Experience with the cochlear implant enhances the neural tracking of spectrotemporal patterns in the Alberti bass

Cochlear implant (CI) users experience diminished music enjoyment due to the technical limitations of the CI. Nonetheless, behavioral studies have reported that rhythmic features are well-transmitted through the CI. Still, the gradual improvement of rhythm perception after the CI switch-on has not yet been determined using neurophysiological measures. To fill this gap, we here reanalyzed the electroencephalographic responses of participants from two previous mismatch negativity studies. These studies included eight recently implanted CI users measured twice, within the first six weeks after CI switch-on and approximately three months later; thirteen experienced CI users with a median experience of 7 years; and fourteen normally hearing (NH) controls. All participants listened to a repetitive four-tone pattern (known in music as Alberti bass) for 35 min. Applying frequency tagging, we aimed to estimate the neural activity synchronized to the periodicities of the Alberti bass. We hypothesized that longer experience with the CI would be reflected in stronger frequency-tagged neural responses approaching the responses of NH controls. We found an increase in the frequency-tagged amplitudes after only 3 months of CI use. This increase in neural synchronization may reflect an early adaptation to the CI stimulation. Moreover, the frequency-tagged amplitudes of experienced CI users were significantly greater than those of recently implanted CI users, but still smaller than those of NH controls. The frequency-tagged neural responses did not just reflect spectrotemporal changes in the stimuli (i.e., intensity or spectral content fluctuating over time), but also showed non-linear transformations that seemed to enhance relevant periodicities of the Alberti bass. Our findings provide neurophysiological evidence indicating a gradual adaptation to the CI, which is noticeable already after three months, resulting in close to NH brain processing of spectrotemporal features of musical rhythms after extended CI use.


Introduction
The cochlear implant (CI) has significantly improved the quality of life for thousands of individuals.This neural prosthesis successfully allows recipients to gain or regain hearing and speech perception abilities.By contrast, listening to the intricate and complex sound of music is challenging.The main reason for this limitation is that the CI is less efficient in transmitting temporal fine structure, has a low spectral resolution and provides a constrained dynamic range (Drennan and Rubinstein, 2008), which leads to poor perception of fundamental musical features, such as pitch, timbre and intensity (Hopyan et al., 2012;McDermott, 2004;Prevoteau et al., 2018;Zeng, 2004).This is reflected in lower levels of music pleasure (Dritsakis et al., 2017;Gfeller et al., 2000;Lassaletta et al., 2008;Looi and She, 2010;Moran et al., 2016) and may exclude CI users from the social benefits and pleasure of listening to music, singing or playing.
On the flip side, behavioral studies report that temporal processing is well preserved with a CI (Cooper et al., 2008;Jiam and Limb, 2019;Limb and Roy, 2014;Limb and Rubinstein, 2012).In fact, rhythm plays such an important role that if CI users are presented with simple melodies in which the temporal cues are removed, they become unable to identify the tunes (Kong et al., 2004).A common limitation of studies investigating rhythm discrimination in CI users is, however, that the performance often reaches ceiling levels due to the simplicity of the rhythmic tasks, as in the case of pattern identification or tempo differentiation (Gfeller et al., 1997;Kong et al., 2004;Limb, 2006;Limb et al., 2010).Some paradigms tried to overcome this limitation by focusing on a more challenging aspect of rhythmic processing: the internal representation of rhythmic periodicities (rhythmic clocking or beat perception; Kim et al., 2010), which requires a more accurate integration of precise temporal cues.Following the same logic, this study aims to investigate the neural correlates of these rhythmic periodicities in CI users and provide an objective measure of their ability to encode the four-tone pattern of the Alberti bass at different levels of CI experience.

Beat perception in CI users
Beat perception consists of detecting and predicting periodicities of events over time (Honing, 2012;Vuust et al., 2022).In normal hearing (NH) individuals, beat perception allows for establishing an endogenous framework of timing and time intervals in which a specific event may occur (London, 2012).These time intervals are the periodicities within rhythmic patterns that direct our body movements during synchronized activities, such as sports, dance or playing music.Musical rhythms are built on these periodicities, with patterns of subdivisions and groupings of these regular events at faster and slower rates (i.e.metrical levels), like in the ternary beat of a waltz or the binary beat of a march.The generation of metrical levels and the perception of beat does not uniquely depend on temporal features (i.e., relative timings; see Ravignani et al., 2018), but also relies on melodic information (e.g., repeated patterns or contour changes; see Hannon et al., 2004) and acoustic features (e.g., intensity accents; Toiviainen and Eerola, 2006).These periodicities that neatly align with the delta frequency band of brain oscillations (~1-3 Hz, Poeppel, 2014) play a key role in language processing: to parse the temporal patterns of the syllabic rate of speech (Kotz and Schwartze, 2010;Meyer, 2018;Gross et al., 2013).The evidence highlighting anomalous beat perception in children with speech and language impairments (Colling et al., 2017;Corriveau and Goswami, 2009) emphasizes the importance of addressing this issue in CI users (Torppa and Huotilainen, 2019), particularly when they are young and have several years of developmental milestones ahead.
It is noteworthy that CI users exhibit beat perception that closely resemble that of NH people.Using a behavioral paradigm, Kim et al., (2010) found that CI users and NH participants did not differ in detecting small perturbations (between ±20 and ±188 milliseconds) in the last sound of a four-beat metronome presented at three distinct tempi (60,120 and 180 BPM).The detection of these small perturbations relied on internal beat synchronization, and CI users even outperformed NH non-musicians in some cases.Using an electroencephalographic (EEG) paradigm, Alemi et al., (2021) showed that a complex tone presented at 2.4 Hz (i.e.144 BPM) elicited measurable cortical steady-state evoked potentials at this frequency rate (i.e. the beat) and its harmonics at frontocentral locations of the scalp (after removing CI-induced artifacts).They found that the cortical neural responses of CI users were smaller compared to those of NH controls.These two studies show that CI users maintain the ability to synchronize internally to metronomic rhythms both behaviorally and neurally.The latter, however, puts frequency tagging forward as a reliable method to investigate the development of beat perception in regular sequences, particularly within music perception.

Adaptation period to the CI
The central nervous system of CI users undergoes an adaptation period to the signal of the intracochlear electrodes once the CI is switched on, and this can take from a few weeks to several months (Glennon et al., 2020).The standard practice is to activate and fit the CI processor approximately one month after the surgery, and here is the point where each user experiences a distinct adaptation process influenced by perceptual and neurophysiological changes.Some studies support that the most dramatic improvements occur within the initial months of CI device activation (Chang et al., 2010;Petersen et al., 2013;Sandmann et al., 2015).Positron emission tomography (Petersen et al., 2013) and EEG (Sandmann et al., 2015;Seeberg et al., 2023) studies revealed that the neural responses to auditory stimuli strengthen with prolonged CI experience.Using a musical multifeature paradigm (MuMuFe), Seeberg et al. (2023) found that mismatch negativity (MMN) responses to deviations in pitch and timbre became stronger after approximately three months of CI experience.They also found that these MMN responses were stronger in experienced CI users (between 1 and 14 years of CI experience) than in recently implanted CI users.In this line, a longitudinal study showed that CI users' performances continue to improve up to 48 months post-implantation (Chang et al., 2010; using a word recognition task), which indicates that the learning phase or adaptation period to the CI can be quite extensive.

Aims and hypotheses
In this study, we aimed to quantify the early and extended adaptation periods for rhythm after cochlear implantation.First, we compared the frequency-tagged EEG responses of recently implanted CI users (CIre) recorded shortly after switch-on of the device (T1) and approximately 3 months later (T2).Second, we compared the responses of experienced CI users (CIex) with the T2 responses of CIre.Third, we compared the T2 responses of CIre and the responses of CIex with those of NH controls.We hypothesized that prolonged CI experience with the CI would be reflected in increased neural synchronization to musical rhythms, indicated by higher frequency-tagged amplitudes related to the periodicities of our musical stimuli.Furthermore, we expected that, over time, the recorded neural responses of CI users would be similar to those of NH people, despite the challenging removal of CI-related artifacts.Finally, we hypothesized that the recorded neural activity not only reflected exogenous responses to intensity or spectral changes between consecutive tones, but also endogenous neural responses to enhance relevant frequencies of the four-tone pattern.We use the term beat-related across the article to refer to the periodicities (i.e., harmonics and subharmonics) related to the tone onset rate of the Alberti bass.
The data were derived from a longitudinal EEG study that investigated the adaptation period following cochlear implantation using a MuMuFe MMN paradigm based on an Alberti bass pattern (Petersen et al., 2020;Seeberg et al., 2023).

Participants
EEG data from the following three groups of participants were included in the analyses: eight recently implanted users (CIre, median age: 63 years, range: 30-85; 2 women), thirteen experienced CI users (CIex; median age = 56 years, range = 18-77 years; 9 women) and fourteen normal hearing adults (NH; median age = 62 years, range = 56-77 years; 7 women).CIre participants were tested twice; within the first six weeks after CI switch-on (T1; median = 20.5 days, time range = 3-42 days), and approximately 3 months later (T2; median = 117 days, time range = 105-188 days).CIex participants were tested once (median CI experience = 7 years, time range = 1-14 years).All participants were non-musicians (<5 years of singing or instrument training), except for one recently implanted CI user who reported 6 years of training in music.For demographic and clinical characteristics of the CIre and CIex participants, see Table S1 in the Appendix.
Inclusion criteria: participants were aged 18 years or older, had postlingual hearing loss (i.e., acquired profound hearing loss on the implanted ear after language acquisition), and reported absence of neurological and severe psychological disorders, and no use of medication affecting brain function.NH participants had to pass an online hearing test for perception of words and numbers in background noise (www.beltonehearingtest.com).
The EEG studies were approved by the Research Ethics Committee of the Central Denmark Region (#55,018).All participants provided informed consent, in accordance with the Declaration of Helsinki.

Stimuli
The CI version of the MuMuFe paradigm presented arpeggiated chord triads following a repetitive four-tone (low-high-medium-high) pattern known as the Alberti bass (see Fig. 1).The stimuli pseudorandomly changed in key (i.e., C, Eb, Gb, or A) every 48 occurrences of the four-tone pattern.To investigate MMN responses, the medium (third) tone always presented a change in terms of intensity, pitch, timbre or rhythm at four possible magnitude levels: small, medium, large and extra-large change (for more details about the differences between deviants, see: Petersen et al., 2020;Seeberg et al., 2023).The auditory sounds were created with the virtual piano Alicia's Keys (Native Instruments) at a frequency sample of 44,100 Hz, using tones between 208 and 659 Hz.The tones had a duration of 200 ms and were presented at an inter-onset interval of 205 ms, i.e., at a frequency of 4.878 Hz (292 BPM).The grouping of two tones thus occurs at a frequency of 2.439 Hz (146 BPM) and the grouping of four tones at 1.219 Hz (73 BPM).These three periodicities constitute the frequencies related to the beat and its metrical levels, which organize the Alberti Bass following binary structures (see Fig. 1).

Procedure
Participants were sitting on a comfortable chair in an electrically and acoustically shielded room in the EEG lab of the Danish Neuroscience Center (Aarhus, Denmark).They watched a muted movie with subtitles while the stimuli of the MuMuFe paradigm were presented for 35 min.They were instructed to ignore the auditory stimuli and focus on the movie.The volume of the sound was adjusted individually to a comfortable level (starting at 65 dB SPL).While NH participants received sounds bilaterally through in-ear Shure headphones, CI users received the sounds unilaterally directly to their implant via audio cables and with microphones switched off.Bilateral CI users chose their preferred implant, and bimodal CI users removed the contralateral hearing aid.The everyday settings of the CI users' speech processor were used during all the recordings.EEG was recorded with a 32-electrode acticap and a BrainAmp amplifier system (BrainProducts, Germany).The active electrodes were placed following the international 10/20 system, using the FCz as reference.The parietal electrodes that could interfere with the CI transmitter coil were disregarded.For more details, see Petersen et al. (2020) and Seeberg et al. (2023).

EEG preprocessing
The EEG recordings were preprocessed as described in Petersen et al. (2020).The FieldTrip Toolbox for Matlab (Oostenveld et al., 2011) was applied to downsample the EEG data to 250 Hz, high pass filter at 1 Hz and low pass filter at 25 Hz.The standard method for weighted interpolation of neighboring channels implemented in FieldTrip was applied for replacing between 0 and 4 bad channels showing high amplitude noise or a flat line.The infomax independent component analysis (Delorme et al., 2007;Makeig et al., 1996) technique was then applied to correct for eye and CI artifacts, and between 1 and 10 artifactual independent components were identified by visual inspection and subtracted.On average, five CI-related components were removed in each CI group.

EEG analyses
To apply the frequency tagging method, the preprocessed EEG data was segmented into 48 trials based on the triggers that signaled changes in tonality in the Alberti Bass.These long trials contained 192 sounds and lasted 39.36 s.Twelve of these trials could not be obtained in one NH participant due to missing triggers, but we still included the participant in the analyses.To discard the evoked potentials related to changes in tonality, we removed the first 820 ms of each trial, i.e., the first four-tone pattern of the Alberti Bass.Subsequently, for each participant and electrode, we averaged the EEG trials to diminish neural activity non-phase-locked to the stimuli.To the resulting averaged epoch, we applied a fast Fourier transform, zero-padded to the next power of two: 2^14.The Fourier transform converts the amplitudes over time (μV/s) into amplitudes over frequencies (μV/Hz), in this case with a frequency resolution of 0.0153 Hz.According to the frequency tagging approach, these amplitudes represent the neural activity of processing the stimuli together with spontaneous stimulus-unrelated activity and background noise (Alemi et al., 2021;Nozaradan, 2014).
To increase the signal-to-noise ratio, we subtracted the mean value of non-adjacent surrounding frequency bins at each frequency bin, i.e., the mean of the amplitudes falling between -76.5 and -45.9 mHz and 45.9 and 76.5 mHz (milliHertz).This noise subtraction procedure stems from the idea that if no steady-state evoked potentials are present, the voltage amplitudes may vary similarly across frequencies, and they will tend towards zero after the subtraction.In contrast, if some periodicities are consistently present in the signal, the amplitudes will be significantly greater than zero, and some peaks will appear in the frequency spectrum (see Fig. 1).To statistically confirm the presence of periodicities related to the Alberti bass, we selected the amplitudes at beat-related frequencies: the sound inter-onset interval (f = 4.878 Hz), two subharmonics of this periodicity (f/4 = 1.219Hz and f/2 = 2.439 Hz) and its first harmonic (2f = 9.756 Hz) and tested whether these amplitudes were greater than zero.The one-sample t-tests against zero confirmed the presence of these peaks across groups (see Table S2 in Supplementary Materials).Accordingly, these four peaks showed the highest amplitudes on the frequency spectra (see Fig. 1,bottom).Additional analyses including more frequencies of interest (FoI; i.e., the 5 first harmonics of the downbeat of the Alberti bass and the 5 first harmonics of the tone onset rate) are reported in Supplementary Materials.
For each participant, we averaged the amplitudes of the four beatrelated frequencies to account for neural synchronization to the rhythm of the Alberti bass.The average or summation of the harmonics of a frequency of interest has recently been reviewed and recommended for frequency tagging (Retter et al., 2021).To prevent potential electrical interference from the CIs and following established literature that identify frontocentral scalp regions as key areas for localizing steady-state evoked potentials associated with tracking rhythmic periodicities (e.g.Alemi et al., 2021;Sauvé et al., 2022), we averaged the activity of six frontocentral electrodes: Fz, F3, F4, FC1, FC2, Cz (see Fig. 2).The topographies in Fig. 2a confirm that this group of electrodes align with the synchronized activity generally localized in frontocentral regions.Thus, the averaged activity at beat-related frequencies of this cluster of electrodes was used in all the subsequent statistical analyses using JASP, version 0.16.00 (JASP Team, 2024).
An additional analysis of the EEG responses to the four tone positions of the Alberti bass is reported in Supplementary Materials.They explore the P50 and N100 components of the event-related potentials of NH participants, to test differences related to the size of pitch contrasts that could have boosted the metrical grouping of the sounds.

Stimuli analyses: intensity and spectral content features
Our natural musical stimuli were created by presenting piano tones sampled from Alicia's Keys.Apart from the intended pitch differences (i.e., spectral content), the consecutive tone contrasts also presented slight differences in terms of intensity (see Fig. 3a).To disentangle their contribution in the processing of metrically organized tones, we extracted these two audio features with the Music Information Retrieval Toolbox (MIR toolbox, version 1.8.1) for Matlab (Lartillot and Toiviainen, 2007).This allowed us to compare the periodicities found in the EEG frequency spectra with the periodicities found in the stimuli, and check if the neural spectra just reflected the tracking of intensity and spectral changes (i.e., exogenous bottom-up responses).
The audio file of the full Alberti bass was decomposed with short-time Fourier transform using a Blackman-Harris window (window size = 100 ms, step size = 10 ms).To increase perceptual validity, the 'Terhardt' option in the mirspectrum function was chosen to suppress the low and high frequency ends of the spectra, related to reduced responsiveness in human hearing (Pampalk, 2004;Terhardt, 1979).First, the sound intensity over time was estimated as the summed amplitude across the short-time Fourier transform frequency bins, which partially relates to the perceived loudness.Second, the spectral flux was measured as the distance between spectra in adjacent time frames, showing peaks at time points with a change in timbre or pitch.To avoid redundancy with the intensity estimates and exclude the influence of sound intensity changes, the 'NormalSpectrum' option was set up as 'Local' in the mirflux function.Finally, to make the intensity and spectral flux estimates comparable to our EEG signal, we converted the sampling rate to 250 Hz and applied a 1 Hz high pass filter and a 25 Hz low pass filter.Next, we converted both estimates from decibel scale into linear scale (amplitude ratio = 10^(dB/20) and applied a fast Fourier transform to obtain the amplitudes at our 4 beat-related FoI.These amplitudes were then normalized with z-scores.

Statistical analyses
To test the hypothesis that prolonged CI experience leads to enhanced neural tracking of the periodicities of the Alberti bass, we compared the frequency-tagged EEG responses of CIre-T1 and CIre-T2 using a one-sided paired samples t-test.Additionally, we compared the frequency-tagged EEG responses of CIre-T2 with those of CIex using a one-sided independent samples Welch t-test.We carried out a correlation analysis between months of using the CI and the frequency-tagged amplitudes of CIre-T2 and CIex.Finally, we compared the neural responses of CIex against NH controls with an independent samples Welch t-test.
To test if the neural responses involved endogenous activity or just reflected the intensity changes and the spectral contrasts of the Alberti bass, we used z-scores to normalize the frequency-tagged EEG amplitudes for each participant and compared them to the normalized estimates of intensity and spectral changes of the stimuli (see Fig. 3).This comparison was achieved by subtracting the z-scores of each audio feature from the z-scores of the participant neural data at each FoI (Lenc et al., 2020(Lenc et al., , 2021;;Nozaradan et al., 2016).In each group, the subtraction outcomes were tested with one-sample tests against zero, and the p-values corrected for multiple comparisons using Bonferroni (p-value x 2 features x 4 FoI).

Comparisons across groups of participants
The one-sided paired sample t-tests comparing the responses of CIre-T1 and CIre-T2 responses revealed that the frequency-tagged amplitudes were significantly higher after approximately 3 months of CI experience (M = 0.043; SD = 0.026) compared to shortly after device switch-on (M (caption on next column) Fig. 1.Applying frequency tagging to the EEG of participants listening to the Alberti bass.(a) Four repetitions of the Alberti bass (in music notation) and the binary metrical groupings of the tones (schematic tree branches).This musical pattern is used in the no-standard musical multifeature (MuMuFe) paradigm, which includes deviant tones (in red).(b) The grand average eventrelated potentials (ERPs) of the participants (NH, CIex and CIre-T2) listening to the four-tone pattern of the Alberti bass.These illustrative ERPs are obtained from six frontocentral electrodes after segmenting the trials into four-tone chunks and averaging them at the individual level using a mean baseline.The onset of each tone (gray bars) is marked as a dotted line.(c) Frequencytagged amplitudes of six frontocentral electrodes averaged across trials and groups.The triangles indicate our four frequencies of interest, directly related to the metrical groupings of the Alberti bass.= 0.036; SD = 0.026; t (7) = 2.357; p = .025;Cohen's d = 0.833).See this comparison in Fig. 2b.
The one-sided independent samples Welch t-test confirmed that the frequency-tagged amplitudes of CIex were significantly higher (M = 0.067; SD = 0.023) than those of CIre-T2 (M = 0.043; SD = 0.026; t (13.9) = 2.167; p = .024;Cohen's d = 0.984).See this comparison in Fig. 2c.Furthermore, there was a significant correlation between the duration of CI experience in these groups and the frequency-tagged amplitudes: Spearman's rho = 0.517, p = .018,N = 21 (see the correlation results in Table S3 and Figure S1 in Supplementary Materials).
The additional analysis comparing the event-related potentials across tone positions showed that the amplitudes of the P50 and N100 components were not directly dependent on the size of the preceding pitch interval (see Supplementary Materials, Figure S4, Figure S5, Table S6 and Table S7).

Comparisons between neural activity and stimuli features
The one-sample tests against 0 comparing the z-scored amplitudes between frequency-tagged EEG and stimuli Intensity changes revealed that there was a significant increase in the EEG at the frequency (f/2 = 2.44 Hz) grouping two tone onsets in NH and CIex (all p < .001).This effect did not survive Bonferroni correction in CIre-T2 (see Table 1 and Fig. 3).The amplitudes related to the frequency of the downbeat of the Alberti bass (1.22 Hz) were less salient in the EEG than in the Intensity estimates of the stimuli in NH (p = .040).The amplitudes related to the frequency of the tones (4.88 Hz) were less salient in the EEG than in the Intensity estimates of the stimuli in CIex (p = .008).
The one-sample tests against 0 comparing the z-scored amplitudes between frequency-tagged EEG and stimuli Spectral flux changes revealed that there was a significant increase in the EEG at the frequency (f/2 = 2.44 Hz) grouping two tone onsets in NH and CIex (all p < .001).This effect did not survive Bonferroni correction in CIre-T2 (see Table 1 and Fig. 3).The amplitudes related to the frequency of the downbeat of the Alberti bass (1.22 Hz) were less salient in the EEG than in the Spectral flux estimates of the stimuli in NH (p = .016).Similarly, the amplitudes at the first harmonic of the tone frequency (9.76 Hz) were less salient in the EEG than in the Spectral flux estimates of the stimuli in CI users (both p = .008).
The one-sample tests against 0 for CIre-T1 are reported in the Supplementary Materials (Table S4).The effects were similar to the ones reported for CIre-T2.

Discussion
Using frequency tagging analysis on EEG data from a previous MMN study, we found that prolonged experience with the CI leads to enhanced neural synchronization to musical rhythms.In other words, the longer the duration of CI use, the larger the frequency-tagged amplitudes at the main periodicities of the Alberti bass.In accordance with our first two hypotheses, this trend is already evident from the initial months of CI usage and increases over time.In contrast to our third hypothesis, the neural responses of experienced CI users increasingly approached those of NH controls (see the event-related potentials and corresponding frequency spectra in Fig. 1), even involving similar scalp topographies (see Fig. 2a) yet remaining significantly smaller in amplitude.Similar findings appeared when we included neural activity elicited across groups at other periodicities relevant to the four-tone pattern (see Supplementary Analyses: Extended FoI).
The Alberti bass that we used presented isochronous sounds (i.e., temporal cues) by repeating them inside a melodic contour with changes in pitch (i.e., spectral cues) and amplitude (i.e., acoustic features), which may have affected the grouping of the events.To test the fourth hypothesis, we compared the frequency spectra of the participants' neural responses against the frequency spectra of the spectrotemporal features of the stimuli (i.e., intensity and spectral fluctuations over time).This revealed that the neural activity was not a linear transformation of the stimuli amplitudes.Rather, it worked like a filter that enhanced or diminished some amplitudes to generate relative enhancements at certain frequencies (e.g.2.44 Hz, in Table 1 and Figure 3; and Table S8 in Supplementary Analyses: Extended FoI).However, we cannot conclusively interpret this effect because of the intrinsic nonlinear nature of the neural system that transforms the not-purely sinusoidal stimulation into neural responses over several frequencies or harmonics (Heinrich, 2010;Retter et al., 2021;Rossion, 2014).
We also found that the neural activity of the participants did not just reflect the processing of stimuli features (i.e., intensity or spectral fluctuations over time), but nevertheless involved a selective enhancement of a frequency comfortable to perceive the musical beat inside a quaternary metrical context: f/2 = 2.44 Hz (see Fig. 3c and 3d).The endogenous binary structuring of the tones was robust in NH and CIex participants, but less consistent in CIre-T2, supporting the idea that the neural responses of CI users approach those of NH with prolonged use of the CI device.This relative neural enhancement reflected the grouping of consecutive tones into binary events, in line with our natural tendency to group events into binary structures (Brochard et al., 2003;C. Møller et al., 2021).At this periodicity, the perceived beat occurs every 410 ms (146 BPM), falling close to human spontaneous motor tempo: 500-650 ms (Desbernats et al., 2023;McAuley et al., 2006;Moelants, 2002).This periodicity also corresponds to the repetition of the 5th note (the second and fourth tone) in the Alberti bass pattern, which appears as a peak at 2.44 Hz in the frequency spectra of the auditory estimates and could have helped the participants to perceptually anchor the beat.In fact, it is the combination of temporal and melodic features (e.g., contour change or melodic repetition) that underlies the perception of the musical meter (Hannon et al., 2004) and its differently accented structure (Toiviainen and Eerola, 2006).

Early adaptation to CI
This study explores how CI users process periodic rhythms.It reveals a subtle yet significant increase in the neural responses synchronized to these rhythms already within the initial three months of CI use.Notably, experienced CI users exhibit larger amplitudes compared to those of recently implanted.The difference in synchronized neural activity can be interpreted as a reflection of the brain's re-learning process (Chen, 2016).Over time, the brain may adjust and adapt to the input provided by the CI (Glennon et al., 2020), although this adaptation is constrained by the age at which the CI intervention occurs (Lazard et al., 2023;Petersen et al., 2015).
The early increase in synchronized activity found in CIre implies swift adaptation to the cochlear stimulation after the CI switch-on.A precise encoding of the musical periodicities facilitates the processing of temporal information because it establishes referential points in time that allow the brain to organize, group and categorize events accurately over time within a metrical framework (Honing, 2013).Through beat perception, the brain is able to optimize temporal information by filtering out irrelevant periodicities while enhancing those that are relevant to the perceived metrical structure.This process is sometimes referred as "periodization" of the input signal (Lenc et al., 2021) and helps to parse acoustic stimuli when they are weakly periodic (Lenc et al., 2020).Tracking a periodic beat is also necessary to reduce perceptual entropy in complex rhythms (Milne and Herff, 2020;Ravignani and Madison, 2017) and even correlates with sensorimotor synchronization skills (Nozaradan et al., 2016).In the present work, one could interpret the increase in amplitude and decrease in variance shown in Fig. 2b as a sign that the neural responses progress towards a more accurate synchronization to the periodicities of the Alberti bass as CIre become more acclimated to their CI device transmitting sounds.Future research involving complex rhythms and both neural and behavioral paradigms could investigate with improved designs how the brains of CI users quickly adapt to the CI inputs, and how neural changes translate into behavioral performances.

Extended adaptation to CI
The greater neural responses found in the group of experienced CI users suggest that the neural adaptation period extends to several years.
Yet, the amplitudes of the CIex group remained smaller than the ones seen in NH controls.This could either indicate that the neural adaptation to the auditory input from the CI is still ongoing, or that there are physiological constraints to the plastic changes occurring postimplantation; perhaps reaching a functional plateau, which may affect the strength of neural activity in different cortical regions (for changes in topography, see Jordan et al., 1997;Sandmann et al., 2015).Another explanation is that we may have been unable to capture the fine-grained cortical changes that may occur after extended CI experience, perhaps due to the applied independent component analyses to remove CI-related artifacts.Nevertheless, our findings are consistent with the findings from Alemi et al., (2021), who also found significant differences in frequency-tagged amplitudes between experienced CI users (~7 years of CI experience) and NH controls listening to single-tone stimuli.The comparison between the neural responses and the changes in the stimuli features suggests that CIex already learned how to interpret the CI input to consistently track the periodicities grouping the tone contrasts.This was less evident for CIre, who were perhaps still learning to decode the signal from the CI.Consistently detected changes in intensity (i.e., CI electrical current) and spectral content (i.e., CI electrode location) may be relevant to generate a clear beat percept; though more research is needed to dissociate the contribution of these two features in neural synchronization (see Wollman et al., 2020).
Altogether, these findings indicate that the neural synchronization to the rhythmic patterns of the Alberti bass improves rapidly within the first months of CI use and may continuously improve over time.Future research should investigate whether these neurophysiological findings correlate with behavioral performances, such as auditory tests with syncopated rhythms that require abstracting a beat at a frequency different from the tone onset rate, or sensorimotor synchronization tasks that involve an accurate coupling of motor outputs to sensory inputs.

Beat perception and synchronization in CI users
Previous behavioral studies on rhythmic perception confirm that the Table 1 One sample tests against zero for beat-related frequency amplitudes obtained by subtracting normalized stimuli features from normalized neural activity.For each group of participants (NH, CIex, CIre-T2), the table reports the Student t-tests (if the mean is different from 0) or the Wilcoxon signed-rank tests (if the median is different from 0) that compare the amplitudes found in neural activity to those present in stimuli features (Intensity and Spectral flux) at four frequencies of interest (1.22,2.44,4.88 and 9.18 Hz).For each group, Bonferroni was applied to correct for multiple comparisons (p-value*8).Note that after this conservative correction the tests still showed an increase at the frequency for the grouping of two tones (f/2 = 2.44 Hz) for the NH and CIex participants.FoI stands for 'frequencies of interest'.adaptation to the CI may occur at an early stage.Experienced CI users perform comparably to NH controls in rhythmic discrimination tasks, such as detecting tempo changes (~11 years of CI experience, in Kong et al., 2004) or beat jittering (~3 years of CI experience, in Kim et al., 2010), and they are capable of replicating previously heard rhythms (~4.5 years of CI experience, in Limb et al., 2010).Despite our observation of reduced response magnitudes in CI users compared to NH controls, this research indicates that general rhythmic discrimination abilities, or at least those that involve simple beat-related tasks, might not be affected by less neural synchronization to the periodicities of the tones.
In contrast, since tracking the beat underlying rhythms involves the activation of auditory-motor areas (Cannon and Patel, 2021;Grahn and Brett, 2007;Kasdan et al., 2022;Merchant et al., 2015), the performance of more demanding sensorimotor synchronization tasks could be affected by the reduction of neural activity related to spectrotemporal periodicities of the stimuli.For example, while CI users (~7 years of CI experience) are able to synchronize their body to a rhythmically complex merengue song, their performance is less accurate than the performance of NH controls (Phillips-Silver et al., 2015).Similar differences in sensorimotor synchronization skills also appear between early pediatric CI users (5-10 years old, with ~4.5 years of CI experience) and NH controls: the CI group is less accurate in tapping along with a metronome, musical excerpts or complex rhythmic patterns (Hidalgo et al., 2021).Based on these findings, one could speculate that higher-level temporal deficits to extract and predict a beat in hierarchically structured patterns could be the cause of these sensorimotor difficulties.Nevertheless, more research is needed to establish any direct link between these rhythmic motor outcomes and neural synchronization to rhythms in CI users.
This frequency tagging study contributes to previous EEG research on phase-locked responses to periodic stimuli in CI users (Alemi et al., 2021) and expands it by incorporating musical tones organized within the quaternary metrical structure of the Alberti bass.Our stimuli offered a richer harmonic context and a binary-based meter to structure auditory events, providing evidence for neural synchronization to musically driven groupings beyond a steady metronome.In line with Alemi et al., (2021), the experienced CI users in our study showed weaker synchronized neural responses as compared to NH controls.Both results point in the same direction: experienced CI users may have lower neural synchronization to the frequencies of the auditory stimuli (at least in the frontocentral regions that include Fz, FCz and Cz), and this could affect behavioral performance, such as the before mentioned sensorimotor synchronization difficulties.This interpretation, however, does not exclude the possibility that relevant neural activity was altered during the removal of CI-related artifacts, nor that other brain regions closer to the CI could compensate for this lower synchronized activity.In fact, a positron emission tomography (PET) study reported greater activity for rhythmic stimuli in experienced CI users than NH controls in certain auditory regions of the temporal lobe (Limb et al., 2010).However, the increased activity found in PET may reflect other primary auditory processes independent from the steady-state evoked potentials keeping the beat in frontocentral regions.

Tracking the underlying beat or lower-level acoustic features?
Here, we used ecologically valid musical stimuli with varying acoustic properties: intensity (different loudness) and spectral content (alternating tone contrasts).Our original aim was to assess beat perception (including metrical levels) in the cortical responses of CI users, but the musical nature of the periodic stimuli made its contribution difficult to discern from the processing of fluctuating spectrotemporal features.While the periodicities of the low-level acoustic features of the Alberti bass helped to generate predictions in the MMN MuMuFe paradigm, the same periodicities unfortunately overlapped with the frequencies related to the musical beat and its metrical groupings.Being aware of this unavoidable constraint, we implemented complementary analyses exploring the spectrotemporal features of the stimuli (Section 3.2) and differences in the event-related potentials of NH participants (Supplementary Materials) with the aim to inform about any possible dissociation between the metrical organization of the beat and the processing of low-level acoustic features.
The comparison of the amplitudes of the neural data against the amplitudes of the acoustic features (Section 3.2) revealed a relative increase of activity at 2.44 Hz (see Fig. 3c and 3d).This increase also appeared in NH and CIex when more FoI were included (see Supplementary Materials, Table S8), though the finding was more consistent when the neural data was contrasted to spectral content fluctuations than to intensity fluctuations.One could attribute the smaller contrast with intensity estimates to the inner function of metrical organization: organizing a beat into patterns of strong and weak events (inherently connected to "intensity").A speculative interpretation of the relative increase at 2.44 Hz could be that it reflects an endogenous grouping of consecutive tones into binary events, which would be in line with our natural tendency to organize events into binary structures (Brochard et al., 2003;Møller et al., 2021).At this periodicity, the perceived beat would occur every 410 ms (146 BPM), falling close to human spontaneous motor tempo: 500-650 ms (Desbernats et al., 2023;McAuley et al., 2006;Moelants, 2002).Another interpretation could be that the increase somehow reflects (or it is boosted by) the repetition of spectral content at this periodicity, since the second and fourth tones of our Alberti bass are identical, but the current analyses do not allow for testing this idea.As previously mentioned, interpretations of increased amplitudes at particular frequencies need careful consideration, due to the non-linear nature of the neural responses processing the stimuli.
From a multifaceted beat perception approach, the metrical structure of musical rhythms arises by combining both temporal and melodic information (Hannon et al., 2004;Honing, 2012Honing, , 2013;;London, 2012;Ravignani et al., 2018;Toiviainen and Eerola, 2006).In terms of melody, our four-tone pattern involved a repetitive alternation of two kinds of pitch contrasts (i.e., large-small-small-large…) emerging from the intervallic distances between the position of the notes in the arpeggiated chord (e.g., C-G-E-G-C…).If the responses elicited by larger pitch contrasts were different from those elicited by smaller pitch contrasts, one could attribute the relatively increased responses at 2.44 Hz to differences in the amplitudes of the components.However, different mean peak amplitudes appeared after the same pitch contrasts, while similar mean peak amplitudes appeared after different pitch contrasts (see Figure S5 and Table S6 in Supplementary Materials).Although these results suggest that the repetitive alternation of pitch contrasts might not be driving the neural enhancement at 2.44 Hz, some limitations weaken this interpretation.First, the fast rate of the tones limited our analyses to early components of the ERP (i.e., P50 and N100), while pitch-related processes could be occurring at a later stage, perhaps overlapping with the following tones.Furthermore, the lack of significant differences between tone 2 and 4 (i.e., same tone after small and large pitch contrast) does not imply proof of similarity between them.Interestingly, the attenuated N100 amplitudes aligned with the attenuated responses for off-beat events discussed in Bower et al. (2024), which were also reported in Fitzroy and Sanders (2015) for metrically-weak sounds.
Given the characteristics of our stimuli, we cannot fully disentangle the contribution of the internal temporal processes organizing the isochronous events and the low-level responses to melodic features (e.g., contour change or melodic repetition) underlying the perception of the musical meter and its differently accented structure.Therefore, our frequency-tagged neural responses may reflect both exogenous responses to low-level acoustic features (e.g., melodic contour repetition, accented amplitudes) and endogenous non-linear transformations enhancing their relevant periodicities.Future research should use more appropriate paradigms to disentangle the interaction of bottom-up and top down mechanisms.

Implications of rhythmic synchronization
Synchronization to structured rhythms is necessary for brain processing of the temporal auditory events, which is crucial for music enjoyment and engagement.The prediction and grouping of upcoming sounds into meaningful rhythmic patterns and melodic motives (Hannon et al., 2004) is essential for musical reward in listeners, and allow music players or dancers to synchronize their movements with recurring temporal events (Vuust et al., 2022).This may have a direct impact on the way music is experienced and the emotions it evokes (Brattico, Brattico and Jacobsen, 2009;Koelsch, 2020).Indeed, a recent study on the subjective emotional experience of music (Yüksel et al., 2023) revealed that both CI users with prelingual and postlingual hearing loss rank lyrics and rhythm ("strong and captivating rhythms") as the features that evoke emotions most strongly.This suggests that the combination of linguistic content and rhythmic patterns is what affects the emotional responses to music regardless of the history of hearing loss.These findings place rhythm training as an important aspect for the development of musical rehabilitative interventions in CI users (see Pesnot Lerousseau et al., 2020).
Temporal encoding of rhythms plays a crucial role in both music and language processing.Rhythmic synchronization not only regulates the processing of musical structure but also assists in delineating the quasirhythmic patterns of syllabic stress in speech, directing attention to the pertinent timing of information (Gross et al., 2013;Kotz and Schwartze, 2010).This phenomenon underlies syntactic and prosodic computations in language (Glushko et al., 2022).Research has demonstrated that presenting rhythmic primes following the stress patterns of spoken utterances enhances phonological processing (Cason and Schön, 2012), and vocal audio-motor training further strengthens this effect by consolidating stressed metrical patterns (Cason et al., 2015a).Interestingly, this effect extends to hearing-impaired children, especially in the context of CI users, where rhythmic cues demonstrate more pronounced benefits compared to those using hearing aids (Cason et al., 2015b).Furthermore, training CI children with regular rhythmic primes also enhances grammatical judgments of morphosyntactic errors (Bedoin et al., 2018;but see McKay, 2021).The effect of rhythmic priming on phonological perception and production is likely due to the engagement of domain-general processes generating temporal expectations (Tillmann, 2012; see also Torppa and Huotilainen, 2019).Hence, the rapid neural rewiring observed here during the initial months of CI use could be explained by cross-fertilization across cognitive domains for which accurate temporal predictions are needed.

Benefits of using EEG frequency tagging in CI research and current limitations
The use of frequency tagging on EEG responses to music can serve as a naturalistic tool to track the recovery of auditory temporal encoding in CI users.This method is accurate for studying consistent periodic activity, such as the temporal expectations of metrical levels underlying rhythmic and melodic motives, but less optimal when the tempo fluctuates inconsistently.Importantly, the combined mismatch responses and frequency tagging analyses of the no-standard MuMuFe paradigm reported here offers the possibility to evaluate complementary aspects of music perception (auditory content: pitch, timbre and intensity; and auditory organization: rhythm, beat-related periodicities) in one single experimental set-up.Frequency tagging the neural responses synchronized to musical rhythms (e.g., Doelling and Poeppel, 2015) could therefore become a useful tool to evaluate how these interventions progress over time, assessing the links between the subjective appraisal of music and the brain responses adapting to the CI.
This work is a re-analysis of EEG data from CI users participating in two studies designed to elicit MMN to identify musical feature discrimination (Seeberg et al., 2023;Petersen et al. 2020) and not beat perception per se, which entails intrinsic limitations.Firstly, in the present MuMuFe paradigm, all the tones of the Alberti bass occur at the same inter-onset interval, except for the rhythm deviant, which randomly occurs earlier in four possible ways.This means that our findings are generalizable to relatively simple rhythmic organizations, involving groupings of the tone onset rate at frequencies already present in the stimuli.This makes any neural enhancement at beat-related periodicities methodologically challenging to be detected (Henry et al., 2017).Future studies should assess beat perception in CI users listening to more complex syncopated rhythms while controlling for differences in the acoustic properties.Secondly, during the EEG recordings, participants were watching a silenced movie as the MMN is a pre-attentive component of the event-related potentials.Previous studies, however, showed that attention to metrical levels of the beat modulates the obtained frequency-tagged neural amplitudes (Nozaradan et al., 2011;Celma-Miralles et al., 2016;Celma-Miralles and Toro, 2019).Crucially, this modulation would also apply to NH controls, leaving the relative difference between groups similar.Thirdly, the poor spatial resolution of the EEG and artifacts caused by the CI makes it challenging to identify the adaptation-specific neural sources.Future studies should record more EEG channels to allow for more precise source reconstruction analyses (see Alemi et al., 2021) or combine EEG with neuroimaging techniques insensitive to the CI artifacts, such as fNIRs (Sherafati et al., 2022) or PET (Limb et al., 2010;Petersen et al., 2013).

Conclusion
Using frequency tagging, we analyzed the EEG data of two MMN studies to obtain reliable neural measures for rhythmic perception in CI users and NH controls.The findings show that steady-state evoked potentials at frequencies related to a repetitive four-tone pattern are detectable shortly after CI switch-on and suggest that their magnitudes continue to increase even after a short period of time.The results also support that the brain adaptation to musical information keeps evolving for several years, as the neural responses to the rhythmic spectrotemporal features of the stimuli increase from the recently implanted CI users to the more experienced CI users.These neural responses were, however, weaker in CI users compared to NH controls, possibly due to CI-related artifacts.In the future, applying frequency tagging to EEG responses could serve as a valuable tool for assessing the neural adaptation in CI users, specifically by targeting the temporal encoding of rhythms.This feature holds significance for both music and language processing.
Pseudonymized data will require a data-sharing agreement.For inquiries or requests contact corresponding author Alexandre Celma-Miralles (a.celma.miralles@clin.au.dk).

Fig. 2 .
Fig. 2. Mean of beat-related amplitudes compared across groups.(a) Topographies of frequency-tagged amplitudes averaged across the frequencies of interest (FoI) related to the main periodicities of the Alberti bass: 1.22, 2.44, 4.88 and 9.18 Hz for recently implanted CI users (CIre) at T1 and T2, experienced CI users (CIex) and normal hearing (NH) controls.The gray dashed triangle depicts the frontocentral cluster of electrodes used for the statistical comparisons in (b) and (c).(b) Vertical raincloud plot depicting the increase of averaged frequency-tagged amplitudes in the recently implanted CI users from shortly after switch-on to after ~3 months of using the CI device.(c) Raincloud plots showing averaged beat-related amplitudes in CIre-T2 (bottom), CIex (middle) and NH (top).*** p < .001,* p < .050.

Fig. 3 .
Fig. 3. Comparison between neural amplitudes and stimuli features: intensity and spectral flux changes over time.(a) First five seconds of the sound waveform of the Alberti bass, its spectrogram and the MIRtoolbox estimates of the Intensity and Spectral flux changes in decibels over time.(b) Frequency spectra of the stimuli features as logarithmic values in the decibel scale (first column) and as linear amplitudes in arbitrary units (second column), with circles signaling our four frequencies of interest.(c) Subtraction of the normalized Intensity amplitudes from the normalized neural amplitudes for each group.(d) Subtraction of the normalized Spectral flux amplitudes from the normalized neural amplitudes for each group.Positive differences from zero indicate that a particular periodicity is enhanced in the EEG (e.g., f/2).Note that asterisks only mark one-sample tests significantly greater than zero ("black" for Bonferroni-corrected p-values; "gray" for uncorrected p-values).*** p < .001,* p < .050.