Several items have been considered in this study: the coding scheme, the influence of noise and the simulation of CI coding in NH subjects.
4.1. On the Coding Strategy
The choice of a coding strategy is a delicate matter and some studies have shown that CI users have a subjective preference for a particular strategy that is not always the one that yields the best performances [
32].
Additionally, many technical parameters concerning the coding scheme, such as stimulation rate, gain control, update rate and filter settings, influence the final results and have an effect on the performances related to the coding strategy [
28,
33].
It is interesting to note that the four manufacturers have taken different stimulation strategies and the results are dispersed; it is therefore difficult to draw definitive conclusions. For any manufacturer, all coding strategies can be implemented within the processor.
From our results with NH listeners, the Bonferroni correction lowered the significance limit but the Cohen analysis indicated that the differences were reliable. The FC strategy presented better recognition percentages than CP, particularly in the SNR range 0 to +6 dB with effect sizes medium (0 and +6 dB) and strong (+3 dB). Moreover, the comparison between the x50% highlighted a strong effect size in favour of the FC strategy.
Of course, our results only stand for a CP strategy with an extraction of 8 channels out of 20 and an FC strategy with 20 channels. However, this seems to be an interesting hint, as the conditions and the subjects were the same for both strategies (CP or FC), used the same random approach, had the same SNRs and had the same signal processing (window length, sampling rate, channel band-pass, quantization, etc.), the subjects were of the same type (range of age, education, etc.) and we were able to use paired comparisons. However, this simulates an “ideal case” where there is no channel interaction and no pitch-shift due to the insertion depth of the electrode array and where all the channels are functional. This is why these results should not be taken on their own without taking into account the experiment conducted with the CI subjects.
With the CI subjects, considering the non-saturation of the percentages, we raised the SNR up to +18 dB and FC led to, albeit not significant, higher scores than CP in the 0–12 dB range. A small inversion of the results was also observed above +12 dB SNR, which can be linked to the significant interaction between SNR and coding scheme. Nevertheless, results are to be taken with caution, considering the wide dispersion of the results and the comparison of relatively small and unbalanced groups. Because of the difference in the number of included patients in each group, we used nonparametric tests to compare them; they are well adapted for this kind of comparisons. While they are generally less powerful than parametric tests, they are more robust. However, like in the experiment with NH subjects, the comparison between the x50% showed an interesting difference in favour of the FC strategies. With a medium effect size, this result would be worth investigating further in another study.
Our results are consistent with studies that showed a superiority of the FC strategy over the CP strategy, particularly in noise [
5]. This was less true when CP strategies, such as ACE, with a high stimulation rate were introduced [
6,
7]. No clear advantage for a particular coding scheme can be identified taking all the literature on the subject.
In many studies, when the FC strategy was used, the stimulation rate was an important factor as the possibility to follow the quick changes in the signal helps the recognition performances mostly for consonants [
34,
35].
4.2. Cochlear Implant Users and Normal-Hearing Subjects
Despite the fact that the groups of CI users were heterogeneous, the general recognition behaviour was the same for CI users and NH subjects, whatever implant was used. With NH listeners, for a SNR of +9 dB, the 100% recognition level was reached.
With CI users, the plateau was not always reached with a SNR of +18 dB; additionally, it was below the 100% measured with NH subjects. For a +9 dB SNR (maximum tested in simulation), the CI users’ performances were below the scores observed with NH subjects; the mean scores with CI users ranged from 50 to 75%; this is consistent with previous studies [
36].
The same results were also seen with the x50% (sigmoid fitting), which was better in NH subjects than with CI users.
With the CI users, an inversion occurred between +12 and +15 dB SNR and it was also seen for +18 dB SNR; performances observed with a CP strategy were higher than the performances with an FC strategy.
The reliability of the data obtained from CI users is a real issue. Is there a link between the plateau and the
x50%? The scatter point diagram of the four CI user populations is shown in
Figure 5. It shows that, for each manufacturer, all possibilities exist, either with a good plateau and a poor
x50% or vice versa. All intermediate situations were found and the correlation coefficients were not significantly different from zero.
As the very goal is to provide an opportunity for every CI user to hear in everyday life [
28], the work ahead is important. The efficiency of a CI is affected by many factors such as the recognition and linguistic skills, the degrees of nerve survival and the technical choices that are made when fitting the device and the variations are wide with every subject.
4.3. Listening in Noise
Listening in noise is a clear challenge, which is not handled in the same way by CI users and NH people. Noise flattens the spectrum and the subsequent structures in the auditory system do not react identically [
37]. The study of speech recognition in noise has become of great interest as it is present in daily listening conditions. Additionally, we can see the coding behaviour for different SNRs (floor and ceiling effect and intermediate situation).
In this study, the CI user group was older on average than the NH group. In general, older people have lower speech perception scores in noise, even with normal or age-related hearing, compared to young people. However, the purpose of the study was to test both groups and see if a similar trend (between FC and CP coding schemes) could emerge and not to compare CI users with NH subjects.
Another finding was the effect of noise on performances for each strategy, which makes this study interesting in that each manufacturer can set any coding scheme on their devices. Consequently, it is worthwhile to evaluate the results through different approaches. Our work may suggest that the strategy is noise-dependent.
The number of channels needed to understand speech in noise or in quiet is an important issue. Studies have indicated that, when speech is processed in the same manner as in CIs and presented in quiet to NH listeners, sentence recognition scores are higher than 90% with as few as 4 channels [
38,
39]. In the literature, results show that more channels are needed to understand speech in noise than in quiet [
10] but selecting more than 12 channels may not yield significant improvements on the recognition performances [
21]. These considerations orientated the choice of our parameters.
In noise, performances of CI users reach a plateau as the number of channel increases; for NH subjects performances continue to increase (up to 100%), suggesting that, CI subjects could not fully utilize the spectral information provided by the number of electrodes, possibly because of the channel interaction [
38]. As indicated above, trends are similar for NH and CI listeners but results are not interchangeable. It is sensible to say that more channels imply more information but this also implies more overlap between the electrodes. This conflict needs to be studied in the future; we can simulate channel interaction with NH subjects.
The acoustical material (in our case the Fournier’s lists and the cocktail party noise) seemed to be well adapted to the situation.