*3.3. Music Test*

Scores decreased with increasing levels of difficulty from tests 1 to 4 for both CrystalisXDP and standard programs (Figure 3). Scores for tests 1, 2 and 3 were above chance level (8.81 ± 0.25 for test 1, *<sup>p</sup>* < 10<sup>−</sup>4, 6.87 ± 0.25, for test 2, *<sup>p</sup>* < 10<sup>−</sup>4, and 5.43 ± 0.20, *<sup>p</sup>* < 0.05 for test 3, *n* = 42, one-sample test). In contrast, the average score for test 4 was not different from the chance level (5.02 ± 0.31, *n* = 42, not significant, one sample test, Figure 3).

The short period of adaptation could have advantaged CrystalisXDP over the standard program in those who already used CrystalisXDP and represented the majority (15 out of 21). A mixed-model analysis (restricted maximum likelihood approach) comparing the results for music tests 1 to 4 with CrystalisXDP and standard strategies in patients who regularly used CrystalisXDP versus those who regularly benefited from the standard program showed a significant effect of the test levels (DFn = 3, DFd = 76, F = 32.15, *p* < 0.001) and the strategy during the test (higher scores for CrystalisXDP versus standard, DFn = 1, DFd = 76, F = 5.76, *p* < 0.05). However, the usual strategy used by the patients before inclusion did not have a significant effect on the test results (CrystalisXDP versus standard, DFn = 1, DFd = 76, F = 0.12, not significant). There was no interaction between these factors (test level\*tested strategy: DFn = 3, DFd = 76, F = 1.52, *p* = 0.214; test level \*initial strategy: DFn = 3, DFd = 76, F = 1.31, not significant; Tested strategy\*usual strategy: DFn = 1, DFd = 76, F = 0.021; not significant; test level\*tested strategy\*initial strategy: DFn = 3, DFd = 76, F = 0.081, not significant). A Tukey's multiple comparison test applied to this model showed a higher level of scores for test 1 in comparison to all other tests (*p* < 10<sup>−</sup>4), a higher score for T2 in comparison to test 4 (*p* < 0.001), and higher scores for T3 versus T4 (*p* < 0.05).

**Figure 3. Scores for music tests with standard (MPIS) and CrystalisXDP sound processing strategies.** Each test was marked out of 10, and the total score out of 40. Bars represent mean ± SEM (*n* = 21). Scores decreased with the difficulty level (\*: *p* < 0.001, mixed model analysis). Patients performed better with CrystalisXDP than with standard program (*p* < 0.05) regardless of their usual strategy (effect not significant). Total scores were also higher with CrystalisXDP than with MPIS regardless of the patients' usual strategy (\$: *p* < 0.05, mixed-effects analysis). Box and Whiskers plot represents first and third quartiles, median, and range. Mean is depicted by (+). Dashed line represents chance level.

As assessed by the total score, patients also performed better with CrystalisXDP than with the standard program regardless of their usual strategy (mixed-effects analysis, DFn = 1, DFd = 38, F = 4.98, and *p* < 0.05 for the effect of the tested strategy; F = 0.644, not significant for the effect of usual strategy, and F = 0.046 not significant for tested strategy\*usual strategy, Figure 3). Higher scores with CrystalisXDP suggested that patients exploit some spectral-based cues in addition to the rhythm to distinguish between happy and sad music.

There was no statistical difference between the total scores at the first and second sessions, suggesting that there was no effect of order (global scores 30.5 ± 5.19 vs. 31.2 ± 5.23, respectively, mean of differences: 1.52, not significant, paired-*t*-test, *n* = 21). The test–retest reliability of the total score was good between the two sessions (Cronbach alpha = 0.87, average R = 0.77).

Musical background was significant in this population. Ten patients used to sing in their childhood (47%). Among these, five continued singing during adulthood and even after CI. Seven declared playing an instrument in their childhood: drums (*n* = 1), flute (*n* = 1), piano (*n* = 3), accordion (*n* = 1), and clarinet (*n* = 1). Only four pursued their hobby as an adult. Five singers also played an instrument. Singing before CI tended to improve scores regardless of strategy (*p* = 0.05, 2-way ANOVA, Table 4), but there was no effect of playing an instrument or training with CI on the scores (not significant, 2-way ANOVA).


**Table 4.** Music test scores as a function of musical experience and training. Total scores for music tests are presented as Mean score ± standard error of mean [range] for each subgroup.

Total music scores were correlated with WDS (Figure 4). Total music scores appeared to be influenced by the number of active electrodes. Although there was no correlation between the number of electrodes and the total score (Figure 5), patients with more than 15 electrodes (*n* = 14) performed better with CrystalisXDP sound processing programs (28 ± 5.89, *n* = 7 for patients with <15 electrodes versus 33 ± 3.93, *n* = 14, t(19) = 2.18, *p* = 0.042, unpaired t-test). With the standard MPIS program, this difference also tended to be significant (26.6 ± 4.12, *n* = 7 versus 30.8 ± 5.10, *n* = 14, t(19) = 2.07, *p* = 0.052, unpaired *t*-test).

**Figure 4. Correlation between musical test total scores and word discrimination scores (WDS) with cochlear implant (CI) only with standard (MPIS) and CrystalisXDP sound processing strategies.** WDS tended to be correlated with total scores in standard condition (right panel, Y = 0.08 \* X + 20.2, R = 0.47, *p* < 0.05, F test) and was significantly correlated to total scores in CrystalisXDP condition (left panel, Y = 0.09 \* X + 20.5, R = 0.58, *p* < 0.01, F-test).

Total scores obtained by patients with unilateral CI were not different from those with binaural or bilateral CI (31.8 ± 4.57, *n* = 17 versus 29.6 ± 5.04, *n* = 4, with CrystalisXDP, and 29.5 ± 7.68 versus 28.5 ± 6.14 without CrystalisXDP, not significant, unpaired *t*-test). Patients with bimodal hearing did not perform better than those with one or 2 CIs in this population (29.1 ± 3.81, *n* = 7 versus 25.5 ± 1.55, *n* = 14, respectively, with standard program, not significant, unpaired *t*-test, data not shown for CrystalisXDP). Similarly, patient who reported musical training during rehabilitation with CI did not perform better than others according to the total score or the scores obtained for each test (data not shown). Patients performed well at tests 1 and 2 and these scores were highly correlated, suggesting the prominence of rhythmical cues even for small differences in tempo in test 2 (Y = 1.00 + 0.67 X, R = 0.73, *p* < 0.001, and Y = −0.31 + 0.81 X, R = 0.67, *p* < 0.001 for standard and CrystalisXDP were Y: test 2 and X: test 1).

In contrast, only nine (43%) patients could categorize above the random level (score > 5) in test 3 (sad versus happy based only on mode) with the standard or CrystalisXDP programs (average scores 6.4 ± 0.59 and 6.9 ± 0.93, respectively, *p* < 0.001, one-sample test for both). In this group, CrystalisXDP, significantly improved the score in comparison to the standard strategy (*p* < 0.05, paired *t*-test, followed by Bonferroni correction). Similarly, only a few patients could distinguish dissonance above chance level (score >5 at test 4): 6 (29%) with standard program (average score 7.0 ± 1.27, *p* < 0.05, one-sample test) and 11 (52%) with CrystalisXDP (average score: 7.0 ± 1.00, *p* < 0.0001, one-sample test). In this group, CrystalisXDP did not improve the scores (not significant, paired *t*-test, followed by Bonferroni correction).

**Figure 5. Total music scores as a function of the number of active electrodes with standard (MPIS) and CrystalisXDP strategies**. Bilateral and binaural cases are depicted with the number of electrodes in one ear (20 and 12, respectively).

Performances for tests 3 (sad/happy only based on mode) and 4 (dissonance) were similar (5.4 ± 0.24 versus 5.0 ± 0.40, respectively, *n* = 21, average of two programs, unpaired *t*-test, not significant), but not correlated (data not shown), suggesting that these two tasks explored different domains. The duration of the hearing deprivation influenced the scores for test 3: patients with a score >5 with CrystalisXDP had a hearing deprivation period <10 years in all cases (*n* = 8), while those who performed poorer had longer deprivation periods (6 out of 12 with deprivation >10 years, *p* < 0.05, chi-2 test). Performances in test 4 were not related to hearing deprivation period (data not shown). Additionally, scores >5 in tests 3 and 4 were not related to age, sex, number of active electrodes, contralateral hearing aid, or previous training (data not shown).

These poor performances contrasted with the questionnaire results in which the majority (18, 86%) declared hearing the melody most (or best) (Table 3). The performances in tests 3 and 4 were not higher in those who declared detecting wrong notes than others (data not shown).

The subjective ease scores decreased with the level of difficulty (Figure 6). Sound processing programs did not influence the ratings of ease, sound clarity or liking (Figure 6). There was a significant correlation between the total music score and the level of ease rated by the participant for the first test (first trial: Y = 2.58 + 0.17X, R = 0.5, *p* < 0.05, second trial: Y = 0.45 + 0.33X, R = 0.6, *p* < 0.01, F-test, X: score, Y: level of ease), but for more difficult levels involving modes and dissonances (tests 2 to 4), this correlation did not exist (data not shown). Clarity and liking ratings were not correlated with total music scores (data not shown) and were not modified by the program (not significant, unpaired *t*-test, Figure 6).

Interestingly, test 3 (happy versus sad based on mode) was rated as easier than test 4 regardless of the program (3.3 ± 0.16 for test 3 versus 2.6 ± 0.20 for test 4, average scores for 2 programs, *n* = 21, *p* < 0.01, paired t-test), while the performances were similarly poor for both tests. Finally, most patients (*n* = 16, 76%, *p* < 0.05, binomial test) preferred CrystalisXDP to the standard MPIS. Among patients (*n* = 15) who used CystalisXDP before the study, 12 kept their usual program and 3 chose the standard program. In the group using MPIS regularly (*n* = 6), three conserved their program and three switched to CystalisXDP.

**Figure 6. Musical test ratings in terms of ease, clarity and melody liking.** Patients scored each item on an auto questionnaire at the end of each test on a Likert scale (1 to 5). Symbols (\*\*\*) represent individual values (*n* = 21) and bars represent mean. Ease scores decreased with the difficulty level, but programs (standard or MPIS versus CrystalisXDP) did not influence ratings (*p* < 0.001 for test levels and not significant for programs, 2-way ANOVA), unpaired *t*-test versus standard.

#### **4. Discussion**

In this study, we showed that music represents a significant daily activity for cochlear implantees. Our original music test, which assessed the hearing performances and explored the emotional aspect of the music, yielded a total score correlated to word discrimination score. It had a good test–retest reliability and did not have a floor or ceiling effect. It was positively influenced by a higher number of active electrodes. As expected, the test revealed a good detection of rhythmical cues but poor performances in detecting dissonances and musical modes. CrystalisXDP improved the musical test results based on both rhythm and spectral cues. Since MPIS and CrystalisXDP have the same basic coding strategy providing the same rhythmical information, and the fitting parameters were identical for both strategies, the results suggest that this improvement is related to modifications in spectral cues.

Musical experience is difficult to describe and analyze since it deals with several intricate factors such as rhythm, pitch, timbre, melody, cultural references, and complex capacities, such as musical sophistication [44]. The latter parameter is defined by the frequency of exerting musical skills or behaviors and the ease, the accuracy or the effect of musical behaviors, and a varied repertoire of musical behavior patterns can be a source of inter individual variability in music tests [44].

Most of the reported music tests evaluate basic features such as pitch, timbre, and rhythm perception [19,45,46]. However, considering the gap between poor musical hearing performances with a CI and a relatively high music enjoyment [47–49], it is interesting to explore higher levels of music perception such as emotions since it can a lead to better understanding of coping mechanisms and neural plasticity in cochlear implantees [50,51].

The effect of Western musical modes on emotions is well known and appears to be effective even in individuals with little or no musical background [for review, see 52]: the major mode evokes dynamism, joy, hope, force and tenderness, and oppositely, the minor mode elicits passivity, despair, sadness, pain, mystery and obscurity. To control the overall difficulty of the trial, we organized the tests in a gradually increasing order of complexity. The rhythmic cue, known to be largely exploited by the CI patients [53], was employed to mitigate the difficulty of the pitch and mode discrimination. As expected, the performances and the level of ease rated by the participants decreased with a lower contribution of rhythm in the categorization. Without this hint, the average score dropped from excellent to chance level for tests 3 (happy versus sad only based on mode) and 4 (dissonance in a melody). This poor performance was in line with the questionnaire in which only 29% of the patients declared being capable of detecting a wrong note. It is noteworthy that CrystalisXDP, which improves spectral cues but provides rhythmical information similar to MPIS, enhanced the happy versus sad categorization performances based on both musical modes and rhythmical information. Previous reports have shown that in cochlear implantees, both place (i.e., electrode position in the cochlea and its assigned frequency band) and temporal cues (i.e., stimulation pulse pattern and rate) are closely related to each other for pitch perception [54,55]. In our study, while place cues remained the same, temporal cues were modified through spectral modifications by CrystalisXDP. The optimization of the temporal cues might influence the pitch perception and provide a possible explanation for the enhancement of sad versus happy categorization.

However, interestingly, a few patients performed relatively well (scores > 5) for these tasks despite the inherent limitations of CI. Better scores for test 3 (happy versus sad based on only mode) were obtained by patients who had a short time of hearing deprivation (<10 years), suggesting the need for an efficient auditory central pathway in music processing [16]. Scores for tests 3 and 4 were not correlated, while scores for tests 1 and 2 (categorization mainly or partly based on rhythm) were highly correlated. This observation suggests that musical modes may involve a different auditory processing task than the detection of a dissonance in a melody. Another important factor, which may explain high performances in tests 3 and 4, is the above-average spectral and pitch resolution related to a higher neural survival in the implanted ear. The quantity of preserved neurons directly influences the number of functional channels, the channel interactions, and the neural capacity to be stimulated at high rates [17,31–33,56].

The distinction of consonant from dissonant notes from a musical instrument or human voices is directly related to the interval between their fundamental frequencies and mainly detected at the cochlear level [57,58]. A dissonant note with fundamental frequency (F0) too close to the reference note to be resolved by the cochlea produces a rapid variation in total amplitude and a sensation of roughness or beating which can be evidenced on the spectrogram [59]. A dissonant note easily distinguishable by the cochlea from the reference has component frequencies that cannot aggregate with those of the reference note producing an inharmonic spectrum. The participation of central auditory processing in this distinction has been suggested based on observation of subjects with amusia [59], but the exact role of peripheral auditory system and the auditory centers are extremely hard to separate in this process. To this end, CI patients represent an interesting pathophysiological model. Observations on CI patients with contralateral normal hearing are in line with this mechanistic explanation. CI patients appear to be sensitive to dissonance by the perception of roughness, and the information related to the temporal envelope plays an important role in distinguishing harmonicity from dissonance [40]. In our study, reducing the spectral distortions without altering the rhythmic information by CrystalisXDP sound processing strategy improved total scores, leading to the hypothesis that by providing discrete cues on roughness and beating, it could enhance global music perception. This phenomenon may be explained by the reduction in spectral smearing and undesired channel interactions in CI patients. Spectral information directly influences the temporal coding within channels. This possible explanation is in line with the observation that reducing the number of harmonics increases the musical enjoyment in both normal-hearing and CI subjects [60].

Despite their poor performances in tests 3 and 4, patients attributed an above-average score to the clarity and the liking of the melodies, and this discrepancy underlines the difference between performance and enjoyment, an observation that has also been reported by others [45,46]. With time, CI patients develop other musical esthetic criteria, and choose types of music which are easier to listen to (more rhythmical cues, less polyphony, and harmonics) as coping strategies [61]. To enjoy music with CI, postlingually deaf patients need time and effort to gather musical experience with new sensations and auditory landmarks. Pleasant music is a skilled mix of predictable events, which drive expectations, and sparse unpredictable developments leading to surprises, and these expectations are related to the experience of musical pleasure [62,63]. Alterations in timber perception and low pitch resolution deteriorate the melody reconnaissance in CI patients [13] and probably also the predictability. With training, these auditory expectations and surprises can be developed in CI patients [7–9]. Another issue is that musical pleasure seems to increase with stimulus complexity (e.g., musical lines, harmonics, timber) up to an intermediate level, and then to decrease with even more complex sounds [64]. Achieving such a level of performance to detect complexity appears possible in some CI patients, since in our population, 9 declared listening to classical music and 5 to opera, reputed as relatively complex, and 15 declared being capable of even comparing performances. However, this ability probably requires a high number of functional channels in the cochlea and a performant central auditory pathway [24,65].

Many variables, such as number of active electrodes, insertion depth, or duration of hearing deprivation may have an impact on the music perception in CI patients [66] and explain the heterogeneity of the results. However, when attempting to control all variables in a very homogeneous population, one might argue that the observations do not apply to other groups of CI patients and the effect is marginal. In addition, one might oppose the fact that other variables such as sex, age, body laterality, ethnicity and cultural background could still interfere. Moreover, it would be difficult, if not impossible, to control parameters such as electrode insertion depth and electrode position or even musical background and experience in such a population. Consequently, we compared sound processing strategies in a paired cross-over design to limit the potential effect of these factors in the outcome. Despite the heterogeneity, which corresponds to the every-day audiology practice, we could observe a quite significant effect of spectral cue enhancement on the music scores. Using only one or both ears could influence the results. However, interestingly, total scores obtained by patients with unilateral CI did not differ from those with binaural or bilateral CIs. Patients with bimodal hearing had marginal acoustic hearing and were tested in CI-only mode; they did not perform better than those with one or 2 CIs in this population. This is consistent with the experimental conditions, which did not disadvantage monaural patients (twin frontal loudspeakers).

In our study, the adaptation period to new sound processing strategies was relatively short. This could have masked the effect or created a bias. However, CrystalisXDP is not a radical change in strategy in comparison to the standard program. It improves the already installed strategy by a better selection of spectral peaks to code, by increasing the spectral contrast, and by fine-tuning the output compression. There is no change in the frequency-place function, frequency band allocation, the loudness or even the basic strategy, which is the MPIS. A previous publication on this sound processing algorithm had shown a rapid adaptation of the patients with significant improvements of WDS in 30 days [38]. This is consistent with the improvement of music scores with CrystalisXDP, which were correlated with WDS in this study. The short adaptation period could have advantaged CrystalisXDP in the majority who used this strategy before inclusion. However, a mixed-model analysis showed that the strategy used regularly before the inclusion did not affect the results.

To our knowledge, there is no validated test for evaluating the emotional aspects of music or musical experience in cochlear implantees. The Munich Music questionnaire has not been validated but was previously published as a relevant tool to evaluate musical perception in CI patients [42]. This questionnaire appeared to provide coherent and consistent results in cochlear implantees from different countries and cultural backgrounds [42,67–69]. This lack of validation imposes precaution in the interpretation of the results related to this tool. In contrast, Likert scales have been largely used as a validated method for the psychometric evaluation of music perception [70] and auditory handicap [71] and provided coherent information regarding the ease of the tests.

In conclusion, the categorization of happy versus sad music samples only based on musical mode or the distinction of melodies with dissonant notes from harmonious ones did not exceed the chance level. CrystalisXDP, which enhances spectral cues, improved performances in the categorization tasks where some rhythmic information was added to the musical mode. This observation, together with the music experience through questionnaires, suggests that CI patients exploit not only rhythmical indications, but also spectral cues to enjoy music and that tests based on intervals, rhythm and melody recognition cannot fully comprehend these cues. Further work on these potential spectral cues will guide the development of next generation sound processing strategies.

**Author Contributions:** Conceptualization, G.L., E.B. and A.B.G.; methodology, A.B.G.; software, G.G. and E.B.; validation, E.B. and A.B.G.; formal analysis, A.B.G.; investigation, G.L. and G.G.; resources, E.B. and G.G.; data curation, G.L. ang G.G.; writing—original draft preparation, A.B.G. and C.G.; writing—review and editing, A.B.G. and C.G.; visualization, A.B.G.; supervision, A.B.G.; project administration, A.B.G.; Funding: A.B.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** The authors acknowledge Oticon Medical for the financial assistance regarding publication fees.

**Institutional Review Board Statement:** The study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (or Ethics Committee) of CCP grand Est III.

**Informed Consent Statement:** All patients were clearly informed and provided their oral and written consent for this study.

**Data Availability Statement:** Not applicable.

**Acknowledgments:** The authors are grateful to Manuel Segovia Martinez, Michel Hoen, and Dan Gnansia from Oticon Medical for their technical assistance.

**Conflicts of Interest:** The authors declare no conflict of interest.
