Robust auditory localization using probabilistic inference and coherence-based weighting of interaural cues.

Area of Science:

• Auditory Neuroscience
• Computational Auditory Scene Analysis
• Acoustic Signal Processing

Background:

• The human auditory system exhibits remarkable sound source localization capabilities, even in complex and noisy environments.
• Existing computational models often struggle with unreliable localization cues and generalization across diverse acoustic scenarios.

Purpose of the Study:

• To propose a computational binaural localization model that handles corrupted or unreliable localization cues.
• To enable generalization across different acoustic situations by dynamically weighting interaural differences.
• To enhance sound source localization performance in challenging acoustic conditions.

Main Methods:

• Developed a computational binaural localization model incorporating interaural coherence (measured as interaural vector strength, IVS).
• Utilized a probabilistic framework where interaural level (ILD) and phase (IPD) differences are dependent on IVS.
• Employed Bayesian computation for direction-of-arrival probability mapping and coherence-weighted cue integration across frequency and time.

Main Results:

• The model demonstrated validity through statistical analysis of interaural parameters.
• Simulated localization experiments showed improved performance by exploiting low-reliability data points (low IVS).
• A temporal integration window of at least 200 ms was found necessary for performance benefits, aligning with psychoacoustic findings.

Conclusions:

• The proposed model effectively integrates interaural coherence to dynamically weight localization cues, improving robustness.
• This approach allows for enhanced sound source localization in challenging acoustic environments and with degraded auditory signals.
• The findings contribute to understanding auditory spatial perception and developing more sophisticated hearing prosthetics.