Background: Several speech-intelligibility (SI) prediction models have been developed for application to a large range of acoustical conditions. Most of the available models were designed and validated based on reference data for normal hearing (NH), for which individual differences in SI are typically small and for which simplistic linear simulations of auditory processing can provide sufficient predictive power. However, although many of these models have later been extended to incorporate aspects of hearing loss, it has remained challenging to accurately predict SI differences between listener groups with NH and hearing impairment (HI) and even more challenging to predict within-group individual differences in the HI population.

Methods: We recently introduced an SI prediction model (Zaar and Carney, 2022, Hear. Res. 426:108553) based on the recently proposed hypothesis that across-frequency fluctuation profiles in auditory-nerve (AN) responses are relevant for discrimination of complex sounds. A phenomenological model provided simulated AN responses for NH and individual HI listeners. The model is here evaluated using several data sets consisting of auditory profiling data and speech reception thresholds (SRTs), measured in a range of noise conditions, all collected in both NH and HI listeners. The model was calibrated using NH data obtained in a single noise condition; predictions were then obtained as a result of differences in the stimuli (different noise conditions) and by incorporating pure-tone thresholds and estimates of outer and inner hair cell (OHC and IHC) impairment into the auditory model (different HI listeners). Special attention was paid to the interpretation of pure-tone thresholds in terms of the underlying OHC and IHC contributions.

Results: The model accounted very well for SI across noise conditions in the NH group and accurately predicted the elevation of SRTs and the reduced masking release due to hearing loss. The measured and predicted SRTs for the HI listeners were strongly correlated for one data set and moderately correlated for another, smaller, data set. The model predictions for the HI listeners were strongly dependent on the interpretation of the pure-tone thresholds with respect to the underlying OHC and IHC contributions. However, individualization of OHC/IHC impairments based on loudness-scaling data did not increase predictive power.

Conclusions: The results indicate that the proposed model accounts well for effects of additive noise and hearing impairment on SI. The differential effects of OHC and IHC impairment in the model warrant further investigation – while individualization based on loudness-scaling estimates was not successful, other diagnostic measures may yield better results.

Funding: LHC was supported by NIH DC010813.