Predicting Speech Intelligibility Based on Across-Frequency Contrast in Simulated Auditory-Nerve Fluctuations

Area of Science:

• Auditory Neuroscience
• Speech Perception
• Acoustic Signal Processing

Background:

• Speech intelligibility is crucial for communication, especially for hearing-impaired individuals.
• Existing models often struggle to accurately predict speech intelligibility in complex noise conditions.
• Understanding the auditory system's processing of speech in noise is essential for developing better hearing technologies.

Purpose of the Study:

• To develop and validate a novel computational model for predicting speech intelligibility.
• To assess the model's accuracy for both normal-hearing (NH) and hearing-impaired (HI) listeners.
• To investigate the model's ability to account for hearing impairment by adjusting peripheral auditory processing parameters.

Main Methods:

• A non-linear auditory periphery model was combined with a decision process based on across-characteristic frequency (CF) modulation analysis.
• The model utilizes the short-term across-CF correlation between speech and noise to predict intelligibility.
• Hearing thresholds were used to adapt the peripheral model for simulating hearing-impaired listeners.

Main Results:

• The model achieved highly accurate predictions for NH listeners across various presentation levels.
• The model demonstrated plausible effects when simulating HI listeners by adapting peripheral parameters.
• The approach showed potential for modeling the impact of outer and inner hair cell loss on speech intelligibility.

Conclusions:

• The proposed modeling approach provides a valuable framework for predicting speech intelligibility in noise.
• The model's adaptability to hearing thresholds suggests its utility in understanding hearing impairment effects.
• Further refinement of HI listener profiles could significantly enhance the model's predictive power.