, this ILD was applied to all stimuli with the intent to produce a centered image at 0-μs ITD, which was mostly successful with a few exceptions (Fig. 1). First, we performed an analysis to account for the carrier signal and averaged across rates for CAPTs and MAPTs. The results indicated that the ILD resulting in a centered image was not significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(2,156) = 1.09, p > 0.05], group [F(1,22) = 0.25, p > 0.05], or a stimulus × group interaction [F(2,156) = 1.11, p > 0.05]. Together, these results suggest that there was not a difference in the mean ILD corresponding to a centered image between stimuli or groups when averaged across rate.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. The results indicated that the ILD resulting in a centered image was neither significantly affected by stimulus [F(1,126) = 1.52, p > 0.05] nor group [F(1,18) = 0.76, p > 0.05], but was significantly affected by rate [F(3,126) = 4.83, p < 0.01]. There was a significant interaction of stimulus × rate [F(3,126) = 3.18, p < 0.05] and rate × group [F(3,126) = 5.59, p < 0.01], but not stimulus × group [F(1,126) = 1.86, p > 0.05]. The ILD resulting in a centered image was also significantly affected by a three-way interaction of stimulus × rate × group [F(3,126) = 4.85, p < 0.01]. Data in Fig. 5(B) show that the ILD resulting in a centered image with 95% within-subject confidence intervals. Together, these results suggest that the mean ILD resulting in a centered image varied depending upon stimulus, rate, and group.

To explore whether the variance of the ILD resulting in a centered image differed between groups, Bartlett's test for heterogeneity was completed for each stimulus (CAPT or MAPT) and rate with Bonferroni corrections to account for multiple comparisons. The results indicated that variance of ILD resulting in a centered image was significantly different between YNH and BICI listeners for CAPTs of 10 Hz [K²(1) = 19.45, p < 0.05], 30 Hz [K²(1) = 7.86, p < 0.05], 100 Hz [K²(1) = 18.90, p < 0.05], and 300 Hz [K²(1) = 7.84, p < 0.05]. There were no significant differences in variance between YNH and BICI listeners for MAPTs with any pulse rate (10 Hz [K²(1) = 1.84, p > 0.05], 30 Hz [K²(1) = 0.59, p > 0.05], 100 Hz [K²(1) = 1.60, p > 0.05], or 300 Hz [K²(1) = 0.85, p > 0.05]). Together, these results show that the variance of ILD corresponding to a centered image varied between YNH and BICI listeners for CAPTs.

4. Standard deviation of lateralization responses

The standard deviation in lateralization averaged across ITD or ILD was also computed and is presented in Fig. 6. Standard deviation of lateralization response reflects uncertainty about the perceived location with a given ITD or ILD. However, the range of lateralization responses also changes depending upon the mean lateralization because as the mean lateralization response increases in magnitude, there is a smaller range of responses that listeners can choose from (see Supplementary Figs. 5 and 6 for YNH listeners' data³). That is, for sounds lateralized far to the left or right, the response range decreases artificially. A logit transformation was completed to attempt to account for decreasing range at more highly lateralized locations. First, 10 was added to each raw lateralization response, which was then divided by 20 such that the updated lateralization responses were bounded between 0 and 1. Next, the logit transform was computed,

logit (p) = log (\frac{p}{1 - p}),

(2)

where p was the transformed lateralization response bounded between 0 and 1. The standard deviations of logit transformed data are presented in Supplementary Figs. 7 and 8, and demonstrate that this technique accounted for dependence of standard deviation on far lateral responses.³ Accordingly, logit transformed lateralization data were used to compare standard deviation in the perceived location of CAPTs and MAPTs by group and rate.

FIG. 6. — (Color online) Logit-transformed standard deviation for (A) ITDs and (B) ILDs with CAPTs (left panels) and MAPTs (right panels). Closed shapes represent YNH listeners and open shapes represent BICI listeners. The x axis corresponds to the pulse rate (for CAPTs) or modulation rate (for MAPTs). The y axis corresponds to the logit-transformed standard deviation of the lateralization response, averaged across ITDs or ILDs tested. For more details, see Sec. II B 4, Eq. (2), and Supplementary Figs. 4 and 5 (see footnote 3). Error bars represent 95% within-subject confidence intervals as described by Baguley (2012) and can be interpreted as informal *post hoc* comparisons at α = 0.05. Individual data are shown offset to the right of the mean.

a. ITDs.

First, we performed an analysis to account for the carrier signal and averaged across rates for CAPTs and MAPTs. The results indicated that standard deviation of lateralization response to ITDs was significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(2,2042) = 3.44, p < 0.05] and group, where YNH listeners showed greater standard deviations than BICI listeners [F(1,19) = 5.60, p < 005]. There was not a stimulus × group interaction [F(2,2042) = 2.37, p > 0.05]. In both groups, standard deviation was neither significantly different between CAPTs and MAPTs [t(2042) = −1.55, p > 0.05] nor CAPTs and carrier [t(2042) = 1.60, p > 0.05], but was significantly greater for MAPTs compared to carrier [t(2042) = 2.47, p < 0.05]. Together, these results suggest that YNH listeners had more variable lateralization responses than BICI listeners, and that MAPTs resulted in greater variability in responses than the carrier.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. The results indicated that standard deviation in responses was not significantly affected by stimulus [F(1,1829) = 1.39, p > 0.05], but was significantly affected by rate [F(3,1829) = 4.15, p < 0.01] and group [F(1,19) = 4.66, p < 0.05]. There was a significant interaction of stimulus × rate [F(3,1829) = 12.52, p < 0.0001], but not of stimulus × group [F(1,1829) = 0.92, p > 0.05] or rate × group [F(3,1829) = 0.96, p > 0.05]. Standard deviation of lateralization was significantly affected by a three-way interaction of stimulus × rate × group [F(3,1829) = 7.01, p < 0.001]. Figure 6(A) shows 95% within-subject confidence intervals for the standard deviation in each of these conditions. The three-way interaction was driven by differences in BICI listener performance according to stimulus. Standard deviation of lateralization responses was relatively consistent across rate in YNH listeners while BICI listeners had opposite patterns across rate for either stimulus. Standard deviation of lateralization responses was greatest at 10 Hz for CAPTs and smallest for MAPTs at 10 Hz. The largest differences between groups in standard deviation of lateralization responses occurred for the rates and stimuli where BICI listeners had the smallest standard deviation of lateralization responses. Together, these results suggest that YNH listeners had more variable responses than BICI listeners, but the effect of rate differed by stimulus depending upon group.

b. ILDs.

First, we performed an analysis to account for the carrier signal and averaged across rates for CAPTs and MAPTs. The results indicated that standard deviation of lateralization response to ILDs was neither significantly affected by the stimulus (CAPTs, MAPTs, or carrier) [F(2,1954)= 2.31, p > 0.05] nor group [F(1,19) = 1.73, p > 0.05], but was significantly affected by a stimulus × group interaction [F(2,1954) = 8.91, p < 0.001]. For BICI listeners, standard deviation of lateralization responses was significantly greater for CAPTs compared to MAPTs [t(1954) = 3.81, p < 0.01], but was neither significantly different for CAPTs compared to the carrier [t(1954) = 1.17, p > 0.05] nor MAPTs compared to the carrier [t(1954) = −1.20, p > 005]. For YNH listeners, the standard deviation of lateralization responses was not significantly different for CAPTs compared to MAPTs [t(1954) = −1.82, p > 0.05], for CAPTs compared to the carrier [t(1954) = −1.87, p > 0.05], or for MAPTs compared to the carrier [t(1954) = −0.72, p > 0.05]. Together, these results suggest that the standard deviation of lateralization responses was greater for CAPTs compared to MAPTs, but only in BICI listeners.

In the second ANOVA, it was of interest to explore the effects of rate. Therefore, the carrier was excluded from the analysis. Effects of stimulus (CAPTs or MAPTs), rate, and group were evaluated. The results indicated that standard deviation of lateralization responses was neither significantly affected by stimulus [F(1,1727) = 3.75, p > 0.05], rate [F(3,1727) = 1.71, p > 0.05], nor group [F(1,19) = 1.42, p > 005]. There was a significant interaction of stimulus× rate [F(3,1727) = 2.27, p < 0.05] and stimulus × group [F(1,1727) = 16.63, p < 0.0001], but there was no significant interaction of rate × group [F(3,1727) = 0.63, p > 0.05]. Standard deviation of lateralization was significantly affected by a three-way interaction of stimulus × rate× group [F(3,1727) = 3.18, p < 0.05]. Figure 6(B) shows 95% within-subject confidence intervals for the standard deviation in each of these conditions. The three-way interaction was driven by differences in BICI listeners' performance according to stimulus and rate. Together, these results suggest that BICI listeners had lower standard deviation of lateralization responses on average for MAPTs compared to CAPTs, with a small decrease with increasing rate in MAPTs, but YNH listeners showed no consistent change in performance.

C. Discussion

This experiment investigated the effects of rate and ITDs or ILDs presented in the temporal fine-structure vs envelope on lateralization of pulse trains in YNH and BICI listeners. It was hypothesized that ITDs and ILDs of amplitude-modulated signals would result in less consistent lateralization responses in BICI compared to YNH listeners because of decorrelation introduced between the ears and poorer binaural temporal processing in BICI listeners. We further hypothesized that ITDs would have a smaller impact on lateralization for BICI compared to YNH listeners because of binaural temporal processing limitations. There are many ways to characterize lateralization performance. We used the lateralization range, bias, and standard deviation of lateralization responses to describe the impact and reliability of ITDs and ILDs.

1. Lateralization of ITDs

a. Lateralization range.

The effects of rate on lateralization range differed depending upon whether pulse trains were constant or modulated in amplitude, and patterns across rate were consistent between groups for the lateralization of ITDs [individual data, Fig. 1; average data, Fig. 3(A) and 3(B); summarized in Fig. 4(A)]. In general, CAPTs resulted in lateralization ranges that were similar at rates ≤100 pps, but decreased as rate increased to 300 pps. This pattern of performance was demonstrated by YNH and BICI listeners, with a lower overall mean in BICI listeners. These results are consistent with previous studies in BICI listeners using direct stimulation to present low-rate CAPTs, showing a decrease in ITD sensitivity for 300 pps and above (Kan and Litovsky, 2015; Laback et al., 2015).

The effects of rate on lateralization range varied depending upon whether ITDs were presented in the fine-structure or envelope. Lateralization range for MAPTs was non-monotonic across rate, resulting in the largest lateralization range for 100 Hz [Fig. 4(A), right panel]. Previous experiments with YNH listeners have shown that ITD sensitivity is best for envelope ITDs with a high-frequency carrier modulated at a rate of 128 Hz (Bernstein and Trahiotis, 2002, 2009). Maximal sensitivity to sinusoidal envelope ITDs at 128 Hz is consistent with experiments investigating the role of off-time (the gap between stimuli in a MAPT, unlike inter-click interval which gives the amount of time between each click onset) and envelope slope on sensitivity to ITDs in NH and BICI listeners (Bernstein and Trahiotis, 2002, 2009; Dietz et al., 2016; Klein-Hennig et al., 2011; Laback et al., 2011). In these experiments, similar results were found between NH and BICI listeners suggesting that the off-time between cycles is the primary predictor of ITD sensitivity for MAPTs (Laback et al., 2011). Bernstein and Trahiotis (2011) provided a phenomenological model for NH relating amplitude modulation rate and depth to lateralization responses based on peripheral filtering and smoothing of the signal. Dietz et al. (2016) provided a physiological explanation in NH for improved ITD coding with sharp envelope attack by demonstrating that stimuli with sharp onsets provide greater ITD sensitivity in a physiologically based model and recordings from inferior colliculus cells. Our results are consistent with these interpretations and extend them by showing that the effects of rate differed for CAPTs and MAPTs within the same YNH and BICI listeners. Surprisingly, the pattern of lateralization range across rate for MAPTs was very similar between YNH and BICI listeners. Our result contrasts with experiments showing that sensitivity to ITDs in discrimination tasks is lower for MAPTs compared to CAPTs in electrical stimulation (van Hoesel, 2007; van Hoesel et al., 2009). Our results further suggest that rate affects NH and BICI listeners in a similar way, with lateralization ranges being smaller overall in BICI listeners for CAPTs. It is possible that MAPTs are processed by both groups in a similar way, unlike CAPTs, which appear to exhibit limitations at lower rates for BICI than YNH listeners. It may be that the physiological processing of MAPTs is more robust to the factors affecting BICI listeners. However, it should be noted that this effect could be driven in part by the difference in spectral representations between YNH listeners (who were presented sinusoidally amplitude-modulated tones) and BICI listeners (who were presented with true MAPTs).

To rule out that lateralization differences could have been driven by a difference in fusion (i.e., the presence of a single sound source), Supplementary Fig. 9 shows the correlation between lateralization range and proportion of single sound sources by individual.³ Recall that the most prominent sound source was used to determine lateralization ranges. Supplementary Fig. 9 demonstrates the lack of correlation between lateralization and fusion within-subjects for both YNH and BICI listeners, which may be driven by a majority of “one sound” responses. If there were a relationship between the stimulus parameters tested in this study and the rate of fusion, a smaller proportion of single sound sources would have been expected.

Finally, lateralization range decreased as rates increased from 100 to 300 Hz by a greater magnitude for CAPTs than MAPTs. This result suggests that the upper-rate limit with acoustical and electrical pulse trains imposed to binaural processing may differ between CAPTs and MAPTs. For example, some sensitivity to ITDs in CAPTs at high pulse rates has been restored by introducing binaurally-coherent jitter to CAPTs (Goupell et al., 2010; Hancock et al., 2012). The mechanisms involved with decreasing sensitivity for CAPTs above 300 Hz remain elusive and will be discussed in greater detail in Sec. IV.

b. Standard deviation of lateralization responses.

One landmark feature of free-field localization performance in BICI listeners is large amounts of variability in their responses (e.g., Jones et al., 2014). Therefore, it was of interest to calculate a metric to capture the variability in perceived locations of stimuli and determine whether this changed across conditions or groups. Standard deviation of lateralization responses was averaged across ITD and logit transformed to account for differences due to higher mean lateralization responses (see Supplementary Figs. 7 and 8).³ Figure 6 shows differences in the effect of rate between CAPTs and MAPTs. In BICI listeners, the standard deviation of lateralization response decreased slightly as rate increased for MAPTs but was relatively consistent across rate for CAPTs. Surprisingly, BICI listeners were less variable in their responses compared to YNH listeners even after the logit transformation [Fig. 6(A)]. There are several reasons this result might occur. BICI listeners tested in this experiment were highly experienced and many had extensive experience in psychophysical research. Related to this, BICI listeners may be highly motivated to perform well in comparison to YNH listeners, for whom the results of research have relatively low impact. Motivation and experience are two factors that are difficult to control in these types of studies (Goupell, 2015). Finally, lateralization bias at 0-μs ITD was corrected in BICI listeners but not YNH listeners.

The relationship between rate and standard deviation of lateralization response varied depending upon the stimulus and group. The three-way interaction between rate, stimulus, and group indicated that BICI listeners had decreasing standard deviation of lateralization responses with increasing rate for CAPTs, and YNH listeners were consistent, with the largest differences between groups occurring at 100 and 300 Hz [Fig. 6(A), left panel]. However, for MAPTs, the largest difference in standard deviation between groups occurred at the lowest rate (10 Hz) and the standard deviation was less consistent across rate within either group. It is clear from these results that the standard deviation of lateralization response across rate and amplitude modulation changed significantly across groups and differed from lateralization range results. Ultimately, changes in the standard deviation of lateralization responses imply subtle changes in the reliability of spatial cues across rate and stimulus (CAPT vs MAPT), though this requires additional and more extensive investigation.

While in general BICI listeners had smaller standard deviations in their lateralization responses, the average lateralization was quite variable across BICI listeners. For example, some BICI listeners required extremely large ITDs (up to 2500 μs) before reaching the full extent of lateralization (Fig. 1). One listener (CAZ) lateralized CAPTs using ITDs at extremely high rates (for the carrier at 1000 pps; Fig. 1). In some studies of BICI listeners, a subset of listeners exhibit highly accurate temporal processing abilities where the reasons for this improved sensitivity relative to other listeners remains unclear (e.g., Goldsworthy and Shannon, 2014; Kong and Carlyon, 2010; Noel and Eddington, 2013).

2. Lateralization of ILDs

a. Lateralization range.

There were substantial differences between groups in the effects of stimulus and rate on the lateralization range of ILDs [Fig. 4(B)]. As rate increased, so did the lateralization range of ILDs applied to CAPTs in BICI listeners. It is possible that this effect was driven by a change in loudness associated with increasing rate. However, in studies with NH listeners, overall level does not have a large effect on ILD discrimination (Dietz et al., 2013). Smaller lateralization ranges for BICI compared to YNH listeners could instead relate to differences in the dynamic range for BICI listeners. For example, a change of 5 CUs might correspond to a different change in loudness for one subject compared to another. Most YNH listeners reached their maximum lateralization responses within the ILDs tested, where their lateralization curve reached the upper and lower asymptotes [Fig. 3(D), Supplementary Fig. 2]. This contrasted from BICI listeners, many of whom did not reach upper or lower asymptote of the lateralization curve (Fig. 2). However, an ILD of ±15 dB would fall far within the dynamic range of NH listeners, yet still resulted in reaching upper and lower asymptotes of the lateralization curve. Finally, while low-rate stimuli would increase reliance on high temporal precision of the inputs from either ear to correctly code for the ILD, higher rates for CAPTs and the higher rate carrier for MAPTs could improve ILD coding in listeners with poorer binaural temporal processing.

Lateralization range for ILDs applied to CAPTs was slightly smaller than MAPTs for YNH listeners, but consistent across rate. This result is opposite of the trend with ITDs, where CAPTs had larger lateralization ranges. The role of rate in ILD processing for YNH listeners is debated, but it appears that YNH listeners are most sensitive to ILDs at 400 pps, with marginally poorer ILD discrimination thresholds (∼0.5 dB) at higher or lower rates (Laback et al., 2017).

b. Bias.

Bias in the lateralization of ILDs was measured because previous experiments suggested that BICI listeners suffer from substantial biases (Fitzgerald et al., 2015; Goupell et al., 2013a; Stakhovskaya and Goupell, 2017). The direction of bias differed from listener to listener (Fig. 5). Unsurprisingly, there was no difference in the mean bias between YNH and BICI listeners when measured via the lateralization offset in the lateralization response (once extreme cases were removed) or the ILD corresponding to a centered image (Fig. 5). Note that the range of lateralization offsets for BICI listeners was often similar to or greater than that of YNH listeners. Tests for heterogeneity of variance revealed that there was significantly greater variability in the ILD corresponding to a centered image for all rates with CAPT stimuli.

Stakhovskaya and Goupell (2017) investigated the lateralization of high-rate 1000-pps CAPTs in BICI listeners and noise bands in YNH listeners. It should be noted that these stimuli result in different spectro-temporal representations, so it is difficult to directly compare between groups. Their study focused on the lateralization of ILDs with single or multiple electrodes (or noise bands in the case of YNH listeners). Their results indicated that the ILD corresponding to a centered image had a larger range for BICI listeners when presented on a single electrode, and more listeners in the present study perceived centered images with ILDs closer to 0 CUs. The YNH listeners in their study had less variable ILDs corresponding to a centered image compared to CAPTs in the present study. MAPTs were not tested in the study by Stakhovskaya and Goupell (2017) and the results using MAPTs in this study resulted in less variability of ILDs corresponding to a centered image, such that there was no difference between YNH and BICI listeners. Thus, our results are consistent with the previously published literature using CAPTs and imply that BICI listeners have more variable biases compared to YNH listeners. Our results further demonstrate that these trends hold for lower-rate CAPTs, not just high-rate pulse trains.

The ILDs corresponding to a centered image from BICI listeners in this study are similar to those presented in Fitzgerald et al. (2015) and Stakhovskaya and Goupell (2017). Despite small sample sizes in each of those studies, the results from the present study and previous studies suggest that BICI listeners tend to have significantly greater variability in measurements of bias, specifically the ILD resulting in a centered image, compared to YNH listeners. The results of the present study extend previous studies by demonstrating that bias associated with lateralization of ILDs appears to be specific to CAPTs, and there tends to be smaller magnitude and less variable biases for MAPTs compared to CAPTs in BICI listeners.

There was a significant effect of rate on the lateralization offset due to the difference between 10 and 100 Hz. We suspect that this effect was the result of chance error in the data as it was not consistent across individuals and may have been driven primarily by a small number of individuals in the CAPT condition.

c. Standard deviation of lateralization responses.

The standard deviation of lateralization responses for ILDs indicated differential effects of stimulus and group. The standard deviation of lateralization responses was lower for BICI than YNH listeners, but only with MAPTs [Fig. 6(B)]. This result suggests that the ILD was more reliable for BICI listeners when listening to MAPTs compared to CAPTs. These results are consistent with the lateralization range results within BICI listeners, implying a relationship between the reliability of ILD and the extent to which the ILD is lateralized within groups.

The standard deviation of lateralization response was similar for CAPTs and MAPTs in YNH listeners. Like the results from standard deviation of responses in ITD lateralization, these results reveal different effects than lateralization range within BICI listeners. Thus, standard deviation of the lateralization response may provide insights into new aspects of perception than other metrics associated with lateralization performance.

3. Conclusions and limitations

Lateralization range for ITDs decreased at rates above 100 Hz [Fig. 4(A)] suggesting decreased impact of ITDs on lateralization for CAPTs and MAPTs in YNH and BICI listeners. Conversely, rates below and above 100 Hz for MAPTs, but not CAPTs, imply decreased impact of ITDs in both groups [Fig. 4(A)]. This result agrees with previous reports investigating the relationship between the shape of the envelope and its effect on ITD sensitivity in YNH and BICI listeners (Bernstein and Trahiotis, 2002, 2009; Dietz et al., 2016; Klein-Hennig et al., 2011; Laback et al., 2011). On average, BICI listeners had smaller lateralization range compared to YNH listeners (Fig. 4). Critically, lateralization range for ITDs worsened for rates above 100 Hz for CAPTs and MAPTs in YNH and BICI listeners, suggesting a shared mechanism dictating performance at high rates, though the magnitude of this change differed between the two stimuli. The physiological basis of limitations to ITD processing at high rates is still being debated, but previous experiments have demonstrated a decrease in the ability to represent temporal fluctuations at the lateral superior olive (Joris and Yin, 1998), where envelope ITDs are computed; Joris and Yin, 1995). Lateralization range when presented with ILDs suggests that increasing rate resulted in increased impact of ILDs in CAPTs for BICI listeners, but little or no effect on MAPTs. In YNH listeners, lateralization range was similarly large across rate for CAPTs and MAPTs, suggesting that ILDs were useful for YNH listeners in all conditions within this experiment. On average, BICI listeners had smaller lateralization range compared to YNH listeners.

It was hypothesized that interaural decorrelation due to differences in loudness growth between the ears would result in greater standard deviation of lateralization responses in BICI compared to YNH listeners. Instead, the standard deviation of lateralization responses was lower in general for ITDs in BICI compared to YNH listeners. Similarly for ILDs, the standard deviation in lateralization response was significantly lower for BICI compared to YNH listeners for MAPTs, but not CAPTs. Though the reason for this is not exactly clear, one explanation is motivation and training effects. If decorrelation is introduced for BICI listeners, it is more likely that the perceived image width would be affected and image width might provide a more sensitive perceptual index of interaural decorrelation (Whitmer et al., 2014).

There were several limitations to the present study. First, no loudness balancing was completed across different rates for CAPTs and MAPTs. This may have introduced undesired differences in stimuli across rate for CAPTs. In BICI listeners, loudness is relatively invariant across rate for amplitude-modulated stimuli (Chatterjee and Oberzut, 2011; Kreft et al., 2010). In general, as level decreases, ITD and ILD sensitivity decrease for NH listeners (Dietz et al., 2013). One way to improve future methods would be to use ILDs that are centered and determined by a loudness growth function for each subject, and balance loudness across rates for CAPTs. Second, differences in stimulus bandwidth may have played a role in these experiments. CAPTs for YNH listeners resulted in 1.5-mm bandwidths, smaller than the estimated 3- to 4.6-mm monopolar CI stimulation bandwidth (Goupell, 2015; Nelson et al., 2008). Additionally, sinusoidally amplitude-modulated tones used in place of MAPTs for NH listeners resulted in substantially smaller bandwidths. Finally, lateralization of ITDs has been shown to change with age in NH listeners (Babkoff et al., 2002; Baumgärtel et al., 2017). Thus, the difference in lateralization performance, particularly the differences in lateralization range between YNH and BICI listeners, could be due in part to poorer temporal processing associated with aging (for review, see Gordon-Salant et al., 2010). The effect of age on lateralization of ITDs and ILDs was explored in experiment 2.

III. EXPERIMENT 2: LATERALIZATION OF CAPTS IN ONH LISTENERS

The results from experiment 1 indicated different performance between YNH and BICI listeners. One important and addressable confound in experiment 1 was the difference in age between groups. That is, reduced lateralization range in BICI listeners may be due to electrical vs acoustical stimulation, poorer temporal processing due to age, or an interaction between both. Thus, in this experiment, a group of ONH listeners completed a subset of conditions in the first experiment to determine the effects of age on binaural temporal processing. The primary conditions of interest were those that exhibited the largest differences between YNH and BICI listeners in experiment 1 (ITD lateralization of CAPTs at rates of 30, 100, and 300 pps). Given the older age of these listeners, we wanted to avoid extensive testing over multiple sessions to complete the experiment.