Predicting speech intelligibility across acoustic conditions and hearing status using a physiologically inspired auditory model

The current study presents an update and extensive evaluation of a previously introduced speech-intelligibility (SI) model by Scheidiger, Carney, Dau, and Zaar [2018, Acta Acust. United Ac. 104, 914-917, doi: 10.3813/aaa.919245]. The model processes the noisy speech stimulus and the noise-alone reference signal through a physiologically inspired nonlinear model of the auditory periphery, followed by a modulation analysis in the range of the fundamental frequency of speech. The decision metric of the model is the mean of a series of short-term across-frequency correlations between population responses to noisy speech and noise alone, with a sensitivity limitation process imposed. This decision metric was (inversely) related to SI using a conversion function obtained from a single fitting condition. The model was first evaluated in speech-in-noise conditions with stationary, fluctuating, and speech-like interferers using data previously obtained in NH and HI listeners, with HI listeners receiving higher broadband presentation level but no individualized amplification. Accurate predictions of NH listeners' speech reception thresholds (SRTs) were obtained across the different noise conditions, and the model also accounted for effects of hearing impairment on the SRTs when adjusting the front-end processing based on individual audiograms. The model was then evaluated using a second dataset collected in NH and HI listeners using stationary and speech-modulated noise, along with several speech interferers, with the HI listeners receiving individualized linear amplification. The model again showed convincing SRT predictions across most conditions and captured the relative effect of hearing status well, across all conditions. However, the model globally overestimated the release from masking induced by speech-like envelope fluctuations in the maskers, and failed to predict the difficulty induced by a same-gender interfering-talker. Overall, the proposed model provides a step towards more accurate predictions of speech-in-noise intelligibility in NH and HI listeners, and — perhaps more importantly — facilitates insights into processes that are crucial for speech understanding.