Spiking neural network model of sound localization using the interaural intensity difference.

Area of Science:

• Computational Neuroscience
• Auditory Processing
• Artificial Intelligence

Background:

• Mammalian auditory systems process sound localization using cues like interaural intensity difference (IID).
• Simulating complex neural processes requires sophisticated computational models.

Purpose of the Study:

• To develop and validate a spiking neural network (SNN) architecture for simulating sound localization based on IID.
• To model the mammalian auditory pathway, inspired by the lateral superior olive.

Main Methods:

• Integrated models of the auditory periphery (cochlea) and medial nucleus of the trapezoid body.
• Utilized leaky integrate-and-fire excitatory and inhibitory spiking neurons with facilitating synapses.
• Trained and validated the SNN using experimentally derived head-related transfer function (HRTF) acoustical data from cats.
• Employed a remote supervision method for supervised learning of azimuthal angles.

Main Results:

• The SNN architecture demonstrated effective sound localization capabilities.
• Performance was optimal for high-frequency sound localization, aligning with biological auditory processing.
• The model exhibited robustness against noise in the head-related transfer function acoustical data.

Conclusions:

• The developed SNN provides a viable computational model for studying sound localization mechanisms.
• The architecture's biological plausibility is supported by its performance characteristics.
• The model shows promise for understanding auditory perception in complex acoustic environments.