*2.2. Population Characteristics*

Twelve women and nine men participated in the test (Table 1). The mean age of the group was 55 ± 2.7 years (23–74). All presented with postlingual deafness. Patients had been implanted for a mean duration of 8 ± 1.2 years [3–19] before inclusion. The hearing deprivation period before implantation was 9 ± 3.1 years [1–48] and the mean age at implantation was 47 ± 2.6 years [19–65].

Seventeen patients (81%) had a unilateral CI (nine right and eight left), three (14%) had a binaural CI, and one (5%) had a bilateral CI. All patients wore their CI more than 12 h per day. Seven patients (33%) with a unilateral CI had a contralateral hearing aid. Before inclusion, 15 patients used CrystalisxDP and 6 used the standard MPIS strategy.

Etiologies of hearing loss were idiopathic in 11 patients (52%), Meniere's disease in 1 (5%), congenital in 4 (19%), advanced otosclerosis in 2 (10%), and traumatic in 2 cases (10%).

**Table 1. Subject demographics.** Age, hearing deprivation and CI experience are expressed in years. Hearing deprivation began by the abandonment of the ipsilateral hearing aid. Process.: Type of Saphyr processor, CI Exp: Cochlear implant experience, F: female, M: Male, L: Left, R: Right, BIN: binaural, BIL: bilateral. Number of active electrodes/total electrodes in BIN and BIL cases are indicated as Right + Left.


The ipsilateral pure-tone average (PTA) was 108 ± 8.8 dB before implantation and 39 dB ± 3.1 in free-field with CI. The aided contralateral PTA was estimated as 79 ± 11.2 dB (*n* = 21) with no response above 1 kHz. The WDS was 6.5 ± 9.88% without CI and with lipreading only, 58.6 ± 22.01 with CI only, and 78.3 ± 19.25 with CI + lipreading.

#### *2.3. Coding and Sound Processing Strategies*

The main peak interleaved sampling (MPIS) strategy was used as the standard strategy in this study [41]. The speech processor (DigiSP) uses a Fourier Frequency Transform (FFT) to extract frequency peaks from the input signal spectrum in the 195–8003 Hz range. Available intracochlear electrodes, or channels (ranging from to 9–20 in this study), are selected for assignment of frequency bands to cover the 195–8003 Hz range using monopolar constant current stimulation. The signal level in each of the bandpass filters is assigned to the active electrodes. Loudness is coded by pulse duration, and pulse amplitude remains constant over time. Active electrodes associated with the highest signal level (spectral maxima) are stimulated in a basal to apical order. The number of transmitted peaks can be modified (default setting: 10 transmitted peaks out of 20 extracted peaks). The number of channels to be stimulated at each cycle is predetermined during fitting. Electrical stimulation rates range from 150 to 1000 pulses per second per electrode (pps/e). The default factory setting is 600 pps/e. Patients in this study used default settings. Only the number of available electrodes changed from one patient to another.

The digital signal processing of CrystalisXDP (Figures 1 and 2) is an evolution of the standard MPIS strategy specifically designed to enhance speech discrimination. It incorporates a multichannel back-end output compression function designated as XDP [38]. The Crystalis coding strategy enhances the FFT analysis by a window analysis in order to suppress artifacts and to extract not only the most salient but also the most relevant peaks to speech discrimination. The signal input spectrum is then processed by a noise

reduction algorithm (Voicetrack®) that is based on a human voice reconnaissance and spectral subtraction. The signal is sent to the XDP transfer function module, which provides an adjustable compression of the electrical dynamic range as a function of the acoustic dynamic range. The knee point can be adjusted independently for four frequency bands: 195–846; 846–1497; 1497–3451; and 3451–8000 Hz, which groups electrodes with a similar energy spectrum for speech. Ninety-five percent of the speech information falls in the area under the knee point in each ambience considered. In this population, a medium preset for the knee point was used (average sound intensity at 70 dB SPL). In comparison to the standard MPIS strategy, CrystalisXDP improves the selection of the most relevant spectral peaks; it enhances the spectral contrast of the signal by a noise-reduction algorithm after the FFT analysis; and finally, it provides fine adjustment of the input–output compression function in order to contain everyday life sounds in a comfortable range.

**Figure 1. Functional Structure of CrystalisXDP.** The system extracts the spectral features of the acoustic input by a 128-point Fast Fourier Transform (FFT). A noise reduction algorithm (VoiceTrack) based on spectral subtraction is then applied to enhance the spectral contrast. The 2 diagrams in the VoiceTrack panel show the simulated electrodograms of a human speech sample (dissyllabic word, 4 s), before (top) and after processing (below), generated by an in-house Oticon Medical simulation program as an example. Finally, the multi-band output compression provides adjustable output levels (*Y*-axis in % of electric dynamic range) as a function of acoustic input (*X*-axis, dB SPL) in 4 frequency bands.
