Auditory Motion Perception by Bilateral Cochlear Implant Users With a Sound Coding Strategy that Synchronizes Peak-Picking Across the Ears

Abstract

Bilateral cochlear implant (BiCI) users do not perform as well as typical hearing listeners when discriminating the direction of sound motion. This is likely due to the lack of synchronization between two independently operating sound processors. The use of bilaterally linked processors can improve the timing of electrical stimulation across the ears. However, these improvements are potentially counteracted by spectral peak-picking in sound coding strategies that may activate different electrodes across the ears, thereby reducing the fidelity of the binaural cues that BiCI users rely on for sound localization. To improve interaural synchrony, a bilateral peak-picking strategy that guarantees coordinated stimulation across the ears was developed and tested using the CCi-MOBILE, a bilaterally linked research processor. The auditory motion tracking abilities of nine BiCI users were measured using the bilateral peak-picking strategy and compared to performance with commercially available, unsynchronized processors. Results showed no effect of bilateral peak-picking. However, a small benefit was observed when tracking the range of motion of a sound when listening with bilateral synchronization which appears to be due to reduced interaural level difference changes with auditory motion. These findings suggest that dynamic auditory perception with BiCI may likely need bilaterally-synchronized hardware.

Introduction

Cochlear implants (CIs) are auditory prostheses that provide access to hearing for individuals who are deaf. Bilateral cochlear implant (BiCI) users (individuals fitted with a CI in each ear) generally demonstrate improved spatial hearing capabilities compared to the use of a single implant (Grantham et al., 2007; Litovsky et al., 2006; van Hoesel & Tyler, 2003). Having two processors gives BiCI users partial access to the binaural cues that typical hearing (TH) listeners use for spatial hearing, namely interaural time differences (ITDs) and interaural level differences (ILDs) (Blauert, 1997; Dunn et al., 2008; Kan & Litovsky, 2015; Seeber & Fastl, 2008). While much research on understanding the spatial hearing abilities of BiCI users has focused on relatively simple tasks such as identifying the location of a stationary sound source, recent studies have investigated the performance of BiCI listeners in auditory motion perception tasks (Archer-Boyd & Carlyon, 2019, 2021; Dwyer et al., 2021; Moua et al., 2019).

Moua et al. (2019) investigated the auditory motion perception abilities of BiCI listeners. Their work revealed that BiCI listeners were worse at discriminating between moving and stationary sounds as well as TH listeners. Furthermore, BiCI listeners were far poorer than TH listeners in accurately reporting the location and range of motion of a sound. In that study, a “virtual acoustic space” methodology was used so that both TH and BiCI listeners were presented with the same stimuli containing controlled binaural cues. Binaural recordings were generated using microphones in the ear canals of a KEMAR mannekin (Knowles Electronics Manikin for Acoustic Research, G.R.A.S. Sound Vibration, Holte, Denmark). The recordings were of “moving sounds” simulated using vector-based amplitude panning (Pulkki, 1997) through an array of loudspeakers that had a 5° resolution. The use of binaural recordings controlled for variations across listener groups such as the difference in microphone location on CIs vs the ear canal in typical hearing. However, binaural recordings using a manikin also denied listeners the potential advantage of using head movements to improve localization performance. Data from Pastore et al. (2018) suggest that BiCI listeners could benefit from using head motion to improve sound localization and resolve front-back confusions. Therefore, there is a need for investigations of auditory motion perception that are truly in the free-field in real-time to determine whether the findings of Moua et al. (2019) generalize to more realistic listening conditions.

There are likely many other reasons for the poorer auditory motion abilities of BiCI users, particularly relating to issues with electrically stimulated binaural hearing (Kan & Litovsky, 2015). One consideration is that BiCIs listeners are typically fitted with two independent, unsynchronized processors. Unsynchronized processors are unlikely to provide consistent ITDs in either the fine structure or envelope, as shown by Gray et al. (2021). Kan et al. (2018) also suggested that the acoustics of the head can interact with the sound coding strategies that employ peak-picking to render ILDs and envelope ITDs not as useful in current clinical processors as compared to direct stimulation. Furthermore, unsynchronized processors can lower interaural synchrony (IS). Reduced IS disrupts spatial release from masking because of unlinked sound processing and reduces access to consistent ITDs because of mismatched quantization across the ears (Francart et al., 2015; Gajecki & Nogueira, 2020; Kan & Litovsky, 2015). Dennison et al. (2022) found that unsynchronized clinical processors reduced the IS of the stimulation patterns of BiCI participants and tested whether bilaterally synchronized hardware, which removed the jitter across the ears and improved IS, has the potential to yield improvements in sound localization and spatial release from masking. However, the use of the Advanced Combination Encoder (ACE) strategy for spectral peak-picking in their testing was likely masking the benefits of using synchronized hardware because it also reduced IS. It may be that coordinated peak-picking across the ears is needed to realize some of the true benefits of synchronized hardware.

The benefits of coordinated peak-picking across the ears were investigated by Gajecki and Nogueira (2020). The authors tested simulations of synchronized vs. unsynchronized processors and evaluated a bilaterally linked band selection technique for speech-in-noise listening. Their algorithm used an estimate of the most favorable ear and selected peaks based on the channels activated in that ear. Target speech was presented from the center while masking noise was presented to the “better” ear. Improvements were found in speech understanding in conditions that used either linked processors or synchronized peak selection. In a follow up study, the authors tested a similar linked “N of M”, or spectral peak-picking, strategy by presenting pre-processed stimuli from virtual loudspeakers spaced at ±5°, ±20°, ±70°, and ±90°. The results suggested that linked processing could have an impact on sound localization at the most lateral locations, but a group level effect was only seen when an exaggerated ILD was applied (Gajecki & Nogueira, 2021). Though these findings may show some promise, all testing was conducted using pre-processed stimuli and delivered via direct stimulation, rather than through processors. As with Moua et al. (2019), it is unclear how well the strategy would translate to more realistic testing conditions with processors running in real-time.

We hypothesized that the inaccuracies observed in the perception of auditory motion of BiCI users in Moua et al. (2019) were caused by low interaural synchrony (the result of unsynchronized processors disrupting the timing of pulses) and poor ILD encoding caused by uncoordinated channel selection (peak-picking) across the ears. If our hypothesis is true, we predict that by using synchronized processors and coordinating which electrode channels are activated in each ear by the ACE processing strategy, BiCI users would demonstrate improvements in sound localization and discrimination of auditory motion. To test this hypothesis, we developed a bilaterally linked peak-picking strategy that would coordinate the selection of channels across the ears in real-time. Our strategy, named Bilateral ACE (BACE), was implemented and tested using the CCi-MOBILE, a bilaterally synchronized research processor developed at the University of Texas at Dallas (Ghosh et al., 2022). The CCi-MOBILE enables rapid development and evaluation of real-time CI strategies (Borjigin et al., 2025; Ghosh & Hansen, 2023; Kan & Meng, 2021) with human participants. BACE was compared with an implementation of ACE on the CCi-MOBILE to compare the benefit of bilaterally linked peak-picking. Testing was also done with unsynchronized clinical processors to assess the benefit of synchronized vs unsynchronized processors for auditory motion perception.

Methods

Processing Strategies

Figure 1 shows a block diagram of the signal processing steps of the ACE and BACE sound coding strategies as implemented on the CCi-MOBILE. The CCi-MOBILE operates at a sampling rate of 16 kHz, with 8 ms (128 samples) input and output buffers. In our implementation of ACE on the CCi-MOBILE, the input buffer is processed in 128-sample frames, with a hop size of 1 ms. A short-time Fourier transform (STFT) analysis is applied to each frame. The frequency bins of each frame of the STFT are weighted and consolidated into 22 channel bands. The magnitude of each channel is calculated to estimate the energy in that channel.

Figure 1.

Signal block diagram for ACE and BACE is shown on the left. In ACE, the left and right ears are independently processed but using the same steps. An example output for each sound coding strategy is shown on the right

To reduce the effects of current spread on speech understanding, ACE picks the channels with the highest energy to activate in each frame. To achieve this, the energy in each channel is sorted and the N channels with the most energy (“largest peaks”) are selected. The amplitudes of the N peaks are then logarithmically compressed to fit into the patients’ electrical dynamic range and used as amplitudes for the electrical pulse train for each channel. Because peak-picking in ACE is done independently in each processor, the number of channels N is dependent on the signal energy at that instant and can be different in each processor. This may lead to incorrect ILDs being presented on channels where stimulation only occurs on one side (Gray et al., 2021; Kan et al., 2018).

To overcome the problems of ACE, BACE was designed to maintain interaural coherence and ILDs at the peak-picking stage. In BACE, the N highest peaks in both ears are pooled together (to give a pool of peaks PN that is between N and 2N), and then a subset of N peaks are chosen from the pool PN. Preference is given to low-frequency channels (electrodes in Cochlear-branded implants are numbered such that higher numbers are assigned to the low-frequency channels) for the final subset of peaks because the low frequencies contain important speech information, ITDs for locating sounds, and ILDs important for near-field sound localization (Brungart et al., 1999; Kan et al., 2009). As an example of the peaks chosen by BACE where N = 8, suppose the 8 highest channels in each ear (sorted from least intense to most intense) are [4,5,6,12, 7, 8,9,13] and [1,2,3,4,5,6,7,9] for the left and right ears, respectively. The union of the set of peaks would then be [13,12,9,8,7,6,5,4,3,2,1], and BACE would choose the first 8 (shown in bold font) of these to stimulate.

The right-hand side panels of Figure 1 shows an example of electrode outputs from ACE and BACE for a sound located to the left. It can be seen that for ACE, there is stimulation in the low frequency electrodes (higher numbered) in the right ear but not in the left, suggesting that there will be rightward-pointing ILD being perceived for some portion of the stimulus. During the equivalent period in BACE, stimulation also occurs in the left ear to maintain the correct ILD.

To confirm that BACE could improve interaural synchrony beyond ACE, electrode outputs (also known as electrodograms) were analyzed to estimate the IS of the electrical pulses that were being encoded and delivered by each strategy. To generate the electrodograms, recordings of the test stimuli were made by placing the CCi-MOBILE microphones on the earpieces of a KEMAR (Knowles Electronics Manikin for Acoustic Research, G.R.A.S Sound and Vibration, Holte, Denmark) manikin. The manikin was placed at the center of the loudspeaker array in the room used for behavioral testing. Five recordings were made for each stimulus condition (described in Stimulus and Apparatus) using the CCi-MOBILE microphones and stored as 16-bit WAV files at a sampling rate of 16 kHz. These sound recordings captured the microphone and head related transfer functions.

The recordings of the stimuli were post-processed using the two sound coding strategies described above. To simulate a condition where a participant is wearing unsynchronized clinical processors, the pulses in each channel at the output of the ACE strategy were jittered based on the jitter measurements from Dennison et al. (2022), e.g. the timing of each pulse was adjusted according to independent draws from the random variable $X=u\left[0,1\right]\frac{1}{2}\frac{1}{900}$ where $u\left[n\right]$ follows a discrete uniform distribution. The output of the processor simulation was a pair of vectors for scheduling pulses, representing which electrodes were to be stimulated and their pulse amplitudes. This output was transformed into a “left” and “right” set of twenty-two signals, one for each channel, by convolving a biphasic pulse shape with an impulse at the time each pulse was scheduled. Biphasic pulses had a phase duration of 25 μs and interphase interval of 8 μs, which is the same duration as a typical electrical stimulation pulse of a Cochlear-branded implant.

Interaural synchrony (IS) was calculated for each channel as the maximum of the normalized cross-correlation function between left and right pulse sequences (Goupell & Litovsky, 2015) with the following equation:

$\rho \left(n,\theta ,\tau \right)=\frac{{\mathrm{\Sigma }}_t{x}_L\left(n,\theta ,t\right)x_R\left(n,\theta ,t\right)\left(t+\tau \right)}{E_L{E}_R},$

where $n$ was the electrode channel number, $\theta$ was the start location of the presented stimulus, $\tau$ was the arbitrary delay for the cross-correlation function, and $E_L$ and $E_R$ were the RMS energy in the left and right signals, respectively. IS was then defined using the same notation as above as

$IS\left(n,\theta \right)=\underset{\tau }{\max }\rho \left(n,\theta ,\tau \right).$

If the $E_L$ or $E_R$ was zero because there were no pulses in that ear, then $\rho$ was set to zero to avoid infinite values. Once IS was calculated for each channel, these values were averaged across n channels and across all five repetitions. This metric was chosen to summarize how well pulses overlapped, but may not be relevant to neural coding of the envelopes carried by those pulses.

Figure 2 shows the IS for each condition. Most notably, IS was improved at all stimulus locations with the use of synchronized ACE over unsynchronized, and near perfect (∼1) IC was observed when using Bilateral ACE. Analysis of the IS as a function of channel revealed that differences in IC across channels were largest at the channels representing the highest frequencies (the lowest electrodes numbers).

Figure 2.

Simulated interaural synchrony (IS) calculated for ACE and BACE strategies implemented on the CCi-MOBILE, as compared to a simulation of clinical processing. A: IS by loudspeaker location, B: IS by electrode channel. Gray bars represent 95% confidence intervals. Stimuli were recorded using the CCi-MOBILE on a KEMAR (Knowles Electronics Manikin for Acoustic Research, G.R.A.S Sound & Vibration, Holte, Denmark) mannequin in the same booth used for experimental testing. The CCi-MOBILE microphone inputs were written to WAV files and processed with both ACE and Bilateral ACE. Unsynchronized clinical processing was simulated by applying jitter to the pulses in each channel for the ACE condition. Each data curve is averaged over twenty-two electrodes and five repetitions; for the plot of IS as a function of channel, data is averaged over five repetitions and nine locations. MAPs were programmed with 22 active electrodes, standard frequency allocation tables, and threshold and comfortable levels of 100 and 150 CUs, respectively

ILDs were also simulated following the procedure established in Borjigin et al. (2025). To estimate ILDs, for each biphasic pulse per channel in a given electrodogram, the current amplitude was transformed into a normalized percentage of the electrical dynamic range (0 to 100%). ILDs were then calculated as the log 10 of RMS power in the right ear over the left ear. Figure 3 summarizes the initial ILD at stimulus onset and the change in ILD over the duration of the stimulus (one second) via linear interpolation of the raw ILDs over time. ILDs were consistent across conditions for initial ILD at onset and when the stimulus was not moving. When the stimulus moved, larger changes in ILDs were estimated with the Clinical condition than the CCi-MOBILE conditions.

Figure 3.

Simulated interaural level differences (ILDs) calculated for ACE and BACE strategies implemented on the CCi-MOBILE, as compared to a simulation of clinical processing. Each column represents a different stimulus starting position along a loudspeaker array, from -80 to 80°. Each row represents a different range of auditory motion, from -40 to 40°. Stimuli were recorded using the CCi-MOBILE on a KEMAR (Knowles Electronics Manikin for Acoustic Research, G.R.A.S Sound & Vibration, Holte, Denmark) mannequin in the same booth used for experimental testing. The CCi-MOBILE microphone inputs were written to WAV files and processed with both ACE and BACE. Unsynchronized clinical processing was simulated by applying jitter to the pulses in each channel for the ACE condition. Each data curve is averaged over twenty-two electrodes and five repetitions. MAPs were programmed with 22 active electrodes, standard frequency allocation tables, and threshold and comfortable levels of 100 and 150 CUs, respectively

Participants

Participants (6 female, 3 male, 20-82 years old) who were bilaterally implanted with Cochlear Ltd. (Sydney, Australia) implants traveled to the University of Wisconsin-Madison for testing, and received travel reimbursement and payment for their time. Experimental procedures followed the National Institutes of Health regulations on research with human participants and the best practice guidelines as outlined in Litovsky et al. (2017), and were approved by the University of Wisconsin-Madison’s Health Sciences Institutional Review Board. Table 1 below includes biographical information for each participant.

Table 1.

Participant Information

ID	Age	Years bilateral	Gender	Pulse rate (pps)	MAPs adjusted
IAJ	73	16	F	1200	No
IAU	72	15	M	900	Yes
IBK	82	13	M	900	No
IBO	54	9	F	1200	No
ICI	61	10	F	900	No
ICM	63	4	F	900	No
IDA	52	5	F	900	No
IDH	20	14	M	1200	No
IDL	67	5	F	900	Yes

Stimuli and Apparatus

Participants sat in a sound booth with internal dimensions of 2.90 x 2.74 x 2.44 m (IAC, RS 254S) and additional sound absorbing foam attached to the inside walls to reduce reflections. A Tucker-Davis Technologies (Alachua, FL) System 3 was used to select and drive an array of 37 loudspeakers (Cambridge SoundWorks, North Andover, MA) arranged on a semi-circular arc of 1.2 m radius. Loudspeakers were positioned in the frontal hemisphere in 5° increments along the horizontal-plane and were hidden behind an acoustically transparent curtain. There were two screens in the sound booth, one mounted at eye level above the arc with instructions for the participant, and one with a touchscreen on a small table in front of the participant. This second touchscreen showed a large “Play Sound” button. An OptiTrack infrared motion tracking system (NaturalPoint Inc., Corvallis, OR, USA) was also installed in the room. The OptiTrack system was used to track the motion of a presenter remote (Amazon, Seattle, WA) that had an OptiTrack-compatible sensor attached on top.

Participants were instructed to press the “Play Sound” button on the touchscreen to play a sound, then use the laser on the presenter remote to indicate the trajectory of where they heard the sound moving. While indicating the trajectory, participants were instructed to press down on a button on the presenter remote to indicate the beginning of the trajectory and release the button to indicate the end of the trajectory. Custom software was written to continuously log the OptiTrack data containing the location of the presenter remote and button presses during a session. The logged data contained the coordinates of the presenter remote in the room, including its location in 3-dimensional space and its rotation, at a rate of 30 frames per second, and the button presses from the presenter remote marked the beginning and end of a trial response.

Stimuli were generated in MATLAB (Mathworks, Natick, MA, USA) and delivered from the loudspeaker array using vector-based amplitude panning (Pulkki, 1997) to simulate a moving sound. Stimuli were white-noise tokens that were generated for each trial. Signal presentation level was set to 60 dB SPL as calibrated by a digital precision sound level meter (System 824, Larson Davis; Depew, NY). This level was below the 65 dB SPL threshold of the clinical processor automatic gain control (AGC) (Vaerenberg et al., 2014) and so there was likely no more than a small effect on comparison across processors. Stimuli moved at three angular ranges: 40°, 20°, or 0° (stationary). Stationary stimuli were presented at loudspeakers -80 (left), -60, -40, -20, 0 (front), 20, 40, 60, and 80 (right) degrees. Moving stimuli began at these locations and moved either to the left or right if possible. If not possible (e.g. moving left from -80 degrees would exceed the range of available loudspeakers), those moving stimuli were not included. In order to ensure the same number of stationary and moving trials overall, there were 150 stationary trials, 80 (40 left/40 right) 20° trials and 70 (35 left/35 right) 40° trials.

Experimental Design

The experiment was conducted using a pseudo-random block design following a Latin Square. Participants were tested in three conditions: listening with (1) ACE and (2) BACE on the CCi-MOBILE, and (3) commercially-available clinical processors. In total, there were fifteen blocks of trials. Each block took about 20 minutes to complete and participants typically took six hours to complete the experiment when accounting for initial fitting and frequent breaks. Testing blocks were arranged in groups of three, meaning that after every three blocks, participants had completed one block of each condition. Each testing block contained 30 stationary trials, 16 trials moving by 20°, and 14 trials moving by 40°, totaling 60 trials in a single block. Due to experimenter error, one participant (IAU) completed an uneven number of stationary and moving trials. Therefore, their data was not considered for any analyses that assumed an equal number of stationary and moving presentations. All stimuli were presented at random within a single block. Participants were not told which CCi-MOBILE strategy was being used during testing and did not know the difference between algorithms. However, they knew when they were using clinical processors vs the CCi-MOBILE because they had to physically change between them. The order of strategies used was counterbalanced across participants to account for learning effects at the group level.

For the clinical processor condition, all participants were fitted with Cochlear N6 processors loaned to the lab from the manufacturer and were programmed with their clinical MAP threshold and comfort levels. Participants needed to have identical frequency allocation tables (FATs) across ears to use the BACE strategy. If participants did not have identical frequency allocation tables, FATs and clinical MAPs were adjusted so that the same electrodes were active in each ear as long as it was safe to do so and in line with best practice (Litovsky et al., 2017). Only two participants required adjustments to their FATs. For the CCi-MOBILE conditions, the same number of channels and FATs were used across ears. The clinical processors used Cochlear’s implementation of the ACE strategy with front end proprietary sound processing such as SCAN, ASC, noise reduction, and ADRO deactivated. Although the sound processing is proprietary, it can be assumed to be relatively similar to the ACE strategy programmed for the CCi-MOBILE because the CCi-MOBILE was based on the implementation provided in the Nucleus MATLAB Toolbox, a derivative of which is now publicly available as the Nucleus Toolbox for MATLAB (Brett Swanson, 2025). AGC could not be deactivated for the clinical processors.

Data Analysis

Because participants were not asked to directly identify a sound as moving or stationary, a Gaussian mixture model was used to classify each participants’ response. A separate Gaussian mixture model was used to classify each individuals’ responses as “stationary,” “moving right,” or “moving left.” These labels can be used to calculate the discrimination ability of each participant with each device. The features used to classify each trajectory were 1) the range (the signed difference between start and end angles), 2) the natural logarithm of the duration of the response, and 3) the velocity (range divided by duration, with no log transform). Data were fit using the MATLAB function fitgmdist with three clusters (stationary, moving left, and moving right), and then grouped with the cluster function.

The proportion of trials identified as moving was calculated from the classified responses. This allowed us to compute two measures to understand the discrimination abilities of the participants. Motion sensitivity (d’) and motion bias criterion (c) were calculated for each participant to understand their ability to discriminate the perceived motion of a sound source and the response bias (Macmillan & Creelman, 2004). Motion sensitivity captures how well a participant can discriminate between a moving sound and a stationary sound, and was calculated as:

$d^{\prime }=Z\left(Hit\hspace{0.17em}Rate\right)-Z\left(False\hspace{0.17em}Alarm\hspace{0.17em}Rate\right),$

where Z() represents the normal transformation, Hit Rate is the proportion of stationary sounds that were accurately identified as stationary, and False Alarm Rate is the proportion of moving sounds that were incorrectly identified as stationary. Sensitivity scores greater than zero represent better than chance discrimination. The motion bias criterion captures whether participants were more likely to indicate that sounds were moving or stationary, and was calculated as:

$c=\frac{1}{2}Z\left(Hit\hspace{0.17em}Rate\right)+\frac{1}{2}Z\left(False\hspace{0.17em}Alarm\hspace{0.17em}Rate\right),$

with positive values representing a bias towards reporting sounds as stationary and negative values representing a bias towards reporting sounds as moving.

All statistical analysis was completed with R version 4.5. Linear mixed effects models were fit using the function lmer to determine if a strategy led to a difference in either $d^{\prime }$ or bias using the models $d^{\prime }\sim 1+Condition+\left(1|Participant\right)$ and $c\sim 1+Condition+\left(1|Participant\right)$ .

To analyze accuracy of responses, two response variables were considered: 1) the root mean square (RMS) error between the start location of a response and the start location of the stimulus and 2) the RMS error between the angular range of a response and the true range of motion. This second metric is different to that considered in Moua et al. (2019) but we believe provides a more holistic view of auditory motion perception than prior metrics. In that study, the authors used multiple metrics for the ability to characterize auditory motion perception. Start and End point RMS values were calculated and analyzed along with the extent of motion. However, these metrics do not capture the absolute range difference (error between true range of motion and reported range of motion). The RMS range difference is analogous to the RMS error as typically reported for sound localization studies, and effectively encompasses both Start and End RMS. The Shapiro-Wilk Test and Levene’s Test were used to verify assumptions of normality and homogeneity of variance, respectively. A linear mixed effect model was fit for each metric using to determine if there was a difference in performance due to strategy. The model equations used were $RM{S}_{Start}\sim 1+Condition+\left(1|Participant\right)$ and $RM{S}_{Range}\sim 1+Condition+\left(1|Participant\right)$ .

Results

Discrimination of Moving Sounds

Figure 3 shows the sensitivity for accurately detecting sound movement and bias for eight participants. It can be seen that for all listening conditions, participants showed a similar amount of sensitivity and bias. Only 3 listeners (IBO, IDA, and IDH) were good at discriminating direction of motion (d’ close or greater than 1). The motion sensitivity scores passed tests for both normality and homogeneity of variance (Shapiro-Wilk: $W=0.96,p=0.47$ ; Levene’s: $F\left(2,21\right)=0.69,p=0.38$ ). The Type II Wald chi-square test revealed no difference due to strategy ( ${\chi }^2\left(2\right)=4.31,p=0.19$ ). The intercept of the model was found to be significantly different from zero ( $t\left(9.25\right)=4.3,p=0.001$ ), with a mean group sensitivity of 0.63. As d’ = 0 is chance performance, this means that as a group, these listeners were above chance at judging whether sounds were moving or not. The full output of the model is available in the Supplemental Material, Supplementary Table 1.

The motion bias criterion scores passed tests for normality and homogeneity of variance (Shapiro-Wilk: $W=0.99,p=0.98$ ; Levene’s: $F\left(2,21\right)=0.13,p=0.88$ ). The Type II Wald chi-square test revealed no difference due to strategy ( ${\chi }^2\left(2\right)=4.62,p=0.09$ ). The intercept of the model was not significantly different from zero ( $t\left(9.9\right)=0.722,p=0.49$ ), indicating that on the group level, participants were not biased to report sounds as either moving or stationary. Taken together with the above results, this means that listeners were able to accurately discriminate whether sounds moved or not but that their performance was not due to a bias in reporting whether sounds were moving or stationary. The full output of the model is available in the Supplemental Material, Supplementary Table 2.

Perception of Range of Motion

Figure 4 shows the RMS error between true start locations and response start locations for all nine participants. Estimated marginal means for RMS errors for start location were 27° for ACE, 29° for Bilateral ACE, and 29° for the Clinical condition. For RMS error for start location, the data was found to violate assumptions of normality (Shapiro-Wilk: $W=0.92,p=0.04$ ) but not homogeneity of variance (Levene’s test: $F\left(2,24\right)=0.135,p=0.87$ ). To address this, the response variable was transformed with a base-10 logarithm, and both tests were passed (Shapiro-Wilk: $W=0.96,p=0.37$ ; Levene’s: $F\left(2,24\right)=0.09,p=0.91$ ). The Type II Wald chi-square test did not reveal a significant effect due to strategy ( ${\chi }^2\left(2\right)=1.41,p=0.494)$ . This indicates that there was no difference in strategy for determining where a sound began. The full output of the model is available in the Supplemental Material, Supplementary Table 3.

Figure 4.

Individual data for sensitivity to auditory motion. A: Discrimination of moving vs. stationary sounds ( $d^{\prime }$ ). B: Bias towards stationary vs. moving sounds ( $c$ )

Figure 5 also shows absolute difference between true range of motion and range of response. The residuals passed tests for both normality and homogeneity of variance (Shapiro-Wilk: $W=0.96,p=0.39$ ; Levene’s: $F\left(2,23\right)=0.62,p=0.55$ ). The Type II Wald chi-square test revealed a significant effect of strategy ( ${\chi }^2\left(2\right)=10.185,p=0.006$ ). The intercept of the model was found to be significantly different from zero ( $t\left(9.95\right)=9.07,p=0.001$ ), with a mean absolute error of 19.4. Estimated marginal means were 22.6° for Clinical processors, 19.4° for ACE, and 19.2° for BACE. Multiple comparisons with Bonferroni method for 3 tests at a confidence level of 0.95 revealed a significant contrast between Clinical and BACE ( $-3.46°,t\left(16\right)=-2.861,p=0.029$ ) and Clinical and ACE ( $-3.22,t\left(16\right)=-2.66,p=0.043$ ). ACE and BACE were not significantly different (Contrast: $0.25°,t\left(16\right)=0.21,p=0.98$ ). This suggests that bilateral synchronization can improve the accuracy of perceived range of motion but bilateral peak-picking may not be necessary. The full output of the model is available in the Supplemental Material, Supplementary Table 4.

Figure 5.

Individual data for error in start response location and range of response. A: RMS error between true start locations and response start locations, B: RMS error between range of motion and response range of motion

Discussion

In this work, the capabilities of bilateral CI users to discriminate auditory motion were measured with clinical processors and a bilaterally synchronized research processor, the CCi-MOBILE. Two conditions were tested with the CCi-MOBILE, ACE and BACE. It was hypothesized that the lack of synchronization and lack of coordination of interaural channel selection in clinical processors lead to the poor auditory motion discrimination ability observed in Moua et al. (2019). Therefore, it was hypothesized that if synchronized hardware was used, auditory motion discrimination and accuracy would be improved by using synchronized processors, and be further improved by bilateral signal processing that coordinates peak-picking across the ears (BACE). While the use of synchronized processors did not lead to significant improvements in discrimination of motion or accuracy in the start position of sounds, an improvement was seen in the perceived range of the moving sound when listening with bilateral synchronization. There was no effect of bilateral peak-picking.

When compared to the investigation of the auditory motion discrimination abilities of BiCI listeners by Moua et al. (2019), participants had improved abilities. On average, participants in Moua et al. (2019) had motion sensitivity $d^{\prime }$ scores near zero, while in this study, participants had $d^{\prime }$ scores closer to 0.7 (averaged across all conditions), suggesting a better than chance ability to discriminate whether or not a sound was moving. Neither group of participants in the two studies approached the d' of 1.7 reported for TH participants (Moua et al., 2019). When considering response bias, Moua et al. (2019) reported that BiCI listeners were more biased to report sounds as moving. This finding was not reproduced here, as bias criterion was not significantly different from zero, meaning that participants did not have a bias towards reporting sounds as moving. There was also no statistically significant difference in discrimination across strategies, suggesting that some factor other than synchronized processors led to this difference from the prior experiment. It is possible that the small improvement of d’ scores and the change of bias in this study as compared to the results of Moua et al. (2019) may have been due to the use of head motion cues. Participants in Moua et al. (2019) listened to virtual acoustic stimuli, while participants in this study had live microphones on their processors and were capable of utilizing head motion cues (Brimijoin & Akeroyd, 2016; Dwyer et al., 2021).

As with motion sensitivity, start location RMS errors were similar across all conditions. Mean RMS errors reported by Moua et al. (2019), around 30° on average, were similar to the present study, which were just below 30°. The small difference is perhaps due to the different method of data collection or the use of virtual stimuli in that study, which limited access to head movements, as discussed above. Moua et al. (2019) had participants respond by drawing the sound trajectory on a touchscreen which may lead to coordinate translation errors (Brungart et al., 2000). In this study, participants used a laser on a presenter remote to draw the trajectory of the sound, which was designed as a more natural way to respond, as it does not require a mental translation of coordinates. The start location RMS error was consistent with errors for static localization and is the same magnitude as previously reported in other studies using different response collection methods (Dorman et al., 2014; Grantham et al., 2007; Smith et al., 2014). As expected based on these previous studies, localization RMS error for BiCI participants was much greater than the errors reported for TH participants, which can be as low as 6.7° (Dorman et al., 2016; Grantham et al., 2007). Dennison et al. (2022) reported no benefit due to using synchronized processors as compared to unsynchronized processors for static localization. The authors conjectured that despite the use of synchronized processors, uncoordinated peak-picking disrupted the benefits of synchronization. The clear improvement in interaural synchrony demonstrated in Figure 2, however, did not lead to better precision when identifying the start location of a sound. This may have been because the initial ILD of the stimuli used here did not change across conditions, as shown in Figure 3.

The only measure that demonstrated a significant effect of strategy was the response range RMS error. Errors in estimating the range of motion were similar with Bilateral ACE and CCi-MOBILE ACE, but were significantly greater with the use of unsynchronized clinical processors. As shown in Figure 3, ILDs changed the most across time in the Clinical condition, which may have driven differences in range perception. Though participants did not demonstrate an improvement in discrimination of motion or accuracy of start location with BACE, the results suggest that participants may experience a benefit in range of motion with synchronized hardware. The current finding adds to that of Dennison et al. (2022), which found that CI users did not seem to derive additional spatial hearing benefits in stationary tasks from synchronized hardware alone. fMRI studies have shown that auditory motion direction is encoded both in the auditory cortex and high-level visual cortex in humans (Alink et al., 2012) and that neural processing of auditory motion may be distinct from processing of stationary sounds (Battal et al., 2019). Hence, it may seem that the benefit of synchronized hardware can only be observed in a dynamic hearing task.

There were some limitations to the study that impact interpretation of the results. Participants were only exposed to the Bilateral ACE strategy for a few hours, and this time period may not have been long enough for participants to acclimate to any differences across conditions. New strategies are sometimes evaluated after months of take-home listening (Carlyon & Goehring, 2021), but this option was impractical for our research device and was outside of the scope of our testing. AGC was not implemented for the CCi-MOBILE conditions, but was active for the clinical processors. Differences in AGC may have led to the small difference in range of motion error, as AGC can impact perception of auditory motion when above the compression threshold (Pastore et al., 2021). By not having AGC in either ear for the CCi-MOBILE, performance may have been more similar to performance with linked AGC, with some authors reporting an improvement in linked over unlinked AGC using Advanced Bionics (Valencia, CA, USA) processors (Dwyer et al., 2021; Pastore et al., 2021) and other authors reporting no difference in stationary sound localization with MED-EL (Innsbruck, Austria) processors (Schleich et al., 2025).

Supplemental Material

Supplemental Material - Auditory Motion Perception by Bilateral Cochlear Implant Users With a Sound Coding Strategy that Synchronizes Peak-Picking Across the Ears

Supplemental Material for Auditory Motion Perception by Bilateral Cochlear Implant Users With a Sound Coding Strategy that Synchronizes Peak-Picking Across the Ears by Stephen R. Dennison, Lingkai Harry Zhao, Alan Kan and Ruth Y. Litovsky in Trends in Hearing.