Predicting speech intelligibility in hearing-impaired listeners using a physiologically inspired auditory model.

Area of Science:

• Auditory Neuroscience
• Speech Perception
• Acoustical Signal Processing

Background:

• A previous speech intelligibility (SI) prediction model was developed by Scheidiger et al. (2018).
• The model predicts SI in speech-in-noise scenarios by comparing processed noisy speech and noise-alone signals.

Purpose of the Study:

• To present a major update and comprehensive evaluation of the existing SI prediction model.
• To assess the model's accuracy in predicting speech-reception thresholds (SRTs) for normal-hearing (NH) and hearing-impaired (HI) listeners across various noise conditions.

Main Methods:

• Signals were processed through a nonlinear auditory periphery model and modulation analysis.
• A decision metric based on across-characteristic-frequency (CF) correlations was employed.
• Model performance was validated using pre-existing SRT data from NH and HI listeners with stationary, fluctuating, and speech-like maskers.

Main Results:

• The updated model accurately predicted SRTs for NH listeners across different noise types, with minor overestimation of masking release for fluctuating maskers.
• For HI listeners, the model, adjusted for individual audiograms (considering inner-hair-cell/IHC and outer-hair-cell/OHC impairment), effectively predicted elevated SRTs and reduced masking release.
• Increased IHC impairment in the model plausibly predicted worsened SI.

Conclusions:

• The enhanced model accurately predicts speech-in-noise outcomes for both NH and HI listeners.
• The model serves as a valuable tool for understanding auditory processes vital for speech comprehension.
• It provides insights into the impact of different types of hearing impairment on speech intelligibility.