Evaluating an auditory model as predictor of speech understanding in hearing-impaired listeners

Speech intelligibility (SI) models that represent effects of hearing impairment (HI) in their processing stages may help understand the link between clinical measures of auditory dysfunction and daily-life challenges with speech-in-noise understanding experienced by listeners. Here, we present a thorough evaluation of a well-established normal-hearing (NH) SI model, the speech-based computational auditory signal processing and perception model [1], as a predictor of SI performance in HI listeners. By modelling three previously published datasets, we evaluated the model's predictive power of (i) the role of audibility in unaided speech reception thresholds, (ii) the masking release obtained in fluctuating (relative to stationary) maskers, and (iii) aided performance in conditions where amplification was provided. The model was evaluated both at the listener group level to assess differences between NH and HI populations, and as a predictor of individual listener performance within the HI population. For comparison, clinical estimates used to fit the model were also analyzed as individual SI predictors. We show that the current model accurately captures effects of audibility and represents the loss of masking release observed in the HI listeners. However, the model is still limited in accounting for data associated with supra-threshold auditory deficits.