Speech in noise threshold measurements in cochlear implant users

We previously established that the sentence recognition scores of cochlear implant (CI) users one month after activation are similar to those for normal listeners listening to 8 12 channel acoustic (vocoder) simulations: Scores were near 100% in quiet and in 10 dB SNR as long as aetiology and duration of deafness were taken into account; that electrode insertion depths were between 300-400 degrees and that there was no scala dislocation of the electrode array (James et al., CIAP, Lake Tahoe, July 2017).

We reduced the results of fixed-SNR-level testing to SNR50 values to avoid ceiling effects in long-term sentence-in-noise data. We validated an analytic sigmoid fit to scores obtained from two SNRs per visit with an additional maximum, in-quiet score parameter. Fifteen CI users were tested twice in one session using French MBAA2 lists presented at progressively reducing SNRs (10, 5, 2, 0) until scores were < 50/100. Test-retest reliability was 0.86 dB using a full Levenburg-Marquardt fit based on 3 or more fixed SNRs, and 0.84 dB using the simplified analytic fit. The mean difference in SNRs test to re-test was 0.14 dB (p=0.65, t[14]=0.462).

The simplified analytic method is useful for clinical testing since it allows the standard collection of scores for fixed SNRs for longitudinal follow-up but also allows expression of results in terms of dB SNR50 to accommodate the large range of subject performance. For example, we will present preliminary analysis of patient- and device-related factors which may influence SpiN scores for CI users over time.

We will also present some clinical uses of the method to determine critical differences; for example to determine the spatial unmasking benefit of a hearing aid used with a CI (i.e. bimodal listening) and to compare measures of speech recognition performance for CI-alone in implanted unilateral hearing loss patients where acoustic leak may be an issue.